Free Radicals in Cyberspace: Complex Liability Issues in Information Warfare

¶ 1 During the weeks leading up to September 11, 2001, successive versions of a particularly destructive and complex Internet worm, named W32/CodeRed, took information warfare to a new level of complexity and danger. By exploiting a common network vulnerability, the rapidly spreading CodeRed slowed down and compromised the security of the Internet, and attempted to launch denial-of-service attacks on the official White House web page.¹ The first version of the worm, which we shall refer to as CodeRed-I, exploited a security vulnerability in Microsoft's Internet Information Services ("IIS") web servers.²

¶ 2 Once CodeRed-I infected a machine, it checked whether the current date was between the first and nineteenth of the month. If that were the case, the worm probed a randomly generated list of machines for vulnerability and continued the infection cycle. Between the twentieth and twenty-eighth of every month, the worm turned its attention from other machines and proceeded to launch a denial-of-service attack on the official White House web page, www.whitehouse.gov. The worm remained dormant between the twenty-eighth and the end of the month.

¶ 3 Due to a programming flaw, CodeRed-I spread more slowly than intended, yet infected enough hosts to cause a significant denial-of-service slowdown in the infected systems. Its attempted attack on the White House web page failed, because the site was moved to a new IP address following an intelligence alert.³ The worm code continued to target the old address, while legitimate traffic was redirected to the new address.

¶ 4 A more destructive sequel, CodeRed-II, soon followed. CodeRed-II was similar to its predecessor, but had a greater impact on the global information infrastructure and did more harm, in part due to its more efficient propagation algorithm.⁴ The new version spread many times faster and also created a backdoor⁵ on infected systems. The backdoor installed by CodeRed-II enabled a hacker to gain access to confidential files and programs on the compromised computer.⁶ CodeRed-II exploited the same vulnerability as its predecessors.⁷

¶ 5 On August 4, 2001, a new worm named CodeRed-III appeared, which exploited the same vulnerability as its predecessors.⁸ CodeRed-III infected its target system, initiated its propagation mechanism, and set up a backdoor into the infected machine. After installing the backdoor, it remained dormant for a day, rebooted the machine, and began to spread. The backdoor allowed remote, administrator-level access to the infected machine. This enabled the compromised system to be used as a launching pad for future denial-of-service attacks, among other hazards.⁹

¶ 6 The CodeRed family of worms was promptly followed by a fast-spreading and complex threat named Nimda. Nimda struck on September 18, 2001, within days of warnings issued by various government agencies, including the Federal Bureau of Investigation and the National Infrastructure Protection Center. Nimda was described as "the most complicated malicious application to strike the Internet to date," and the office of the U.S. Attorney General predicted that Nimda would be more harmful than CodeRed.¹⁰

¶ 7 Nimda was a worm, but, like CodeRed, differentiated itself from other Internet worms in its exploitation of network security flaws, use of multiple vectors of infection and propagation, and resulting efficiency and speed. Among many exploits, Nimda searched automatically for vulnerable Microsoft IIS web servers to infect, and it used backdoors created by CodeRed. The different attack vectors resulted in multiple points of damage, which made clean-up particularly difficult after a Nimda attack.¹¹ Nimda's infection and attack vectors were not novel individually, but their combination constituted a new level of complexity in malevolent code.

¶ 8 CodeRed and Nimda were the forerunners of a new genre of modern malevolent software, known as "blended attacks." Information security scholars have described the succession of blended attacks ushered in by CodeRed and Nimda as the "fourth wave of modern worms." The fourth wave followed the initial experimental wave of the 1980s, the second wave of polymorphic viruses and virus toolkits, and the third wave of mass e-mail viruses, such as Melissa, of the late 1990's.¹²

¶ 9 Blended threats are diverse, but they have two main characteristics in common: (i) exploitation of one or more security vulnerabilities, and (ii) multi-vector malevolent code with multiple destructive properties. The combination creates synergies that make blended threats significantly more hazardous than their predecessors.¹³

¶ 10 Blended attacks create complex liability issues. In a negligence action¹⁴ involving a blended attack, the two most likely defendants are: (i) the person responsible for the security vulnerability, and (ii) the person who programmed and distributed the virus or worm to exploit the vulnerability. The former—the original tortfeasor—may be a software designer or commercial vendor, and as such, likely to be solvent and able to pay a tort judgment. The second tortfeasor—the hacker or virus distributor—on the other hand, is often judgment-proof.

¶ 11 The liability of an original tortfeasor is usually cut off by an intervening crime or intentional tort. Suppose, for instance, a technician negligently fails to properly fasten the wheels of a plaintiff's car. A wheel comes off, leaving the plaintiff stranded on a busy highway. The stranded plaintiff is subsequently struck by a passing driver who failed to pay attention. The technician and the inattentive driver were both negligent and would both be held liable for the plaintiff's harm. The inattentive driver's inadvertent negligence would not cut off the liability of the mechanic.¹⁵ Suppose, alternatively, a passing driver is in a mood to kill someone, and seeing the stranded motorist as a convenient target, shoots him. In this scenario, the passing driver's intentional or criminal intervention would cut off the liability of the negligent mechanic and shift liability exclusively to the second wrongdoer. Such a liability shift from a solvent original tortfeasor to a judgment-proof cyber attacker may leave a victim without recourse. The original tortfeasor's liability may be preserved, however, under the Encourage Free Radical ("EFR") doctrine.

¶ 12 The EFR doctrine preserves the liability of an original tortfeasor who encourages those individuals who are shielded from liability by anonymity, insufficient assets, lack of mental capacity, or lack of good judgment.¹⁶ Such trouble-prone individuals are termed "free radicals" because of their tendency to bond with trouble. Examples of free radicals include children, anonymous crowds, criminals, terrorists, and mentally incompetent individuals.¹⁷

¶ 13 The EFR doctrine recognizes that the prospect of negligence liability is ineffective against defendants who are shielded from or otherwise undeterred by the prospect of liability. The deterrence rationale of negligence law would be defeated if responsible people who encourage free radicals were allowed to escape judgment by shifting liability to undeterrable free radicals. Common law negligence rules therefore impose liability on the first tortfeasor—the encourager of the free radicals—even when intentional or criminal behavior by a free radical intervenes.

¶ 14 The analysis in this Article suggests that virus authors and distributors have properties commonly associated with free radicals. The analysis informs, furthermore, that the factors that influence courts in holding a defendant liable for encouraging free radicals are present in a typical blended attack. We conclude that liability will be preserved against a primary tortfeasor whose negligence was responsible for a vulnerability that was intentionally exploited by a free radical cyber attacker. The primary tortfeasor is likely a solvent commercial entity, while the attacker is often judgment-proof or otherwise shielded from liability. The result is therefore significant, especially for the victim of a blended attack seeking to recover damages related to the attack.

¶ 15 The role of free radicals and their encouragers has been recognized in cyberlaw. Initial litigation in online music copyright infringement, for instance, was directed at web sites that "encouraged" alleged infringers by matching music uploaders with downloaders.¹⁸ Subsequent copyright suits targeted venture capital firms that funded the web sites, as well as ISPs who provided the infrastructure for the sites' services.¹⁹

¶ 16 Although state and federal statutes allow compensatory damages and injunctive relief directly against cyber attackers,²⁰ critics have argued that a more appropriate target for such lawsuits would be the deep-pocketed entities whose lax security practices facilitated the attacks.²¹ It has also been proposed that ISPs should not be immune from liability for contributing to online malfeasance, such as virus and worm attacks.²² The reasons commonly cited for imposition of liability on ISPs for the malfeasance of those using their services emphasize the free radical nature of the wrongdoers.²³

¶ 17 Although ISPs continue to enjoy immunity, there is a general awareness in the information security community of the explosive mix of security vulnerabilities, information inefficiencies, and free radical cyber rogues with destructive weapons standing ready to take advantage of the situation. Corporations with lax security practices, such as the failure to fix common security vulnerabilities, have been held liable for encouraging cyber rogues to launch attacks on their networks,²⁴ and a recent class action suit against Microsoft accuses the software giant of releasing information about security vulnerabilities in the Windows operating system in such a way that the information aided free radical cyber rogues more than network administrators.²⁵

¶ 18 This Article is organized as follows: Section II introduces the principles of malevolent software, blended attacks, and the (currently) most commonly exploited security vulnerability, the buffer overflow. Section III discusses liability issues in blended attacks. Section IV analyzes the EFR doctrine in the context of blended attacks. A final section discusses these issues and concludes.

II. Malevolent software: Background

¶ 19 Malevolent software is a term for code that is intended to cause damage to or disrupt the operation of a computer system. The most common of these rogue programs is the computer virus and its common variant, the worm. Other forms of malicious software include so-called logic bombs, Trojan horses, and trap doors.²⁶

¶ 20 The term "virus," Latin for "poison," was first formally defined in terms of computer programs by Dr. Fred Cohen in 1983,²⁷ even though the concept goes back to John von Neumann's studies of self-replicating mathematical automata in the 1940s.²⁸ Dr. Cohen describes a computer virus as a series of instructions, or in other words, a program, that: (i) infects other computer programs and systems by attaching itself to a host program in the target system; (ii) executes when the host is executed; and (iii) spreads by cloning itself, or part of itself, and attaching the copies to other host programs on the system or network. In addition, many viruses have a so-called payload capable of harmful side-effects, such as data corruption.²⁹

¶ 21 A worm³⁰ is a special type of virus. It is similar to a virus in most ways, except that it is self-replicating. A worm does not need to attach itself to a host program to replicate and spread. Like viruses, worms often carry payloads capable of destructive behavior, such as deleting files on the system through which it propagates. Worms without a destructive payload can nevertheless slow down a system significantly through the network traffic it generates with its prolific replication and spreading.³¹

¶ 22 The first worm was implemented by scientists at Xerox PARC in 1978.³² The so-called Morris worm, created by Cornell University graduate student Robert T. Morris, was the first worm to become a household name. The 1989 Morris worm used a security flaw in a UNIX program to invade and shut down much of the Internet. By some accounts, this event first woke the world up to the dangers of the computer vulnerability known as a buffer overflow.³³

¶ 23 As the definition suggests, computer viruses consist of three basic modules or mechanisms, namely an infection mechanism, a payload trigger, and the payload. The infection mechanism allows the virus to replicate and spread analogously to a biological virus. This is the most salient property of a computer virus.³⁴ The infection module first searches for an appropriate executable host program to infect. It then installs a copy of the virus onto the host, provided the host had not yet been infected.

¶ 24 When the host program executes, the virus is also executed. Upon execution, the virus typically performs the following sequence of actions. It replicates ("clones") by copying itself to other executable programs on the computer.³⁵ During execution, the virus program also checks whether a triggering condition is satisfied. When the condition is satisfied, the virus executes its harmful component, the so-called payload module. Triggering events come in a variety of forms, such as a certain number of infections or the occurrence of a particular date. The Friday-the-13th virus, for instance, only activated its payload on dates with the cursed designation.³⁶ More recently, the first CodeRed worm alternated between continuing its infection cycle, remaining dormant, and attacking the official White House web page, depending on the day of the month.

¶ 25 Execution of the payload may produce harmful side-effects, such as destruction or corruption of data in spreadsheets, word processing documents and data bases, or theft of passwords.³⁷ Some effects are particularly pernicious because they are subtle and undetectable until substantial harm has been done. Subtle harmful viral effects include transposing numbers, moving decimal places, and stealing passwords and other sensitive information.³⁸ Payloads are not necessarily destructive and may involve no more than displaying a humorous message.³⁹ Some virus strains do not destroy or corrupt information but consume valuable computing resources.⁴⁰ Viruses and worms used in blended attacks, however, are harmful by design.

¶ 26 A virus may infect a computer or a network through several possible points of entry, including via an infected file downloaded from the Internet, web browsing, an infected e-mail attachment, or infected commercial shrink-wrapped software.⁴¹ Fast-spreading worms, such as CodeRed and Blaster, can only infect new hosts that contain one or more exploitable vulnerabilities.⁴² The recent trend in virus transmission has been a decrease in infected diskettes and an increase in infection through e-mail attachments. In a 1996 national survey, for instance, approximately nine percent of respondents listed e-mail attachments as the means of infection of their most recent virus incident, while seventy-one percent put the blame on infected diskettes. In 2003, the corresponding numbers were eighty-eight percent for e-mail attachments, and zero for diskettes.⁴³

¶ 27 It was once believed that viruses could not be transmitted by data files, such as e-mail attachments. Viruses such as the infamous Melissa taught us otherwise. Melissa typically arrived in the e-mail inbox of its victim, disguised as an e-mail message with a Microsoft Word attachment. When the recipient opened the attachment, Melissa executed. First, it checked whether the recipient had the Microsoft Outlook e-mail program on its computer. If Outlook was present, Melissa would mail a copy of itself to the first fifty names in Outlook's address book, creating the appearance to the fifty new recipients that the infected person had sent them a personal e-mail message. Melissa would then repeat the process with each of the fifty recipients of the infected e-mail message (provided they had Outlook), by automatically transmitting clones of itself to fifty more people. A Melissa attack frequently escalated and resulted in clogged e-mail servers and system crashes.⁴⁴

¶ 29 Anti-virus technology comes in two broad categories, namely "virus-specific" and "generic." Virus-specific technologies, such as signature scanners, detect known viruses by identifying patterns that are unique to each virus strain. These identifying patterns, known as "signatures," are analogous to human fingerprints. Generic anti-virus technology, on the other hand, detects the presence of a virus by recognizing generic virus-like behavior, usually without identifying the particular strain.

¶ 30 A virus-specific scanner typically makes a specific announcement, such as "the operating system is infected with the Cascade virus," while its generic counterpart may simply state "the operating system is (or may be) infected with an (unidentified) virus." Virus-specific technology is more accurate with known strains and produces fewer false positives, but generic technology is better at detecting unknown viruses.

¶ 31 Technical anti-virus defenses come in four varieties, namely scanners, activity monitors, integrity checkers, and heuristic techniques.⁴⁵ Scanners are virus-specific, while activity monitors and integrity checkers are generic. Activity monitors look out for suspicious, virus-like activity in the computer. Integrity checkers sound an alarm when detecting suspicious modifications to computer files. Heuristic techniques combine virus-specific scanning with generic detection, providing a significantly broadened range of detection.

¶ 32 Scanners are the most widely used anti-virus defense. A scanner reads executable files and searches for known virus patterns. These patterns, or "signatures," are the most reliable technical indicator of the presence of a file-resident virus in a computer system. A virus signature consists of patterns of hexadecimal digits embedded in the viral code that are unique to the strain.⁴⁶ These signatures are created by human experts, such as researchers at IBM's High Integrity Computing Laboratory, who scrutinize viral code and extract sections of code with unusual patterns. The selected byte patterns then constitute the signature of the virus.⁴⁷ The scanner announces a match with its database of known viral signatures as a possible virus.

¶ 33 The virus signature pattern is selected to be a reliable indicator of the presence of a virus. An ideal virus signature gives neither false negatives nor false positives.⁴⁸ In other words, it ideally should always identify a virus when present and never give a false alarm when a virus is not present.⁴⁹

¶ 34 Scanners are easy to use, but they are limited to detecting known virus signatures. A scanner's signature database has to be continually updated, a burdensome requirement in an environment where new viruses appear rapidly. Use of scanners is further complicated by the occurrence of false positives. This occurs when a viral pattern in the database matches code that is in reality a harmless component of otherwise legitimate data. A short and simple signature pattern is all too often found in innocent software, leading to many false positives. Viruses with longer and more complex patterns will less often give a false positive, but at the expense of more false negatives.⁵⁰ Finally, as the number of known viruses grows, the scanning process will inevitably slow down as the system evaluates a larger set of possibilities.⁵¹

¶ 35 Activity monitors are resident programs that monitor activities in the computer for behavior commonly associated with viruses. Suspicious activities include operations such as attempts to rewrite the boot sector, format a disk, mass mail multiple copies of itself, or modify parts of main memory. When suspicious activity is detected, the monitor may simply halt execution and issue a warning to alert the user, or take definite action to neutralize the activity.⁵² Activity monitors, unlike scanners, do not need to know the signature of a virus to detect it. It works for all viruses, known as well as unknown. Its function is to recognize suspicious behavior, regardless of the identity of the culprit.

¶ 36 The greatest strength of activity monitors is their ability to detect unknown virus strains, but they also have significant weaknesses. They can only detect viruses that are actually being executed, possibly after substantial harm has been done. A virus may, for instance, become activated before the monitor code, and escape detection until well after execution has begun. A virus may also be programmed to alter monitor code on machines that do not have protection against such modification. A further disadvantage of activity monitors is the lack of unambiguous and foolproof rules governing what constitutes "suspicious" activity. This may result in false alarms when legitimate activities resemble virus-like behavior. Recurrent false alarms may ultimately lead users to ignore warnings from the monitor. Conversely, not all "illegitimate" activity may be recognized as such, leading to false negatives.⁵³

¶ 37 Integrity checkers look for evidence of file tampering, such as "unauthorized" changes in system areas and files. The typical integrity checker is a program that generates a code, known as a "checksum," for files that are to be protected from viral infection. A file checksum may, for instance, be some arithmetic calculation based on the total number of bytes in the file, the numerical value of the file size, and the creation date. The checksum effectively operates as a "signature" of the file. These checkcodes are periodically recomputed and compared to the original checksum. Tampering with a file will change its checksum. Hence, if the recomputed values do not match the original checksum, the file has presumably been modified since the previous check, and a warning is issued. Since viruses modify and change the contents of the files they infect, a change in the checksum may be a sign of viral infection.⁵⁴

¶ 38 The advantage of integrity checking is that it detects most instances of viral infection, as an infection must alter the target file. The main drawback is that it tends to generate many false alarms, as a file can change for legitimate reasons unrelated to virus infection.⁵⁵ On some systems, for instance, files change whenever they are executed. A relatively large number of false alarms may trigger compliance lapses, as users may ignore warnings or simply not use the utility. Integrity checking works best on static files, such as system utilities, but is, of course, inadequate for files that naturally change frequently, such as Word documents.

¶ 39 A fourth category of virus detectors uses heuristic detection methods. Heuristic rules are rules that solve complex problems "fairly well" and "fairly quickly," but less than perfectly. Virus detection is an example of a complex problem that is amenable to heuristic solution. It has been proven mathematically that it is impossible to write a program that is capable of determining with 100 percent accuracy whether a particular program is infected with a virus, from the set of all possible viruses, known as well as unknown.⁵⁶ Heuristic virus detection methods accept such limitations and attempt to achieve a solution, namely a detection rate that is "pretty good," albeit below the (unachievable) perfect rate.

¶ 40 Heuristic virus detection methods examine executable code and scrutinize its structure, logic, and instructions for evidence of "virus-like" behavior. Based on this examination, the program makes an assessment of the likelihood that the scrutinized program is a virus by tallying up a score. Instructions to send an e-mail message with an attachment to everyone in an address book, for instance, would add significantly to the score. Other high-scoring routines include capabilities to replicate, hide from detection, or execute some kind of payload. When a certain threshold score is reached, the code is classified as malevolent, and the user is so notified.

¶ 41 The assessment is necessarily less than perfect and occasionally provides false positives and negatives. Many legitimate programs, including even some anti-virus programs, perform operations that resemble virus-like behavior.⁵⁷ Nevertheless, state-of-the-art heuristic scanners typically achieve a seventy to eighty percent success rate at detecting new and unknown viruses.⁵⁸

¶ 42 A heuristic scanner typically operates in two phases. The scanning algorithm first narrows the search by, for instance, identifying the location most likely to contain a virus. It then analyzes the code from that location to determine its likely behavior upon execution. A static heuristic scanner typically compares the code from the "most likely" location to a database of byte sequences commonly associated with virus-like behavior.⁵⁹ The algorithm then decides whether to classify the code as viral.⁶⁰

¶ 43 A dynamic heuristic scanner uses CPU emulation.⁶¹ It typically loads suspect code into a virtual computer, emulates its execution, and observes its behavior. Because it is only a virtual computer, virus-like behavior can safely be observed in what is essentially a laboratory setting, with no need to be concerned about real damage. The program is monitored for suspicious behavior while it runs.⁶²

¶ 44 Although dynamic heuristics can be time-consuming due to the relatively slow CPU emulation process, they are sometimes superior to static heuristics. This will be the case when the suspect code: (i) is obscure and not easily recognizable as viral in its static state, but (ii) clearly reveals its viral nature in a dynamic state.

¶ 45 A major advantage of heuristic scanning is its ability to detect viruses, including unknown strains, before they execute and cause damage. Other generic anti-virus technologies, such as behavior monitoring and integrity checking, can only detect and eliminate a virus after exhibition of suspicious behavior, usually after execution. Heuristic scanning is also capable of detecting novel and unknown virus strains, the signatures of which have not yet been cataloged. Such strains cannot be detected by conventional scanners, which only recognize known signatures. Heuristic scanners are capable of detecting even polymorphic viruses, a complex virus family that complicates detection by changing their signatures from infection to infection.⁶³

¶ 46 The explosive growth in new virus strains has made reliable detection and identification of individual strains very costly, making heuristics more important and increasingly prevalent.⁶⁴ Commercial heuristic scanners include IBM's AntiVirus boot scanner and Symantec's Bloodhound technology.

¶ 47 CodeRed and Nimda were the forerunners of a new wave of modern malevolent software known as the "blended attack."⁶⁵ Blended attacks are more sophisticated, complex, and dangerous than their predecessors. They exploit computer security vulnerabilities and often create new vulnerabilities to enhance their destructiveness. The earlier generation of viruses, such as LoveLetter, Melissa, and Michelangelo, in contrast, exploited only the regular functionality of the systems they targeted.

¶ 48 Blended threats are diverse, but they have two main characteristics in common, namely: (i) the exploitation of security vulnerabilities, and (ii) malevolent code with multiple destructive properties.

¶ 49 Blended threats are designed to take advantage of security vulnerabilities to gain access to and compromise a system.⁶⁶ The buffer overflow is currently (and has been for over a decade) the most commonly exploited vulnerability to obtain unauthorized access to a system.⁶⁷ A buffer overflow vulnerability allows executable malevolent code to be copied into the memory of a target computer. A skillful attacker can then exploit the vulnerability to manipulate the computer to remotely execute the malevolent code.⁶⁸

¶ 50 Other security flaws, such as input validation vulnerabilities, are also frequently exploited by blended threats. A web page exhibits an input vulnerability, for instance, if it asks for user input, such as an e-mail address, without verifying that the user-provided address is in the proper form. Such a flaw may enable a hacker to manipulate the system by providing a specially formatted input. The uncensored input may cause the system to perform in a way that compromises its security.

¶ 51 Vendors are usually quick to issue patches to fix vulnerabilities as soon as they are discovered, but users tend to be slow in implementing them, and even if several vulnerabilities are patched, some may remain, inviting exploitation. By some estimates, even if ninety percent of the users of a particular technology with a newly discovered vulnerability could be trusted to implement the security patch issued by the vendor, the remaining unpatched systems could still allow enough hijackings to launch a denial-of-service attack on millions of other systems and networks.⁶⁹ Successive generations of CodeRed plagued the Internet despite the fact that each attack and the role played by the vulnerability were widely publicized, and that a security patch to fix the vulnerability had been made available even before the first CodeRed attack.

¶ 52 Blended attacks employ viruses and worms with multiple destructive properties. The properties are usually not individually novel, but their combination in one virus or worm is unique. The payloads of blended threats are multidimensional and harmful by design.

¶ 53 Many conventional virus strains do little but create a mild nuisance. The earlier Italian PingPong virus, for instance, merely displayed a bouncing ball, and the W95/LoveSong/998 virus was designed to trigger a love song on a particular date.⁷⁰ Blended threats, in contrast, are destructive by design. Blended threats carry a variety of payloads, including mechanisms capable of triggering DoS agents, deleting data files, and setting up backdoors in infected systems.⁷¹ The CodeRed blended attack attempted to launch a full-scale denial-of-service attack on the official White House web page. The Slammer worm infected more than ninety percent of computers within ten minutes with a particular buffer overflow vulnerability, and caused significant disruption to financial, transportation, and government institutions, including widespread ATM failures, canceled airline flights, and interference with elections.⁷²

b. Blended threats propagate by multiple methods, attack from multiple points, and spread without human intervention.

¶ 54 The typical blended threat attacks its target via multiple attack methods and attack points, enabling it to spread more rapidly and efficiently, consume more computational resources and network bandwidth, and spread within a shorter time period. The Nimda worm, for instance, attacked via five vectors, including via e-mail propagation using its own SMTP engine and via backdoors left by worms such as CodeRed.⁷³

¶ 55 Blended threats attack from multiple points, including injecting malicious code into executable files on a system and targeting and infecting visitors to compromised web sites, often through an innovative use of mass e-mail. Mass-mailing worms in blended attacks frequently bypass existing e-mail applications by using their own e-mail servers to spread. Such a worm could infect a computer with Microsoft Outlook, for instance, and spread via e-mail without using the Outlook application.⁷⁴

¶ 56 Blended attacks do not require user intervention to trigger and spread, whereas traditional viruses depend on such intervention. Melissa, for instance, required users to actually open an e-mail attachment before the virus could execute and continue its infection cycle. Blended attacks exploit vulnerabilities that allow them to dispense without such interaction. A buffer overflow vulnerability in the e-mail servers Microsoft Outlook and Outlook Express, for instance, enabled an e-mail worm to spread automatically. The malicious code in the infected e-mail message could be executed merely by reading an HTML message, without opening an attachment. The recipient could therefore not protect herself by declining to open any attached files.

¶ 57 Blended threats are programmed to automatically search for and exploit new vulnerabilities. Such vulnerabilities often surface in new and emerging technologies, such as instant messaging technology, wireless local area networks, personal digital assistants, peer-to-peer networks, and networked cellular telephones.⁷⁵ Corporations and government departments and agencies, which rely increasingly on such vulnerable new technologies to conduct business, are particularly at risk. Many of these organizations and agencies are crucial elements of the national critical information infrastructure, including banking, transportation, communications, and energy provision systems.

¶ 58 We now turn to a discussion of the buffer overflow, the most commonly exploited security vulnerability.

¶ 59 Buffers are data storage areas in memory with a limited capacity. Buffers often function as temporary storage for data that is to be transferred between two devices that are not operating at the same speed. The purpose of the temporary storage is to coordinate speed differentials between the adjacent devices. A printer, for instance, is not capable of printing data at the speed that it receives the data from the computer feeding it. A buffer in the interface between the computer and printer typically resolves this bottleneck. Instead of feeding the printer directly, the computer sends the data to the buffer. While the buffer relays the information to the printer, at the printer's speed, the computer is freed up to carry on with other tasks.⁷⁶

¶ 60 A buffer overflow occurs when a program attempts to fill a buffer with more data than it was designed to hold. A buffer overflow is analogous to pouring ten ounces of water into a glass designed to hold eight ounces. The water must obviously overflow somewhere and create a mess. The glass represents a buffer, and the water the application or user data.⁷⁷ The excess data typically flows into adjacent memory locations, where it may corrupt existing data, possibly changing the instructions and resulting in unintended executions. The unintended executions may be harmless, but may also be malicious by design. In the most benign scenario, the buffer overflow will cause the program to abort, but without much further harm.⁷⁸ In a darker scenario, a buffer overflow could allow a hacker to remotely inject executable malicious code into the memory of a target computer and execute it.

¶ 61 Suppose, for instance, the adjacent area ("overflow area") contains an instruction pointer, which defines the instruction to be executed next. By overwriting this pointer, the attacker can influence the program's next execution. The attacker may, for instance, fill the buffer with malicious code, such as a virus or worm, and overwrite the pointer with the address of the buffer. This would cause the execution path to change and cause the program to execute the viral code in the buffer.⁷⁹

¶ 62 The following depicts the most basic sequence of events in a buffer overflow attack:

4. The new return address now points to the new data in the buffer, which may be malevolent instructions.

¶ 63 In 1989, the so-called Morris Worm used a buffer overflow vulnerability in a UNIX program to invade and shut down much of the Internet. It was the first worm of its kind to become a household name, and, by some accounts, brought the destructive potential of the buffer overflow to the attention to the computer community.⁸¹

III. Liability issues in blended attacks

¶ 64 A civil action involving a blended attack would most likely be pursued under a negligence theory, the most widely used theory of liability in the law of torts.⁸² Negligence is generally defined as a breach of the duty not to impose an unreasonable risk on society.⁸³ This theory applies to any risk that can be characterized as unreasonable, including the risks associated with malevolent software. A victim of a blended attack may therefore bring legal action under a negligence theory against anyone who contributed to the risks associated with the attack, as well as against those who failed in their duty to reduce or eliminate the risk.⁸⁴ Although most cases claiming damages resulting from viruses and other malevolent software never reach a determination of the merits, courts are increasingly recognizing such claims.⁸⁵

¶ 65 Blended threats are diverse, but they have two main characteristics in common: (i) they exploit security vulnerabilities, and (ii) they contain malevolent code with multiple destructive properties. This suggests that the most likely defendants in a blended attack would be the original tortfeasor responsible for the security flaw, usually a solvent commercial vendor, and the virus distributor, the intervening party who programmed and distributed the virus or worm to exploit the vulnerability. The virus distributor is in practice often judgment-proof and shielded by the anonymity of cyberspace. The liability of the original tortfeasor is therefore likely of greater interest to a prospective plaintiff.

¶ 66 The plaintiff in a negligence action has to prove the following elements to establish her claim:

A legal duty on the part of the defendant not to expose the plaintiff to unreasonable risks.

A breach of the duty, namely a failure on the part of the defendant to conform to the norm of reasonableness.

A causal connection between defendant's conduct and plaintiff's harm. This element includes actual as well as proximate cause. Defendant's negligence is the actual cause of the plaintiff's harm if, but for the negligence, the harm would not have occurred. Proximate causation means that the defendant's conduct must be reasonably closely related to the plaintiff's harm.

¶ 67 Generally, a duty exists: (i) where someone sells a product, (ii) where someone has committed an affirmative act, (iii) when a special relationship exists, (iv) when a special kind of contract exists that benefits the plaintiff, and (v) where there is an undertaking by the defendant. Duty is also not an impediment to the plaintiff when a defendant has acted maliciously to destroy property.⁸⁶

¶ 68 Courts have occasionally used a finding of "no duty" to limit the expansion of certain cybertorts. In Lunney v. Prodigy Services Co.,⁸⁷ for instance, the court held that the defendant was not negligent for allowing an imposter to send threatening e-mail messages on a Prodigy account. The court declined, as a matter of public policy, to impose a duty on an ISP to screen all their e-mail communications.⁸⁸ In contrast, in Diversified Graphics v. Groves,⁸⁹ the court held computer professionals to an elevated duty of care because of their specialized expertise.⁹⁰

¶ 69 Courts require a plaintiff to prove breach of duty by identifying an untaken precaution that would have prevented the harm, and by showing that the untaken precaution would have yielded greater benefits in accident reduction than its cost. The issue of breach in the context of a blended attack is discussed and analyzed in Section IV.C (stating that the "encouragement of free radicals must be negligent"). The issue of damages in a virus context, including the economic loss rule, has been analyzed in related articles.⁹¹

¶ 70 We now turn to proximate causality, which is the most complex and interesting liability issue in blended attacks.

¶ 71 Mark Grady explains that proximate cause applies to two broad categories of cases, including those involving: (i) multiple risks, and (ii) concurrent efficient causes.⁹² A Multiple Risks case typically involves two risks, both of which would have been reduced by the defendant's untaken precaution. The first is the primary risk, which was clearly foreseeable to a reasonable person, and the second an ancillary risk, which was not reasonably foreseeable.⁹³ Suppose, for instance, a surgeon performs a vasectomy negligently, and a child is born. The child grows up and sets fire to a house. The owner of the house sues the doctor for negligence. This is clearly a multiple risks case. The primary risk consists of foreseeable medical complications due to the incompetent vasectomy, including an unwanted pregnancy. The ancillary risk is the (unforeseeable) risk that the conceived child may grow up to be a criminal.⁹⁴ The proximate cause issue is whether the defendant should be held liable for the harm due to the ancillary risk.

¶ 72 Grady notes that a Concurrent Efficient Causes case involves multiple causes, all of which are actual causes of the same harm.⁹⁵ In a typical Concurrent Efficient Causes case, an original wrongdoer and a subsequent intervening party are both responsible for the plaintiff's harm.⁹⁶ Suppose, for instance, a technician negligently fails to properly fasten the wheels of a plaintiff's car. A wheel comes off, leaving the plaintiff stranded on a busy highway. The stranded plaintiff is subsequently struck by a passing driver who failed to pay attention. The technician and the inattentive driver were both negligent and are concurrent efficient causes of the plaintiff's harm. The proximate cause issue is whether the second tortfeasor's intervening act should cut off the liability of the first. We now show that proximate cause is analyzed best when viewed as a dualism, consisting of two separate doctrines.

¶ 73 As Grady clarifies, proximate cause is a dualism consisting of two separate doctrines or tests. One doctrine applies to Multiple Risks cases and the other to Concurrent Efficient Causes cases.⁹⁷ Some accidents involve purely multiple risks, while others involve purely concurrent causes. In some cases, however, both doctrines apply. When both situations—Multiple Risks as well as Concurrent Efficient Causes—exist in the same case, both proximate cause doctrines apply, and the requirements for both must be satisfied for a finding of proximate cause.⁹⁸

¶ 74 The Reasonable Foresight doctrine applies to cases of multiple risks, where a primary and ancillary risk both caused the plaintiff's harm. This doctrine establishes the conditions under which the tortfeasor who created the primary risk will be liable for the actual harm resulting from the ancillary risk.⁹⁹ The bungled vasectomy is a typical Reasonable Foresight case. The Reasonable Foresight doctrine determines whether the surgeon would be held liable for damage caused by the ancillary risk, namely the risk that an unwanted pregnancy may produce a future criminal.

¶ 75 The Direct Consequences doctrine of proximate cause applies to cases involving multiple efficient causes. The doctrine "examines concurrent causes to determine whether the person responsible for the second cause has cut off the liability of the person responsible for the first cause."¹⁰⁰ The "loose wheel" case is a typical Direct Consequences case. The Direct Consequences doctrine would determine whether the intervening tortfeasor (the inattentive driver who struck the stranded plaintiff) would cut off the liability of the original tortfeasor (the negligent automobile technician.)

¶ 76 The Direct Consequences doctrine applies to blended attacks. A blended attack has two efficient causes, namely the security vulnerability and the virus distributor who exploited the vulnerability to launch the attack. The vulnerability and the intervening hacker are both essential to, and but-for causes of, the attack. Fast-spreading worms, such as CodeRed or Nimda, could not infect a system without an exploitable vulnerability. A system without the vulnerability or with an effective patch properly installed would be immune to these worms.¹⁰¹

¶ 77 In the proximate cause analysis of a blended attack, the buffer overflow vulnerability is the original cause for which one of the defendants, the software designer, is responsible. Subsequently, an intervening defendant committed a second tort, namely transmitting a virus programmed to exploit the vulnerability. The second tort is a possible supervening tort which may cut off the liability of the first tortfeasor.

¶ 78 The direct consequences doctrine of proximate cause determines when the second concurrent efficient cause, the virus distributor, would cut off the liability of the person responsible for the first, the buffer overflow vulnerability. The liability of the virus distributor is not an issue, as she will always be liable, as long as the elements of duty, breach, and actual causation are satisfied. However, a plaintiff would usually be more interested in suing the solvent original tortfeasor than the judgment-proof hacker.

¶ 79 Analysis of the Direct Consequences doctrine becomes clearer when broken down into five mutually exclusive paradigms. If a case falls clearly within one of the paradigms, its proximate cause analysis is normally straightforward.

¶ 80 Any direct consequences case belongs to one of five mutually exclusive paradigms, namely: (i) No Intervening Tort, (ii) Encourage Free Radicals, (iii) Dependent Compliance Error, (iv) No Corrective Precaution, and (v) Independent Intervening Tort.¹⁰²

¶ 81 The No Intervening Tort paradigm is the default paradigm. It preserves proximate cause if no tort committed by a third party has intervened between the original defendant's negligence and the plaintiff's harm, as long as the type of harm was foreseeable.¹⁰³ In this paradigm, the original tortfeasor is not only the direct cause of the harm, but also the only wrongdoer.¹⁰⁴ A speeding and unobservant driver who strikes a pedestrian walking carefully in a crosswalk is a clear example of a case within the No Intervening Tort paradigm. The original wrongdoer is clearly liable under this paradigm, and is also the only wrongdoer. A blended attack does not fit into this paradigm because of the intervening tort of a second wrongdoer, the cyber attacker.

¶ 82 Under the Encourage Free Radicals ("EFR") paradigm, proximate cause is preserved if the defendant's wrongdoing created a tempting opportunity for free radicals.¹⁰⁵ Proximate cause is preserved under the Dependent Compliance Error ("DCE") paradigm if the defendant's wrongdoing has increased the likelihood that the victim will be harmed by someone else's inadvertent negligence.¹⁰⁶ A blended attack would not fall into the DCE paradigm if the second wrongdoer acted intentionally.

¶ 83 Proximate cause is broken under the No Corrective Precaution paradigm if a third party with an opportunity and duty to prevent the plaintiff's harm intentionally fails to do so. If, for instance, the plaintiff intentionally fails to take a corrective precaution that would have prevented the harm, such failure would cut off the original tortfeasor's liability.¹⁰⁷

¶ 84 As the name suggests, the Independent Intervening Tort paradigm cuts off the original tortfeasor's liability if an independent intervening tort caused the plaintiff's harm. Under this paradigm, the original tortfeasor's liability will be cut off if the relation between the original tortfeasor's negligence and the second defendant's subsequent negligence is coincidental.¹⁰⁸

¶ 85 The victim of a blended attack, as the plaintiff in a negligence action, would be interested in preserving the liability of a solvent original tortfeasor. There are three direct consequences paradigms that, if applicable, may preserve the liability of the original tortfeasor: the No Intervening Tort ("NIT") paradigm, the Dependent Compliance Error ("DCE") paradigm, and the EFR paradigm. The NIT and DCE paradigms do not apply to the typical blended attack case, leaving the EFR doctrine. The EFR doctrine would preserve the liability of an original tortfeasor if she encouraged free radicals and if the factors that influence courts in holding a defendant liable for encouraging free radicals are present in the case. If, however, the second tortfeasor is a responsible person who deliberately omitted a reasonable precaution or committed an intentional tort or crime, the original tortfeasor's liability will be cut off.

¶ 86 The EFR paradigm, to which we now turn, is therefore the most relevant paradigm in the liability analysis of a blended cyber attack.

¶ 87 Courts hold rational and "irrational" defendants equally liable for their torts. Actors with a severe mental illness, for instance, are not exempted from liability. Mentally incompetent people are held to the standard of normal people, even though they could never achieve it. In Polamtier v. Russ,¹⁰⁹ for instance, a legally insane paranoid schizophrenic defendant was held liable for shooting his father-in-law.¹¹⁰ The court reasoned that, in spite of his mental illness, he could nevertheless form the intent to commit his unlawful act.¹¹¹

¶ 88 We observe the same pattern in negligence law. People with mental illnesses are held to the negligence standards of normal people. In Breuning v. American Family Insurance Co.,¹¹² a person started experiencing delusions, but continued driving her car and caused an accident.¹¹³ The court reasoned that a reasonable person should have seen the delusions as a danger signal, and that continuing to drive therefore constituted negligence.¹¹⁴

¶ 89 The courts do distinguish between rational and irrational actors when the actor is encouraged by a rational defendant. Courts hold a rational defendant liable for encouraging or provoking an irrational person, but cuts off the encourager's liability when the provoked actor is rational.¹¹⁵ The rationale for this distinction appears to be rooted in the deterrence and insurance goals of tort law.

¶ 90 Negligence law is the most basic form of safety regulation, but it is an ineffective deterrent against defendants who are shielded from liability by anonymity, insufficient assets, lack of mental capacity, or lack of good judgment. Such trouble-prone individuals are termed "free radicals" because of their tendency to bond with trouble. Examples of free radicals include children, anonymous crowds, criminals, mentally incompetent individuals, and in the cyber realm, hackers and cyber rogues, such as computer virus authors and distributors.¹¹⁶

¶ 91 Free radicals are not deterred by the threat of tort liability. Judgment-proof free radicals have insufficient assets to pay for the harms they cause, while other free radicals simply lack the good judgment or mental capacity to care about the consequences of their actions.¹¹⁷ For instance, terrorists may be blinded to the threat of liability by ideological or religious motivations. The deterrence rationale of negligence law would therefore be defeated if responsible people who foreseeably encourage free radicals to be negligent were allowed to escape judgment by shifting liability to the latter. Common law negligence rules have responded to this policy dilemma with the Encourage Free Radicals ("EFR") doctrine. The EFR doctrine imposes liability on the encourager, even when intentional or criminal behavior by a free radical intervenes.¹¹⁸

¶ 92 Satcher v. James H. Drew Shows, Inc.¹¹⁹ illustrates the Free Radicals paradigm. In Satcher, the plaintiff bought a ticket for a ride on the bumper cars in an amusement park. A group of mental patients on an outing joined the plaintiff's group on the bumper cars. When the ride started, the patients surrounded the defendant and repeatedly crashed into her from all sides, injuring her neck permanently. The plaintiff filed suit, alleging that the defendant owner and operator of the ride had been negligent in allowing the patients to target and injure her. The appellate court reversed the trial court's decision for the defendant, on the grounds that the defendant had encouraged free radicals.

¶ 93 If the plaintiff had sued the mental patients, she likely would have won. Their mental illness would not have been a defense, although they may not have had the assets to pay a judgment. Their mental illness is a critical factor in the liability of the owner defendants. If the bumper car drivers were rational individuals, instead of free radicals, the defendant would probably not have been held liable.¹²⁰ Tort law focuses liability on responsible people, which is where its policy goals will be best promoted.

¶ 94 Weirum v. RKO General, Inc.¹²¹ presents another free radicals case. The defendant radio station broadcast a contest in which a disc jockey would drive throughout Los Angeles. He would stop occasionally and announce his location on the radio. Teenagers would race to meet the disc jockey, and he would give a prize to the first one who reached him. Eventually, two overeager racing teenagers were involved in a road accident, killing the plaintiff's deceased. There were two concurrent efficient causes of the accident, namely the organizers of the contest and the reckless teenage drivers. The court determined that the radio station was negligent in encouraging the free radical teenagers to drive recklessly. The wrongdoing of the teenagers did not therefore cut off the defendant radio station's liability. The court held the radio station jointly liable with the teens and, as the deeper pocket, was likely required to pay most of the damages.

¶ 95 The EFR doctrine is not a modern development but has a long history. Mark Grady details the ways the doctrine developed a critical mass throughout the nineteenth century, as did negligence cases generally.

¶ 96 One of the earliest cases in which a court applied the EFR doctrine is the famous 1773 English case Scott v. Shepherd.¹²² The defendant threw a lighted squib, made of gunpowder, into a crowded marketplace. The squib was picked up and thrown away successively by several people, until it landed elsewhere in the market where it exploded and injured the plaintiff. The verdict turned on whether the harm was direct (trespass vi et armis) or consequential (trespass on the case.) If the original throwers had acted out of self-defense or necessity, the harm would be considered to have been a direct consequence of the defendant's first throw of the squib. If the intermediate throwers, on the other hand, had acted to "continue the sport" as true free radicals would, then the harm would be consequential or indirect. Justice Blackstone argued for the free radical interpretation, but was outvoted by his colleagues. The court held for the plaintiff because he had decided to plead trespass vi et armis.¹²³

¶ 97 The English case Dixon v. Bell¹²⁴ may have been the original EFR case.¹²⁵ The defendant sent his thirteen-year-old servant to fetch a loaded gun he had kept in his apartment. Assuming the gun was unloaded, the servant playfully pointed it at the plaintiff's son and pulled the trigger. The gun went off, injuring the boy. The plaintiff's declaration based its claim of liability on the allegation that the defendant had encouraged a free radical: the "too young" servant was "an unfit and improper person to be sent for the gun."¹²⁶ In particular, the allegation claimed that the defendant had wrongfully sent a juvenile servant to fetch a loaded gun, fully aware that it was inappropriate and dangerous.

¶ 98 Lynch v. Nurdin¹²⁷ was decided in 1841, during the full swing of the Industrial Revolution. The defendant had left his horse and cart unattended on a usually busy street on a day when it was even more bustling than usual. The defendant knew that groups of children would be coming down the street and that they would be attracted to his horse and cart. A young child was injured when another boy, who was playing on the cart, caused it to jerk forward and run over the boy's leg. The Queen's Bench held the defendant liable for the boy's injuries because he had provided an opportunity and encouragement to free radicals.¹²⁸

¶ 99 Guille v. Swan¹²⁹ was possibly the first EFR case in the United States. In Guille, the defendant descended in a balloon over New York City into the plaintiff's garden in a manner that attracted a crowd. The defendant's balloon dragged over the plaintiff's garden, but the stampeding crowd did far more damage to the garden. The defendant argued that he should be responsible only for his share of the damages and not for that caused by the crowd, but the court held him responsible for all the damages. People who are otherwise perfectly rational may behave differently when they are shielded by the anonymity and diminished accountability of a crowd, and the crowd acted as a free radical in that particular situation. Chief Justice Spencer's statement that the defendant should be held responsible because his manner of descent would foreseeably draw a crowd with predictable consequences is a classic description of the EFR doctrine.¹³⁰

IV. Free radicals, the buffer overflow and blended attacks

¶ 100 In a negligence action, the liability of an original tortfeasor for encouraging a second tortfeasor will be preserved under the EFR doctrine, if: (i) the second tortfeasor is in fact a free radical, and (ii) the case exhibits the factors that influence courts in holding a defendant liable for encouraging free radicals. We now turn our analysis to these two issues in the context of a blended attack.

¶ 101 The EFR doctrine only applies when a free radical is involved. If the encouraged person is not a free radical, and if the defendant's encouragement is insufficient to make him a co-actor with the immediate wrongdoer, the defendant is immune from liability. A defendant would, for instance, not be held liable for encouraging a responsible citizen. If Bill Gates had responded to the Weirum radio broadcast by racing to collect the prize, his intervening conduct would almost certainly have cut off the radio station's liability.¹³¹ Likewise, in the unlikely event that Bill Gates would use a virus kit to create a virus that exploits a security flaw in Windows, the creator of the kit would escape liability. If, however, a free radical, such as a judgment-proof hacker, did the same, the hacker's intervention would likely not break proximate causality.

¶ 102 Seith v. Commonwealth Electric Co.¹³² presents a case where a non-free radical intervened, cutting off the liability of the defendant. In Seith, because of negligent maintenance, a live electric wire broke and fell on a sidewalk.¹³³ Two police officers came to investigate, and one of them flipped the wire with his club towards the plaintiff, a bystander.¹³⁴ The plaintiff caught it reflexively and suffered a severe electric shock as a result.¹³⁵ The trial court found for the plaintiff, but the Illinois Supreme Court reversed on the grounds that the police officer, as a model of propriety and responsibility, was not a free radical.¹³⁶ If a free radical, such as a child or mentally incompetent person, had flipped the wire to the plaintiff, the defendant would likely have been held liable.¹³⁷

¶ 103 The second tortfeasors in a blended attack are typically virus authors and distributors who have exploited a security vulnerability to launch a cyber attack. The liability of the original tortfeasor will be preserved if the exploiters of the vulnerability are free radicals. We now turn to an analysis of virus authors and distributors as free radicals.

¶ 104 Virus authors and distributors have properties commonly associated with free radicals. They are often judgment-proof and shielded by the anonymity of cyberspace. Furthermore, virus attacks are under-reported and under-prosecuted, and the probability of catching a hacker or virus author is comparatively low. Virus authors appear undeterred by the threat of legal liability and often seem unconcerned about the problems caused by their creations. Most virus authors would either be unaffected or, perversely, actually encouraged by stricter anti-virus legislation. All of these factors are consistent with a free radical profile.¹³⁸

¶ 105 The Internet provides users with a degree of anonymity that has emboldened cybercriminals to commit crimes they would not otherwise consider.¹³⁹ The anonymity of cyberspace complicates the task of detecting computer crimes and tracking down offenders. It also makes it harder to obtain evidence against a wrongdoer such as a virus author or distributor.¹⁴⁰ Cyberspace provides the technology and opportunity to a skilled operator to assume different identities, erase his digital footprints, and transfer incriminating evidence electronically to innocent computers, often without leaving a trace.¹⁴¹

¶ 106 Suppose, for instance, a virus were transmitted from the e-mail account of someone named Jill Smith, and a copy of an identical virus were tracked down in the same account. This evidence may look like the proverbial smoking gun, but would likely not prove by a preponderance of the evidence that the owner of the account was the actual culprit. Someone may have hacked into the Smith account, used it to launch a virus, and stored incriminating files in the account.¹⁴²

¶ 107 Perpetrators of denial-of-service attacks employ similar tactics to hide their identities. The most common form of a distributed denial-of-service attack¹⁴³ consists of flooding a network with bogus information packets, thereby preventing legitimate network traffic. The source addresses of this illegitimate network traffic are usually spoofed to hide the true origin of the attack, making it difficult to identify the attacker. This is especially true with distributed attacks.¹⁴⁴

¶ 108 As the number of machines connected to the Internet increases, the ability of hackers to elude detection is enhanced. Subverting multiple machines makes it difficult to trace the source of an attack. An attacker can take a circuitous route and hide his tracks in the adulterated log files of multiple machines, which would reduce the likelihood of detection and allow the attacker to remain hidden from law enforcement.¹⁴⁵

¶ 109 The anonymity of cyberspace has contributed to virus authors' graduation from cyber-vandalism to organized crime. Virus writers are increasingly cooperating with spammers and hackers to create viruses to hack into computers to steal confidential information, often hiding their identity by spoofing the identity of the legitimate owner. Spammers are using viruses, for instance, to mass distribute junk mail by sending out viruses to take over computers and e-mail accounts and then using them to mass-distribute spam messages.¹⁴⁶ The owner of the hijacked computer usually does not know that it has been hijacked, although there are often subtle indications, such as a slower Internet connection.¹⁴⁷

¶ 110 E-mail plays a prominent role in computer security. It is currently the most widely used mechanism for virus transmission,¹⁴⁸ as well as a prime means to install backdoors and other malicious programs in target systems.¹⁴⁹ E-mail is a popular mechanism for transmitting viruses embedded in Word macros (such as Melissa), infected attachments (such as Love Bug), and viruses embedded in HTML mail. Technology that enables anonymous e-mail transmission would therefore be a significant tool in the hands of cyber rogues.

¶ 111 E-mail anonymity is substantially enhanced by the use of anonymous remailers. Remailers are servers which forward electronic mail to network addresses on behalf of an original sender who wishes to remain anonymous. An e-mail message usually carries a header with information about its starting point, its destination, and the route it has taken. This information makes the true source of the message traceable. The purpose of a remailer service is to disguise an e-mail's true source by delivering the message without its original header and with a fictitious return address. This ensures almost total anonymity for the original sender.¹⁵⁰

¶ 112 The remailer typically receives a message from A that is intended to be transmitted to B. The remailer then transmits the message to B, but in such a way that the true source (A) is obfuscated.¹⁵¹ Some remailer services enable the recipient to reply to the true source, but without revealing the identity of the source. A virus may be circulated anonymously in this manner, by remailing an e-mail message with an attachment containing the virus.¹⁵²

¶ 113 Remailers come in different varieties and levels of anonymity.¹⁵³ Some remailers maintain an internal list of the true identities of their clients. Any client of the remailer is in principle identifiable by someone with access to the internal master list. The former anonymous remailer, penet.fi, operated in this way, which ultimately led to its demise when a court ordered that the client list be made available to a plaintiff in a lawsuit. Pseudonymous remailers, generally termed "nym servers," use cryptography to provide the same service but with a greater degree of confidentiality.

¶ 114 The purpose of keeping a list of client identities is to facilitate two-way interaction. When the remailer receives a message intended for one of its clients, the remailer consults the list and forwards the message to the client. If e-mail users are willing to forego two-way interaction, such a master list is no longer necessary and greater confidentiality can be achieved. When a message is remailed anonymously under such an arrangement, it leaves no information behind that can be used to trace it to the original sender. A determined sender can use "chained remailing" as an additional line of defense, namely a combination of anonymous remailers and encryption techniques to make it virtually impossible to trace her communications.

¶ 115 Remailers fulfill an important function in attacks that rely on e-mail as a propagation mechanism, as is often the case in blended attacks. A buffer overflow vulnerability in the e-mail servers Microsoft Outlook and Outlook Express, for instance, enabled an attacker to invade a target computer by sending an infected e-mail message. The malicious code could be executed merely by reading the transmitted HTML message, without opening an attachment. As soon as the recipient downloaded the infected message from the server, Outlook would crash and the viral code would activate. The infected e-mail message would then be sent to all the contacts in the victim's address book. The process would repeat itself, repeatedly causing e-mail clients to crash and occasionally paralyzing Internet traffic as a result.¹⁵⁴

¶ 116 Anonymous e-mail protected and encouraged the perpetrators of this exploit and countless others. The anonymity provided by remailing drastically reduces accountability and deterrence on the Internet and is increasingly a hindrance to law enforcement efforts.¹⁵⁵

¶ 117 Perpetrators of virus attacks appear to be undeterred by the threat of legal action. In a leading study on the subject, Dr. Sarah Gordon examined the correlation between the number of new viruses in the wild and the high profile prosecutions of virus authors, as a measure of the deterrence value of prosecution. Dr. Gordon reports that high profile prosecutions have had a limited deterrent effect.¹⁵⁶

¶ 118 Dr. Gordon's conclusions were corroborated by a survey by the same author, in which virus authors and anti-virus researchers were asked whether the arrest and prospective sentencing of the Melissa author would have any impact on the virus writing community. All virus authors interviewed stated that there would be no impact, immediate or long-term, while the anti-virus researchers were evenly split between whether the arrest would or would not have any impact. These results are consistent with those of comparable surveys by other researchers.¹⁵⁷

¶ 119 The results of a subsequent survey on the impact of anti-virus legislation on virus authors, suggest that new laws may, perversely, result in more viruses than before. According to the survey results, a majority of virus authors would either be unaffected or actually encouraged by anti-virus legislation. A significant number of the virus authors interviewed claimed that the criminalization of virus writing would actually encourage them to create computer viruses, perhaps as a form of protest or civil disobedience.¹⁵⁸

¶ 120 Laws against virus authors cannot be effective unless virus incidents are reported and perpetrators are prosecuted. There is evidence that virus crimes are seriously under-reported and, as a consequence, under-prosecuted.¹⁵⁹ Companies tend to be reluctant to report security breaches, such as virus attacks, perhaps to avoid negative publicity.¹⁶⁰ Firms seem particularly reluctant to report and prosecute cybercrimes that originate from overseas.¹⁶¹

¶ 121 Commenting on the ineffectiveness of the law to combat computer viruses, Grable writes, "[b]oth the federal and New York state criminal statutes aimed at virus terror are ineffective because... [t]he combination of the lack of reporting plus the inherent difficulties in apprehending virus creators leads to the present situation: unseen and unpunished virus originators doing their damages unencumbered and unafraid."¹⁶²

¶ 122 Law enforcement is continually hampered by the law's lagged response to technological advances. Philippine authorities were unable to prosecute the creator and distributor of the destructive Love Bug virus, for instance, because at the time there existed no criminal statute under which he could be prosecuted.¹⁶³

¶ 123 Creative cybercriminals often find a way to circumvent one or more elements of newly enacted statutes. In United States v. LaMacchia,¹⁶⁴ for instance, an MIT student was prosecuted for distributing copyright-protected software for free over the Internet. The case was dismissed because of lack of evidence of financial gain by the defendant, an essential element of the criminal wire fraud act under which he was prosecuted. This loophole was eventually eliminated by the No Electronic Theft Act of 1997.¹⁶⁵

¶ 124 The conclusion that may be drawn is that virus authors have the Teflon-like deterrence properties commonly associated with free radicals. They are often judgment-proof and shielded by the anonymity of cyberspace, are increasingly motivated by crime, and appear unconcerned about the problems caused by their creations. Currently, most virus and blended attacks depend on e-mail. Such attacks are aided by numerous technologies that enable anonymous e-mail transmission. Furthermore, virus attacks are under-reported and under-prosecuted, and virus authors, to a significant degree, appear to be unconcerned and, perversely, often encouraged by the threat of legal liability and tougher laws.

¶ 125 There are additional factors, besides the requirement that the second tortfeasor be a free radical that courts look at before they hold a primary tortfeasor liable for encouraging free radicals. As a threshold requirement, the defendant's encouragement of free radicals must have been negligent before liability will be imposed.

¶ 126 A defendant will not be held liable for encouraging free radicals unless the encouragement was negligent. The encouragement must therefore have been a breach of duty to the plaintiff. Courts require a plaintiff to prove breach of duty by identifying an untaken precaution that would have prevented the harm and showing that the untaken precaution would have yielded greater benefits in accident reduction than its cost. The CodeRed attack presents an illustrative example.

¶ 127 The Windows IIS vulnerability that enabled the CodeRed attack sequence was discovered on June 18, 2001.¹⁶⁶ A security patch to fix the vulnerability was promptly issued by Microsoft. The first version of CodeRed that exploited the vulnerability appeared approximately one month later.¹⁶⁷ Due to a programming flaw, the first version of CodeRed did not spread as quickly nor as widely nor do as much harm as its creator had apparently hoped for.

¶ 128 At this stage, after the first exploitation, the existence of the vulnerability was common knowledge in the IT community, a patch to fix it had been made available, and the first CodeRed attack, at the very least, alerted the IT community to the exploitability of the vulnerability and the harm it could cause. The IT community was also aware of the programming flaw in CodeRed that limited its effectiveness and that the flaw could easily be fixed. The damage that a debugged version of CodeRed could do with the assistance of the Windows IIS vulnerability was therefore foreseeable.

¶ 129 A second, more virulent version of CodeRed appeared on July 19, 2001. The programming flaw that plagued its predecessor was fixed in this version, and the second version, predictably, caused substantially more harm than its predecessor.

¶ 130 An IT manager who failed to implement a security patch to fix the Windows IIS vulnerability, after the first CodeRed attack, may be held liable for negligently encouraging free radicals who subsequently exploited the flaw in his system to cause harm to other users. A breach of duty analysis would begin by considering an untaken precaution that would have avoided the second CodeRed infection. The most logical and probably most effective precaution would be implementation of the security patch provided by Microsoft. The CodeRed worms could, for instance, not infect a system that had the Microsoft MS01-033 patch installed.¹⁶⁸ A commentator opined that "[t]here was so much publicity and so many published articles by the time CodeRed-II hit, that any competent server manager would have had ample opportunity to patch their systems in time."¹⁶⁹

¶ 131 The plaintiff would need to show that implementing the patch would have yielded greater benefits in accident reduction than its cost. The benefits would include avoidance of the foreseeable harm from further exploitation of the vulnerability. After the appearance of the first version of CodeRed, a reasonably competent IT professional knew or should have been able to infer the potential harm from further and more efficient exploitation of the vulnerability.

¶ 132 The expected harm avoided must be weighed against the cost of implementing the patch. Although security patches are usually made available for free to users, implementing them may be costly and difficult, especially in large corporations with complex systems. Patches also tend to interact with and affect the systems to which they are applied, sometimes impairing their performance.¹⁷⁰

¶ 133 Although a conclusive resolution of the cost-benefit tradeoff would require a detailed numerical analysis of the costs and benefits involved, it appears that the defendant in this hypothetical was likely negligent in failing to implement the security patch. Implementing the security patch and dealing with and fixing bugs introduced by interaction between the patch and the regular system appear minor compared to fixing the harm from an attack by a blended threat, such as CodeRed and its potential successors.

¶ 134 In the only two significant cases known at the time of writing that involved failure to apply security patches, such failure was considered a breach of duty. In April 2003, the Maine Public Utilities Commission concluded that the telecommunications company Verizon had acted unreasonably by failing to apply a security patch issued six months earlier by Microsoft.¹⁷¹

¶ 135 In a suit against the clothing retailer Guess?, the U.S. Federal Trade Commission ("FTC") claimed that Guess? had failed to patch a web site vulnerability that made it susceptible to SQL injection attacks by the SQLSlammer worm. The complaint alleged that the Guess? web site had been vulnerable to commonly known or reasonably foreseeable attacks from attackers seeking access to confidential customer information.¹⁷² The vulnerability was successfully exploited by a cyber attacker, resulting in access to a database containing 191,000 credit card numbers. The FTC alleged that Guess? had been aware of the vulnerabilities since October 2000, that the stolen information was sensitive, and the fix relatively cheap and easy. This essentially alleges failure to take a cost-effective precaution, which suggests a breach of duty. The case was subsequently settled in June 2003.¹⁷³

¶ 136 In addition to the negligence requirement, the following factors influence courts in holding a defendant liable for damage caused by encouraging free radicals:¹⁷⁴

The fact that the defendant's encouraging behavior was deliberate, as opposed to inadvertent, was considered important in some cases.

The defendant had a special relationship with the free radical, the victim, or both.

¶ 137 The defendant's encouragement of the free radical must have been substantial for liability to be imposed. Courts have interpreted "substantial encouragement" in terms of the likelihood of provoking harmful behavior by free radicals.

¶ 138 Contrast, for instance, Segerman v. Jones¹⁷⁵ with Home Office v. Dorset Yacht Co.¹⁷⁶ In Dorset, seven boys who had been sentenced to work in a boot camp for juvenile offenders were working under supervision of three Home Office guards. One evening, in breach of their instructions to watch the boys, the guards simply went to bed, leaving the boys unsupervised. The boys swam out to an unattended yacht moored nearby and managed to set it in motion. They collided with another yacht owned by the plaintiffs, who sued Home Office for the resulting damage. The trial court ruled in favor of the plaintiff, and the Court of Appeal affirmed.

¶ 139 Lord Reid stated that Home Office would be liable if it appeared very likely, ex ante, that the boys would damage property if they were to escape from supervision. It is plausible to assume a highly foreseeable likelihood of escape and harmful behavior by the delinquents. The boys were juvenile offenders, with records including convictions for breaking and entering, larceny, and grand theft auto. Given, in addition, that five of the seven had a record of previous escapes from boot camp, the inference of substantial encouragement appears justified.

¶ 140 In Segerman, the defendant teacher left her classroom for a few minutes. During her brief absence, one student kicked out the teeth of one of his classmates. The Maryland Supreme Court held that the teacher was not liable. The extent of her encouragement of the children was leaving them to their own devices for a few minutes. This encouragement was too slight to impose liability. The free radicals in this case were ordinary school children and were obviously not in the league of the juvenile offenders of Dorset.

¶ 141 Similar cases have denied liability for leaving a stake at a construction site,¹⁷⁷ for leaving a screwdriver out in a yard,¹⁷⁸ and for leaving a load of dirt clods out in a backyard.¹⁷⁹ In these cases, the court apparently considered the likelihood of harm in each situation to be insignificant and the encouragement therefore insubstantial.

¶ 142 In a blended attack, the original tortfeasor typically encourages free radical cyber rogues by making a tempting and exploitable security vulnerability available. The likelihood that a vulnerability would be found and exploited to perpetrate a cyber attack is a proxy for the substantiality of the encouragement inherent in the vulnerability. We argue that buffer overflows are likely to be discovered rapidly when they become available, and, once identified, promptly exploited. This conclusion is supported by empirical evidence.

¶ 143 It is unlikely that a valuable and exploitable computer security vulnerability, such as a buffer overflow, will remain undiscovered for long. Worms and viruses employed in blended threats are programmed to automatically search for and locate exploitable vulnerabilities. Furthermore, technologies are available to assist software designers in identifying security vulnerabilities in their products.¹⁸⁰ Although such vulnerability-identifying technologies are intended to assist designers of "legitimate" software in troubleshooting and debugging, the technologies are, of course, equally available to designers of malevolent code.

¶ 144 Once an appropriate vulnerability is identified, it will likely be exploited. In his recent treatise on buffer overflow attacks, James Foster comments that "[i]t's no coincidence that once a good exploit is identified, a worm is created. Given today's security community, there's a high likelihood that an Internet worm will start proliferating immediately. Microsoft's LSASS vulnerability turned into one of the Internet's most deadly, costly, and quickly proliferating network-based automated threats in history. Multiple variants were created and released within days."¹⁸¹ In fact, current trends and patterns of infection suggest that the time lag between discovery of a vulnerability and its exploitation is shrinking.¹⁸²

¶ 145 Factors that contribute to the prompt exploitation of buffer overflow vulnerabilities include their ease of exploitation and their convenient properties that give cyber rogues exactly what they need. These properties include a buffer overflow's convenient configuration as a gateway to inject and execute attack code and then assume unauthenticated remote control of a system or network, including root control.¹⁸³

¶ 146 A vulnerability may be considered easy to exploit if no special programming skills are necessary to take advantage of it, or if the necessary exploit code is publicly available.¹⁸⁴ Writing a successful buffer overflow exploit takes considerable programming skills, but the code for such an exploit is often publicly available and accessible, even to individuals without technical sophistication. As new buffer overflow vulnerabilities are discovered, exploits are habitually published shortly after the discovery.¹⁸⁵ Technical articles continuously appear, describing vulnerabilities and how to exploit them, often in substantial detail.¹⁸⁶

¶ 147 Advisories have reported multiple buffer overflow vulnerabilities that were either trivial to exploit or for which exploit code was publicly available. A vulnerability in the Solaris KCMS Library Service System, for instance, was easy to exploit. Exploitation of this vulnerability could be accomplished by drawing on a standard and widely available software tool and on basic computer literacy.¹⁸⁷

¶ 148 A buffer overflow vulnerability in a version of the commercial program Hypermail was, likewise, easily exploitable.¹⁸⁸ Hypermail is an open-source program that converts e-mail messages into cross-linked HTML pages. The program contained a vulnerability that could be exploited simply by sending a malicious e-mail with an overly long attachment name. A detailed sample e-mail message that could trigger the overflow and control Hypermail's execution had been posted on the Internet.¹⁸⁹

¶ 149 Not all buffer overflow vulnerabilities are easy to exploit. A recent advisory describes, for instance, the SpamAssassin buffer overflow as "challenging to exploit," depending on the target computing environment. This particular vulnerability is not exploitable on all platforms. To succeed, a would-be attacker would have to identify and target victims who are using a vulnerable spam filter.¹⁹⁰

¶ 150 In summary, a security vulnerability such as a buffer overflow likely constitutes substantial encouragement to perpetrators of cyber crimes such as blended attacks. A valuable and easily exploitable vulnerability, such as the buffer overflow, is likely to be promptly discovered and exploited. The discovery of vulnerabilities is facilitated by technology: viruses used in blended attacks are often programmed to search for new vulnerabilities, and specialty software designed to identify security weaknesses in computer systems and networks is freely available. Attackers have a strong incentive to find and exploit vulnerabilities such as buffer overflows, because buffer overflows are easy to exploit and give attackers exactly what they need to launch a blended attack. Empirically, exploitation of buffer overflows is pervasive, both in an absolute sense as well as measured as a percentage of all blended attacks.¹⁹¹

¶ 151 The substantial encouragement inherent in security vulnerabilities has been recognized in the common law. For instance, in its suit against the retailer Guess? for failing to patch a security vulnerability, the FTC emphasized the foreseeability and high likelihood of exploitation of the vulnerability.¹⁹² The analysis in this section suggests that a buffer overflow vulnerability as substantial encouragement to free radical cyber rogues is closer to Dorset (encouragement likely to incite free radical juvenile delinquents) than to Segerman (encouragement insufficient to incite normal school children).

¶ 152 Courts are more likely to impose liability when the defendant has created a tempting opportunity that does not normally exist for the free radical. If a free radical already has several opportunities available for harmful behavior, the defendant's encouragement does not amount to a scarce opportunity.

¶ 153 A person flashing a wad of $100 bills, for instance, would probably not be liable for the harm caused by a fleeing thief who runs into and injures someone. Because of the availability to the thief of many other similar opportunities, the flash of money was not an unusually tempting opportunity to the free radical. If the person had not flashed the money, a determined thief would have found another equally attractive opportunity.¹⁹³

¶ 154 In Stansbie v. Troman,¹⁹⁴ the defendant, an interior decorator, neglected to lock the door of the house of a client. A burglar entered through the open door and stole the plaintiff's jewelry. The court held the defendant liable for the loss. The defendant had created an opportunity for the thief that did not normally exist, since valuables are normally kept locked up.

¶ 155 In a similar case, the defendant put a scaffold in place next to the plaintiff's apartment building. Armed robbers used the scaffold to gain entry to the plaintiff's apartment and stole his goods. The New York Supreme Court denied the defendant's petition for summary judgment. The defendant had encouraged free radicals by making a scarce and tempting opportunity available to them.¹⁹⁵ In an analogous case, involving information security, the bookseller Barnes and Noble allegedly permitted cyber rogues to gain unauthorized access to confidential client information through security vulnerabilities in its web site. Barnes and Noble entered into a settlement agreement with the New York Attorney General in April 2004.¹⁹⁶

¶ 156 Not all encouragement of free radicals necessarily presents a rare opportunity. In Gonzalez v. Derrington,¹⁹⁷ the defendant sold free radicals five gallons of gasoline in an open pail, in violation of a municipal ordinance which prohibited the sale of gasoline in open containers and in excess of two gallons. The free radicals subsequently used the gasoline to commit arson.¹⁹⁸ The court held the defendant not liable.¹⁹⁹ Providing the gasoline to the free radicals did not constitute a rare opportunity, as they could have siphoned the gasoline they needed from a car.²⁰⁰

¶ 157 This pattern of case law suggests that a scarce opportunity is one that: (i) is not part of the free radical's normal opportunity set, and (ii) is more tempting than existing opportunities, perhaps because it lowers the transaction costs of the free radical's harmful behavior. A thief can always use brute force to break into an apartment to do his business. However, an unlocked door or conveniently placed scaffold would be an unusual opportunity to lower his transaction costs because it requires less physical exertion, presents faster results, and is less likely to attract attention than a more forceful entry. The unlocked door and scaffold therefore fit the common law profile of a scarce opportunity. Selling gasoline to a free radical in an open pail, on the other hand, does not constitute a scarce opportunity. The alternative, siphoning the gasoline from a car, is not significantly more burdensome or costly.

¶ 158 Computer security vulnerabilities appear to be scarce opportunities. A buffer overflow vulnerability is analogous to the unlocked door in Stansbie. The unlocked door provides the convenient access normally reserved for someone with valid authentication, such as possession of a key. Similarly, a buffer overflow yields remote, unauthenticated, and root access to a target computer system, as well as the opportunity to inject malicious code into the system.²⁰¹ This kind of privileged access is normally reserved for the system administrator.

¶ 159 The unlocked door provides additional encouragement to the bricks-and-mortar intruder by lowering his transactions costs. Analogously, security vulnerabilities lower the cyber attacker's transaction costs by making attacks faster and more efficient, conveniently remotely executable, and less likely to be blocked before substantial harm is done.²⁰² Skillful exploitation of security vulnerabilities therefore allows attackers with fast-spreading worms, such as Nimda, to achieve more destruction in less time and with lower computational expenditure.²⁰³ Security vulnerabilities conform to the common law pattern of scarce opportunities in free radical cases. They present opportunities not available in the normal functionality of computers and, analogous to an unlocked door or a scaffold, offer privileged access and convenience to an attacker. Security vulnerabilities also lower the transactions costs of a cyber vandal by offering greater potential harm in less time and with a lower likelihood of timely detection. The opportunities presented by a buffer overflow therefore appear to be more closely related to Russo (scaffold) or Stansbie (unlocked door), than to Derrington (pail of gasoline).

¶ 160 Foreseeability is a touchstone of proximate cause. A crisp formulation of the proximate cause requirement is that the realized harm must be within the scope of risk foreseeably created by the defendant, and that the plaintiff must belong to the class of persons foreseeably put at risk by the defendant's conduct.²⁰⁴

¶ 161 Consistent with the spirit of proximate cause, the behavior of free radicals must be foreseeable or predictable to hold their encouragers liable. If a radio station organizes a contest that encourages teenagers to race to catch up with a roving disc jockey, and they do in fact race and cause an accident, the organizers of the contest will likely be held liable.²⁰⁵ If the free radical goes too far or otherwise acts in an unpredictable manner, the defendant will escape liability. If one of the contestants had shot the other, for instance, the radio station would not be held liable.

¶ 162 In Bansasine v. Bodell,²⁰⁶ the defendant provoked a driver known for his aggression. The provoked driver shot the plaintiff's deceased. In spite of the driver's known aggressive tendencies, the defendant was found not liable.²⁰⁷ The court stressed that the defendant could not foresee that a driver would fire a gun at him for shining his high beams on the driver.²⁰⁸ The free radical driver's reaction, even for a person known for his fiery temperament, went beyond the encouragement of the defendant.²⁰⁹

¶ 163 The exploitation by cyber attackers of vulnerabilities, such as the buffer overflow, is foreseeable. It was argued in Section IV.C.2.a. ("Substantial encouragement") that buffer overflows are likely to be discovered rapidly when they become available and, once identified, promptly exploited.

¶ 164 Foreseeability of exploitation of buffer overflows is confirmed by the empirical pervasiveness of such exploits and the IT community's awareness of it. Buffer overflows are currently, and have been for a decade or so, the most commonly exploited security vulnerability and the most common way for an attacker outside of a target system to gain unauthorized access to the target system.²¹⁰ If buffer overflows were eliminated, the incidence of security breaches would be drastically reduced.²¹¹ The computer security community is indeed aware of the exploitability of and hazards associated with buffer overflows. James Foster opines that "[b]uffer overflow vulnerabilities are the most feared of vulnerabilities from a software vendor's perspective. They commonly lead to internet worms, automated tools to assist in exploitation, and intrusion attempts."²¹²

¶ 165 Statistics reported in security advisories confirm the dominance of the buffer overflow as the exploit of choice. A third of the investigative advisories spanning from September 2002 through March 2004 were related to buffer overflows (224 out of 659.)²¹³ In the year 2003 alone, approximately seventy-five percent of all CERT advisories were related to buffer overflows.²¹⁴ A Symantec security bulletin reported that blended attacks accounted for sixty percent of malicious code submissions during the first half of 2003, most employing a buffer overflow.²¹⁵ Other advisories and security reports exhibit a similar pattern.²¹⁶

¶ 166 Serious buffer overflow problems are not limited to small and resource-deprived companies, but have also plagued the products of large and well-known software vendors. Advisories of companies such as Apple and Oracle were nearly all related to buffer overflows, while half the advisories of Cisco, Microsoft, and Sun were related to overflows.²¹⁷

¶ 167 Exploitation of buffer overflow vulnerabilities is foreseeable, because it is well known that new vulnerabilities are likely to be discovered and exploited, and that actual exploitation is in fact pervasive. Courts ruling in lawsuits against encouragers of free radical cyber attackers have emphasized foreseeability of the free radical behavior. In April 2003, the Maine Public Utilities Commission denied Verizon Maine's request for a waiver of performance standards on its network for the month of January 2003.²¹⁸ Verizon had failed to meet certain performance standards on its network because of an attack by the SQL Slammer worm.²¹⁹ The Commission concluded that the company had failed to take reasonable precautions to fix the vulnerability that allowed the worm to penetrate its network.²²⁰ The Commission cited evidence of the foreseeability of the Slammer worm attack. The complaint also referred to security alerts issued by Microsoft, including recommended use of a software patch that would have prevented the attack.²²¹

¶ 168 The courts do not usually allow free radicals to recover for injuries they have caused to themselves. The courts only hold encouragers liable when free radicals injure third parties.

¶ 169 In Gilmore v. Shell Oil Co.,²²² the defendant's employee left a loaded gun within easy reach of a teenager. The teenager took the gun and shot and killed himself.²²³ Although the teenager had shot himself intentionally, it was clear from the circumstances of the case that, but for his ready access to the gun, he would not have done so.²²⁴ The trial court nevertheless entered summary judgment for the defendant, and the Alabama Supreme Court affirmed.²²⁵ Deterrence of this type of behavior is outside the policy objectives of tort law. The case may have been decided differently if the deceased had been a young child, as the EFR doctrine protects children against themselves.²²⁶

¶ 170 Courts are more likely to hold a defendant liable if the foreseeably encouraged harm is serious. Someone who has left explosives around children,²²⁷ for instance, is more likely to face liability than someone who has left a pile of dirt clods;²²⁸ and someone who fails to supervise juvenile delinquents is more likely to face liability than a school teacher who leaves ordinary school children momentarily to their own devices.²²⁹

¶ 171 The severity of a computer security breach resulting from a system vulnerability is a function of the following factors:

A. The degree of control over the affected system or network given to an exploiter of the vulnerability. The degree of control afforded by the vulnerability depends on: (i) the level of access it confers; (ii) the degree of remote exploitability it allows; (iii) its ease of exploitation; and (iv) the degree to which it enables circumnavigation of authentication requirements.²³⁰

B. Once the attacker has gained control, the kind and degree of harm such control allows the attacker to unleash.

¶ 172 The ultimate gift to a cyber attacker would be the most privileged level of access, namely full root-level access to the target system. "Root" is the conventional name of the superuser who has all rights in all modes on the computer system.²³¹ This is usually the system administrator's account. The superuser has privileges that an ordinary user does not have, such as authority to change the ownership of files, install and run programs, change web server databases, add, change, or delete system files or data, and change or replace web pages.²³² An attacker who gains system-level access inherits these privileges. Hence, if a program is already running with root privileges, a buffer overflow could hijack the program and transfer root control to the attacker.²³³ The attacker would then effectively become the administrator of the system. Exploitation of buffer overflows commonly yields root access to the attacker. The Linux application DosEMU, for instance, had a buffer overflow vulnerability that assisted an attacker in gaining root access.²³⁴

¶ 173 A vulnerability allows remote exploitability when it enables a user to access and execute commands on a remote system, as if the user were connected to a direct terminal on the system. Buffer overflow vulnerability is the perfect springboard to gain remote access to a target system because it allows a remote attacker to inject malevolent code, such as an e-mail worm, directly into the execution path of the remote system.²³⁵ The viral code could then create further opportunities for other remote attackers. The Nimda worm, for instance, attacked via backdoors left by worms such as CodeRed.²³⁶

¶ 174 Remote exploitation of buffer overflows has recently been reported in well known products, such as Sendmail, various Microsoft products, and, ironically, PGP.²³⁷ A vulnerability in Microsoft's Internet Explorer browser, for instance, allowed a properly formatted HTML document to cause a buffer overflow. This flaw could be exploited to allow an attacker to execute arbitrary code on the affected system, including malicious code, with the privileges of the user running Internet Explorer. This vulnerability was remotely exploitable.²³⁸

¶ 175 As discussed in Section IV.C.2.a. ("Substantial encouragement"), a vulnerability is easily exploited if a would-be attacker does not need technical sophistication or a complex exploit to use it, or if a suitable exploit is publicly available. We have argued that although not all buffer overflow vulnerabilities are necessarily easy to exploit, there are many that are easy. Even for difficult-to-exploit vulnerabilities, exploits are frequently publicly available soon after the vulnerability is discovered.²³⁹

¶ 176 The term "authentication" refers to the procedures by which a computer system verifies the identity of a party from whom it has received a communication. The login procedure is probably the best-known example of an authentication procedure. A login prompt asks the user to identify herself, followed by a request for a password. The system then authenticates the stated identity of the user by validating the password to see if the password and identity match. If they do not match, the user is restricted from accessing the system. Other examples of authentication include the requirement of a confirmation e-mail to activate an on-line account, ATM access, cryptographic authentication of a digitally signed contract, and biometric identification in applications such as Internet banking.

¶ 177 Authentication provides a line of defense against unauthorized access to a restricted system. A vulnerability that allows unauthenticated access may allow an attacker to bypass this line of defense. Network vulnerabilities, including buffer overflows, allow unauthenticated remote access to attackers without authentication.²⁴⁰

¶ 178 A remotely exploitable buffer overflow in Microsoft Data Access Components ("MDAC"), a system that provides database access for Windows platforms, was recently reported. The vulnerability enabled an attacker to run unauthenticated arbitrary code on an affected system.²⁴¹ The unauthenticated arbitrary code could, of course, be malicious.²⁴²

¶ 179 A vulnerability in Fusion News, a news management program for web servers, allowed remote unauthenticated attackers to create arbitrary user accounts on the Fusion News server by sending a specially crafted request to the server. If properly structured, the request could also be used to gain administrative access. Exploitation of this vulnerability was trivial. A ready-to-use sample server request was, for instance, available on the Internet.²⁴³ This vulnerability contained all the critical elements favorable to a cyber attacker: no authentication barriers, system administration-level (root) access, ease of exploitation, and the ability to execute malicious code.

¶ 180 The ultimate measure of the severity of a cyber attack is its economic impact. By this measure, blended threats, aided by buffer overflow vulnerabilities, are capable of considerable harm. The CodeRed family of blended attacks, although not the first of its kind, woke us up to the risks of remotely launched buffer overflow attacks.²⁴⁴ The first CodeRed worm caused billions of dollars of damage in just a few days, despite corporate firewalls and other defensive efforts. Worldwide harm caused by CodeRed is estimated at $2.62 billion.²⁴⁵ Subsequent blended attacks, such as Nimda, continued the trend. According to an estimate by consulting firm Computer Economics, Nimda infected more than 2.2 million servers and PCs in a twenty-four hour period during September 2001, causing damage of more than $590 million worldwide.²⁴⁶ A study by computer and communications consulting firm Aberdeen Group reports that annual productivity loss due to viruses and blended threats averages more than $200 per employee in the financial industry.²⁴⁷ Buffer overflow vulnerabilities are well represented in the SANS Top Twenty List of CERT security vulnerabilities,²⁴⁸ and are ranked fifth in the Top Ten Vulnerabilities by Orthus Information Security Solutions.²⁴⁹

¶ 181 In conclusion, security vulnerabilities, especially the buffer overflow, present opportunities for free radicals to do serious harm. The severity of the harmful behavior encouraged by a buffer overflow vulnerability is due to: (i) the degree and level of control over the affected system or network it affords an attacker; and (ii) once the attacker has control, the potential harm such control allows the attacker to unleash. Empirical data suggest that blended attacks do in fact exploit vulnerabilities to do considerable economic damage.

¶ 182 Negligence law distinguishes between deliberate and inadvertent failure to use a reasonable precaution, in EFR cases. A defendant is more likely face liability if he deliberately encouraged a free radical to do harm. However, even inadvertent encouragement will yield liability when the threatened harm is sufficiently serious and probable.

¶ 183 In Mills v. Central of Georgia Ry.,²⁵⁰ the defendant had left a signal torpedo on its tracks. A signal torpedo is an explosive device which blows up upon impact, such as when hit by an oncoming train. The purpose of putting a signal torpedo on the tracks is to warn crews working on railroad tracks of an approaching train. Upon workers leaving for the day, if a torpedo had not been detonated, it is supposed to have been picked up and put away. Contrary to this precaution, however, the torpedo in question was inadvertently left on the tracks. The plaintiff's sons found the torpedo, played with it, and injured themselves when it exploded. The Georgia Supreme Court ultimately found for the plaintiff.²⁵¹ Although the defendant created the opportunity inadvertently, the harm threatened was sufficiently serious and probable to justify imposing liability.

¶ 184 Liability is more likely when the defendant had a special relationship with the victim, the free radical, or both. This is consistent with general principles of negligence law. A hotel or common carrier, for instance, has a special duty to guard the best interests of its customers. Likewise, an airline would likely be held liable for negligently maintaining a highly disorganized baggage claim area that leads to injury of a passenger.²⁵²

¶ 185 A defendant who has encouraged free radicals through a nonfeasance as opposed to a misfeasance will not be liable, unless there was a special relationship. An individual who has advance knowledge of a cyber attack and who fails to warn an unrelated plaintiff, will not be liable to the unrelated plaintiff for any harm from the attack.

V. Discussion and conclusion

¶ 186 Information security threats are diversifying and evolving into multi-threat weapons that combine a variety of attack technologies and exploitation of security vulnerabilities. The blended attack exploits synergies between a multi-vector virus or worm and a computer security vulnerability, such as the buffer overflow, to enhance the effectiveness and destructiveness of its payload.

¶ 187 Blended attacks vary in complexity and technology, but they have two elements in common, namely a multi-vector worm or virus and exploitation of a security vulnerability. The skillful combination of the two elements creates synergies that make such attacks more hazardous than previous generations of malevolent code. The two salient elements of a blended attack focus the spotlight on the two most likely defendants in a civil action involving a blended attack: (i) the original tortfeasor responsible for the security vulnerability; and (ii) the second tortfeasor responsible for the malevolent code that exploited the vulnerability. The tortfeasors are concurrent efficient causes of the harm of the victim of a blended attack.

¶ 188 The direct consequences doctrine of proximate cause examines concurrent efficient causes to determine whether the second tortfeasor (the virus distributor) has cut off the liability of the first (the software vendor). An intervening crime or intentional tort, as is often the case in a cyber attack, normally cuts off the liability of the first tortfeasor. This is significant because the second tortfeasor, the exploiter of the vulnerability, is often judgment-proof or otherwise immune to liability, in contrast to the original tortfeasor. If liability were fixed exclusively on the second tortfeasor, it would leave the victim of a blended attack without recourse.

¶ 189 The Encourage Free Radicals ("EFR") paradigm of the direct consequences doctrine creates an exception if the second tortfeasor is a free radical. It fixes liability on the primary tortfeasor if she created an opportunity for free radicals to do harm. The policy objective of the EFR doctrine is to preserve the liability of individuals who are deterred by the threat of liability by preventing a solvent defendant from shifting liability to a judgment-proof individual who is not so deterred.

¶ 190 The analysis in this Article shows that virus authors and distributors who exploit security vulnerabilities to launch blended attacks have properties commonly associated with free radicals. An analysis of the technology and mechanism of blended attacks suggests that the factors that influence courts in finding a defendant liable for encouraging free radicals are present in a typical blended attack. Software designers and commercial vendors who are negligently responsible for security vulnerabilities in their products would likely be held liable for the harm caused by cyber rogues who exploit such vulnerabilities. This result is especially significant to plaintiffs who have suffered harm in a blended attack.

ENDNOTES

^* John Landerer Faculty Fellow, University of New South Wales, School of Law, Sydney, NSW 2052, Australia. mdv@unsw.edu.au. The author would like to acknowledge Mark Grady, Peter Szor, and the audience at Georgia Tech.

¹CodeRed infected over 300,000 machines within twenty-four hours. Peter Szor, The Art of Computer Virus Research and Defense 398 (Symantec Press 2005).

²The attacks occurred shortly after Microsoft had discovered the vulnerability and issued a patch to fix it. Microsoft, A Very Real and Present Threat to the Internet, http://www.microsoft.com/technet/archive/security/news/codealrt.mspx (last visited Oct. 31, 2005). Section II.C. infra discusses the principles of the buffer overflow vulnerability.

³Jeremy D. Baca, Windows Remote Buffer Overflow Vulnerability, SANS' Information Security Reading Room, Sep. 10, 2001, at 4, http://www.sans.org/rr/whitepapers/malicious ("With plenty of warning about the coming attack, the site's IP address was moved from 198.137.240.91 to 198.137.240.92.").

⁴CodeRed-II infected more than 359,000 computers in fourteen hours. Cooperative Association for Internet Data Analysis, CAIDA Analysis of Code-Red, http://www.caida.org/analysis/security/code-red (last visited Oct. 31, 2005).

⁵A backdoor is a method of gaining remote access to a computer, and is usually not detectable by casual inspection. See, e.g., Backdoor, Wikipedia, http://en.wikipedia.org/wiki/Back_door. A backdoor may, for instance, consist of a password recognition routine installed in the computer, perhaps as a modification of a legitimate program. The routine would enable a hacker who provided the right input to gain access to confidential files and programs on the computer.

⁶Baca, supra note 3, at 5.

⁷Attacks occurred despite the fact that a patch had been issued for the vulnerability by Microsoft before the first attack. The CodeRed-I as well as CodeRed-II worms could, for instance, not infect a system that had the MS01-033 patch installed. See, e.g., Baca, supra note 3, at 6 ("There was so much publicity and so many published articles by the time CodeRed-II hit, that any competent server manager would have had ample opportunity to patch their systems in time.") CodeRed-II also compromised devices with web interfaces, such as routers, switches, DSL modems, and printers. See, e.g., Cisco Systems, Inc., Cisco Security Advisory: Code Red Worm—Customer Impact, http://www.cisco.com/warp/public/707/cisco-code-red-worm-pub.shtml (last visited Oct. 31, 2005).

⁸eEye Digital Security, Analysis: CodeRed-II Worm, Aug. 2001, http://www.eeye.com/html/Research/Advisories/AL20010804.html.

⁹A denial-of-service attack ("DoS attack") is an attack on a computer system or network that causes a loss of service to users, typically the loss of network connectivity and services by consuming the bandwidth of the victim network or overloading the computational resources of the victim system.

¹⁰George V. Hulme, Nimda's Biography, InformationWeek, Sept. 19, 2001, http://www.informationweek.com/showArticle.jhtml;jsessionid=UICAMSSIJSSXAQSNDBGCKHSCJUMEKJVN?articleID=6506613 (last visited December 3, 2005).

¹¹Thomas M. Chen & Jean-Marc Robert, The Evolution of Viruses and Worms, in Statistical Methods in Computer Security 278 (William W.S. Chen ed., 2005), available at http://engr.smu.edu/~tchen/papers/statmethods2004.pdf ("Even if found, the worm [Nimda] was very difficult to remove because it makes numerous changes to Registry and System files. It creates an administrative share on the C drive, and creates a guest account in the administrator group allowing anyone to remote login as guest with a blank password.").

¹³Szor, supra note 1, at 366 ("Security exploits, commonly used by malicious hackers, are being combined with computer viruses, resulting in very complex attacks that sometimes go beyond the general scope of antivirus software."); id. at 542 ("[H]ighly infectious worms [such as CodeRed and Slammer] jump . . . over the Internet using buffer overflow attacks on networked services. Because the files need not be created on disk and the code is injected into the address space of the vulnerable processes, even file integrity systems remain challenged by this particular type of attack.").

¹⁴Negligence is the most widely used theory of liability in the law of torts. See James A. Henderson, Jr., Why Negligence Law Dominates Tort, 50 UCLA L. Rev. 377 (2002). See also Gary T. Schwartz, The Vitality of Negligence and the Ethics of Strict Liability, 15 Ga. L. Rev. 963 (1981); Gary T. Schwartz, The Beginning and the Possible End of Modern American Tort Law, 26 Ga. L. Rev. 601 (1992).

¹⁵Hairston v. Alexander Tank and Equipment Co., 311 S.E.2d 559 (N.C. 1984).

¹⁶For a discussion of the effects of the judgment-proof problem on the deterrence and insurance goals of torts, see, e.g., Steven Shavell, The Judgment Proof Problem, 6 Int'l Rev. L. & Econ. 45 (1986); Henry Hansmann & Reinier Kraakman, Toward Unlimited Shareholder Liability for Corporate Torts, 100 Yale L.J. 1879, 1882-83 (1991); William R. Keeton & Evan Kwerel, Externalities in Automobile Insurance and the Underinsured Driver Problem, 27 J.L. & Econ. 149, 149-50 (1984); John Summers, Comment, The Case of the Disappearing Defendant: An Economic Analysis, 132 U. Pa. L. Rev. 145 (1983).

¹⁷Mark F. Grady, Proximate Cause Decoded, 50 UCLA L. REV. 293, 306-12 (2002).

¹⁸See, e.g., A&M Records, Inc. v. Napster, Inc., 239 F.3d 1004 (9th Cir. 2001); Metro-Goldwyn-Mayer Studios, Inc. v. Grokster, 259 F.Supp. 2d 1029 (C.D. Cal. 2003).

¹⁹See, e.g., Recording Indus. Ass'n of Am., Inc. v. Verizon Internet Servs., Inc., 351 F.3d 1229 (D.C. Cir. 2003).

²⁰See, e.g., Computer Fraud and Abuse Act, 18 U.S.C. § 1030 (2000) (allowing civil cause of action against violators of the Act); VT. Stat. Ann. tit. 13, § 4106 (Supp. 2005) (permitting civil action for victims of cyber crimes).

²¹See, e.g., Ethan Preston & John Lofton, Computer Security Publications: Information Economics, Shifting Liability and the First Amendment, 24 Whittier L. Rev. 71, 132 (2002) ("Several commentators have proposed holding network administrators liable when their negligent failure to secure the computers under their control either allows computer criminals access to third party data stored there . . . or permits computer criminals to stage an attack on third party networks."). See also id. at 132 n.362-63.

²²See, e.g., Doug Lichtman & Eric Posner, Holding Internet Service Providers Accountable (John M. Olin Law & Economics Working Paper, No. 217, 2004); Douglas Lichtman, Holding Internet Service Providers Accountable, Regulation, Winter 2004-2005, at 54. But cf. Jim Harper, Against ISP Liability, Regulation, Spring 2005, at 30.

²³See, e.g., Michael Lee et al., Electronic Commerce, Hackers, and the Search for Legitimacy: A Regulatory Proposal, 14 Berkeley Tech. L. J. 839, 875-76 (1999) ("Numerous reasons may be cited for the proposition that ISPs should be jointly liable for the tortious hacking activities of those who use their services," including the fact that hackers are generally judgment-proof, difficult to identify and detect, and tend to be unattractive defendants due to jurisdictional problems.).

²⁴Inquiry Regarding the Entry of Verizon-Maine into the InterLATA Telephone Market Pursuant to Section 271 of Telecommunication Act of 1996, No. 2000-849 (Maine Public Utilities Commission Apr. 15, 2003).

²⁵Complaint at 10, Hamilton v. Microsoft Corp., No. BC303321 (Cal. Super. Ct. Sept. 30, 2003), available at http://www.sans.org/resources/mscomplaint.pdf ("[W]hile Microsoft has issued strings of alerts, they cannot be understood by the General Public and the method of delivery of the warning has actually increased the probability of harm from hackers who are educated by the information about the flaws and potential breach in the operating systems as described by Microsoft.").

²⁶See Dorothy E. Denning & Peter J. Denning, Internet Besieged 75-78 (ACM Press 1998).

²⁷Fred B. Cohen, Computer Viruses (Dec. 1986) (unpublished Ph.D. dissertation, University of Southern California) (on file with Learning Resource Center, Central College of Iowa).

²⁸Jeffrey O. Kephart et al., Fighting Computer Viruses, Scientific American, Nov. 1997, at 88. Dr. Gregory Benford published the idea of a computer virus as "unwanted code" and apparently wrote actual "viral" code, capable of replication. Denning, supra note 26, at 74.

²⁹ John McAfee & Colin Haynes, Computer Viruses, Worms, Data Diddlers, Killer Programs, and Other Threats to Your System 26 (1989). Dr. Frederick B. Cohen, A Short Course on Computer Viruses 1-2 (2d ed. 1994).

In his PhD dissertation, Dr. Cohen defined a virus simply as any program capable of self-reproduction. This definition appears overly general. A literal interpretation of the definition would classify even programs such as compilers and editors as viral. Denning, supra note 26, at 75.

³⁰The concise Oxford dictionary defines a worm as "[a] self-replicating program able to propagate itself across a network, typically having a detrimental effect." The concise Oxford dictionary (Judy Pearsall ed., 10^th ed. 1999).

³¹See generally John F. Shoch & Jon A. Hupp, The "Worm" Programs—Early Experience with a Distributed Computation, Comm. of the ACM, Mar. 1982, at 172.

³²Palo Alto Research Center, PARC History, http://www.parc.xerox.com/about/history/default.html (last visited Oct. 31, 2005).

³³Ari Takanen et al., Running Malicious Code by Exploiting Buffer Overflows: A Survey of Publicly Available Exploits, 2000 EICAR Best Paper Proc. 158, 162 ("The day when the world finally acknowledged the risk entailed in overflow vulnerabilities and started coordinating a response to them was the day when the Internet Worm was introduced, spread and brought the Internet to its knees.").

³⁴Lance J. Hoffman (ed.), Rogue Programs: Viruses, Worms, Trojan Horses (Van Nostrand Reinhold, 1990), at 247 ("The ability to propagate is essential to a virus program"); Denning, supra note 26, at 73-75.

³⁵Potential target hosts include application and system programs and the master boot record of the hard disks or floppy disks in the computer.

³⁶See, e.g., Eric J. Sinrod & William P. Reilly, Cyber-Crimes: A Practical Approach to the Application of Federal Computer Crime Laws, 16 Santa Clara Computer & High Tech. L.J. 177, 217 n.176 (2000).

³⁷ Jan Hruska, Computer Viruses and Anti-Virus Warfare 17-18 (1990). (In addition to self-replicating code, viruses often also contain a payload. The payload is capable of producing malicious side-effects.) See also Cohen, supra note 29, at 1-2 (examples of malignant viruses and what they do); McAfee & Haynes, supra note 29, at 61.

³⁸MacAfee & Haynes, supra note 29, at 61.

³⁹Sinrod & Reilly, supra note 36, at 218 (describing the W95.LoveSong.998 virus, designed to trigger a love song on a particular date).

⁴⁰ Viruses can cause economic losses, e.g., by filling up available memory space, slowing down the execution of important programs, locking keyboards, adding messages to printer output, and effectively disabling a computer system by altering its boot sector. The Melissa virus, for instance, mailed copies of itself to everyone in the victim's e-mail address book, resulting in clogged e-mail servers and even system crashes. See, e.g., Philip Fites, Peter Johnston & Martin Kratz, The Computer Virus Crisis 23-24 (2d ed.1992) (By filling all available storage capacity with copies of itself, the Christmas card virus stopped a major international e-mail system until the copies could be isolated and removed); Sinrod & Reilly, supra note 36, at 218 (describing the Melissa virus).

See Section III, infra, for an analysis of damages from computer virus infection. For examples of benign viruses and how they operate, see, e.g., Cohen, supra note 29, at 15-21.

⁴¹ There are three mechanisms through which a virus can infect a program. A virus may attach itself to its host as a shell, an add-on, or as intrusive code. A shell virus forms a shell around the host code, so that the latter effectively becomes an internal subroutine of the virus. The host program is replaced by a functionally equivalent program that includes the virus. The virus executes first, and then allows the host code to begin executing. Boot program viruses are typically shell viruses. Most viruses are of the add-on variety. They become part of the host by appending their code to the host code, without altering the host code. The viral code alters the order of execution, by executing itself first and then the host code. Macro viruses are typically add-on viruses. Intrusive viruses, in contrast, overwrite some or all of the host code, replacing it with its own code. See, e.g., Denning, supra note 26, at 81; Fites, supra note 40, at 73-75.

⁴²Szor, supra note 1, at 365-421 (extensive discussion of viruses that use exploits to spread themselves). See also discussion infra Section II on blended attacks that depend on vulnerabilities to spread.

⁴³Larry Bridwell, ICSA Labs, 9th Annual Computer Virus Prevalence Survey 14 (2004), https://www.icsalabs.com/icsa/docs/html/library/whitepapers/VPS2003.pdf.

⁴⁴David Harley et al., Viruses Revealed: Understand and Counter Malicious Software 406-10 (Osborne/McGraw-Hill 2001).

⁴⁵See, e.g., Denning, supra note 26, at 90-93; Ken Dunham, Bigelow's Virus Troubleshooting Pocket Reference 78-83, 102-08 (McGraw-Hill 2000).

⁴⁶See Hruska, supra note 37, at 42.

⁴⁷Jeffrey O. Kephart Et Al., Automatic Extraction of Computer Virus Signatures, Proceedings of the 4th Virus Bulletin International Conference (R. Ford, ed. 1994), Virus Bulletin Ltd., Abingdon, England, at 179-94, at 2.

⁴⁸A false positive is an erroneous report of the activity or presence of a virus where there is none. A false negative is the failure to report the presence of a virus when a virus is in fact present.

⁴⁹Hruska, supra note 37, at 42. For short descriptions and hexadecimal patterns of selected known viruses, see Hruska, supra note 37, at 43-52; Jeffrey O. Kephart Et Al., Blueprint for a Computer Immune System, IBM Thomas J. Watson Research Center Report, at 11 ("[A] signature extractor must select a virus signature carefully to avoid both false negatives and false positives. That is, the signature must be found in every instance of the virus, and must almost never occur in uninfected programs.") False positives have reportedly triggered a lawsuit by a software vendor, who felt falsely accused, against an anti-virus software vendor. Jeffrey O. Kephart Et Al., Blueprint for a Computer Immune System, IBM Thomas J. Watson Research Center Report, at 11. The IBM High Integrity Computing Laboratory has developed an optimal statistical signature extraction technique that examines all sections of code in a virus, and selects the byte strings that minimize the incidence of false positives and negatives. Jeffrey O. Kephart et al., Automatic Extraction of Computer Virus Signatures, Proceedings of the 4th Virus Bulletin International Conference e, R. Ford, ed., Virus Bulletin Ltd., Abingdon, England, 1994, pp. 179-94.

⁵⁰Dunham, supra note 45, at 78-83; Jeffrey O. Kephart et al., Fighting Computer Viruses, Scientific American, Nov. 1997. See also Sandeep Kumar & Eugene H. Spafford, A Generic Virus Scanner in C++, Technical report CSD-TR-92-062, Dept. of Computer Science, Indiana University, at 6-8.

⁵¹See, e.g., Pete Lindstrom, The Hidden Costs of Virus Protection, Spire Research Report, June 2003, at 5 ("In this day of 80,000+ known viruses and frequent discovery of new ones, the size of the signature file can be large, particularly if the updates are sent out as cumulative ones. Large updates can clog the network pipelines . . . and reduce the frequency that an administrator will push them out to the end users.").

⁵²Kumar, supra note 50, at 3-4.

⁵³Hruska, supra note 37, at 75.

⁵⁴Fites, supra note 40, at 69-76, Fig. 5.2-5.5; Dunham, supra note 45, at 79. See also Kumar, supra note 50, at 5-6.

⁵⁵Fites, supra note 40, at 125.

⁵⁶Carey Nachenberg, Future Imperfect, Virus Bulletin, August 1997, at 6. See also Francisco Fernandez, Heuristic Engines, Proceedings of the 11th International Virus Bulletin Conference, Sept. 2001, at 407-44; Diomidis Spinellis, Reliable Identification of Bounded-Length Viruses is NP-Complete , IEEE Transactions On Information Theory, Jan, 280, 282 (Jan. 2003) available at http://www.dmst.aueb.gr/dds/pubs/jrnl/2002-ieeetit-npvirus/html/npvirus.pdf (stating that theoretically perfect detection is in the general case undecidable, and for known viruses, NP-complete); David M. Chess & Steve R. White, Undetectable Computer Virus, Virus Bulletin Conference, Sept. 2000, available at http://www.research.ibm.com/antivirus/SciPapers/VB2000DC.htm (last visited November 29, 2005).

⁵⁷Fernandez, supra note 56, at 409 ("Many genuine programs use sequences of instructions that resemble those used by viruses. Programs that use low-level disk access methods, TSRs, encryption utilities, and even anti-virus packages can all, at times, carry out tasks that are performed by viruses.").

⁵⁸Nachenberg, supra note 56, at 7.

⁵⁹Certain byte sequences are, for instance, associated with decryption loops to unscramble a polymorphic virus when an infected routine is executed. If it finds a match, e.g., the scanner detects the presence of a decryption loop typical of a polymorphic virus, it catalogues this behavior.

⁶⁰See Kumar, supra note 50, at 4-5.

⁶¹The CPU, or central processing unit, of a computer is responsible for data processing and computation. See Hruska, supra note 37, at 113; D. Bender, Computer Law: Evidence and Procedure (1982), §2.02, at 2-7, 9.

⁶²Kumar, supra note 50, at 4.

⁶³Polymorphic viruses have the ability to "mutate" by varying the code sequences written to target files. To detect such viruses requires a more complex algorithm than simple pattern matching. Denning, supra note 26, at 89.

⁶⁴Nachenberg, supra note 56, at 9.

⁶⁵The wave of blended attacks introduced by CodeRed and Nimda is commonly described as the fourth wave of modern worms. The first wave consisted of the experimental viruses and worms of the time period, 1979 through 1990; the second wave introduced polymorphism and virus toolkits; the third wave, roughly spanning 1999 through late 2000, brought the mass e-mail viruses, such as Melissa. Chen, supra note 11, at 274.

⁶⁶3 Symantec Internet Security Threat Report 34, 35 (2003) (by exploiting IT vulnerabilities, blended threats are frequently able to bypass conventional security practices such as requiring strong, non-default passwords, as long as systems have the type of vulnerability exploited by the attack) available at http://enterprisesecurity.symantec.com/Content.cfm?articleID=1964&EID=0 (last visited Nov. 28, 2005).

⁶⁷Eric Chien & Peter Szor, Blended Attack Exploits, Vulnerabilities and Buffer-Overflow Techniques in Computer Viruses. Symantec White Paper, available at http://enterprisesecurity.symantec.com/content/displaycolateral.cfm?colID=425 (last accessed Nov. 28, 2005), reprinted from Virus Bulletin, Sept. 2002., at 3., The buffer overflow is discussed in the next subsection.

⁶⁸See discussion infra Part II.C.

⁶⁹George V. Hulme, One Step Ahead, InformationWeek, May 20, 2002, at 57.

⁷⁰Some earlier viruses were, of course, destructive. The 1987 South African Friday-the-13th virus, for instance, was programmed to delete its host program, if invoked on Friday the 13th.

⁷¹Blended Threats: Case Study and Countermeasures. Symantec White Paper, at 2, available at http://enterprisesecurity.symantec.com/content/displaycolateral.cfm?colID=152&EID=0. (last accessed Nov. 28, 2005). See also Hulme, supra note 69 (noting the destruction capabilities of a blended threat include destruction of files, creation of backdoors, and installation of Trojan horses, and so-called zombie programs that can later be used to launch denial of service attacks); eEye Digital Security, CodeRed Worm Analysis, Aug. 2001, available at http://www.eeye.com/Research/Advisories/AL20010804.html (last accessed Nov. 28, 2005).

⁷²David Moore et al., Inside the Slammer Worm, 1 IEEE SECURITY AND PRIVACY 33, 33 (2003).

⁷³Nimda's other attack vectors were infection of Microsoft IIS web servers via a buffer overflow exploit, infection of network shares, and infection via Javascript added to web pages. See, e.g., Thomas Chen, Presentation at Southern Methodist University Department of Electrical Engineering (proceedings available at http://www.cisco.com/web/about/ac123/ac147/ac174/ac236/about_cisco_ipj_archive_article09186a00801c5a9c.html.). Other attack points frequently used in blended attacks include injecting malicious code into .exe files on a target system, creating world readable network shares, making multiple registry changes, and adding script code to HTML files.

⁷⁴3 Symantec Internet Security Threat Report at 36 (2003) (describing the operation of the typical mass-mailing worm: First, they exploit a known IT vulnerability to infect the system. Then, they collect e-mail addresses from the infected system. Finally, they spread via their own e-mail system, which is independent of the client e-mail system. This methodology enables the worm or virus to spread and propagate without user intervention. Users whose systems have been hijacked in this manner are often unaware that they are being used as launching pads for infected e-mails. In addition, these viruses frequently spoof the "From" address on e-mails, obscuring the origin of the infected e-mail.), available at http://www.symantec.com/region/hu/huresc/download/2003_02_03SymantecInternetSecurityThreatReport.pdf (last accessed Nov. 28, 2005).

⁷⁵M. Grey, Instant Messaging in the Enterprise Will Remain a Puzzle, Gartner Research Report COM-18-7979 (22 Nov. 2002), available at http://www.gartner.com (on file with author); K. Dulaney & B. Clark, E-Mail/PIM Is Still No. 1. Gartner Research Report SPA-18-5839 (20 November 2002) ("The 'always on' nature of the connectivity, remote access to critical sensitive data, and the increasingly computational nature of mobile devices, set the stage for a potential virus or worm of significance."), available at http://www.gartner.com (on file with author). For a discussion of vulnerabilities in instant messaging technology, see, e.g., Securing Instant Messaging, Symantec White Paper, http://securityresponse.symantec.com/avcenter/reference/secure.instant.messaging.pdf (last accessed Nov. 28, 2005).

⁷⁶William S. Davis & T.M. Rajkumar, Operating Systems: A Systematic View 27, 28 (6 ^th ed. 2004).

⁷⁷Mark E. Donaldson, Inside the Buffer Overflow Attack: Mechanism, Method & Prevention, Sans Institute White Paper (Sans Institute) April 3, 2002, at 4,

http://www.sans.org/rr/whitepapers/securecode/.

⁷⁸The effect of a buffer overflow would be to abort the application program, resulting in a segmentation fault and terminating with a core dump.

⁷⁹Microsoft Corporation defines a buffer overflow attack as "an attack in which a malicious user exploits an unchecked buffer in a program and overwrites the program code with their own data. If the program code is overwritten with new executable code, the effect is to change the program's operation as dictated by the attacker. If overwritten with other data, the likely effect is to cause the program to crash."Donaldson, supra note 77 at 3.

⁸⁰Rob Enderle and Jasmine Noel, The New Approach to Windows Security, Enderle Group White Paper (Enderle Group) June 25, 2004, at 7, http://www.amd.com/us-en/assets/content_type/DownloadableAssets/New_Approach_to_PC_SecurityFinal_ (2).pdf.

⁸¹Takenen, supra note 33, at 162. ("The day when the world finally acknowledged the risk entailed in overflow vulnerabilities and started coordinating a response to them was the day when the Internet Worm was introduced, spread and brought the Internet to its knees.")

⁸²See e.g. James A. Henderson, Why Negligence Law Dominates Tort, 50 UCLA L. Rev. 377 (2003). See also Gary T. Schwartz, The Vitality of Negligence and Ethics of Strict Liability, 15 Ga. L. Rev. 963 (1981); Gary T. Schwartz, The Beginning and Possible End of Modern American Tort Law, 26 Ga. L. Rev. 601 (1992).

⁸³Prosser & Keeton on the Law of Torts § 31(5th ed. 1984); Restatement (Second) of Torts, § 282 (1965). (Describing negligence as conduct "which falls below the standard established by law for the protection of others against unreasonable risk of harm.")

⁸⁴Dan B. Dobbs, The Law of Torts Vol. 1, 258 (West Group., 2001).

⁸⁵See e.g. N. Tel v. Brandl, 179 B.R. 620 (Bankr. Minn. 1995)(Affirming default judgment for damage resulting from transmission of a virus to computer system known to be vital to plaintiff's business operations.)

Authors of malevolent software are rarely prosecuted, although high-profile offenders are occasionally brought to justice. Robert T. Morris, the author of the first worm that had a significant impact on the Internet was convicted under 18 U.S.C. § 1030. See e.g. Katie Hafner and John Markoff, Cyberpunk: Outlaws and hackers on the computer frontier, 251, 339-340 (1991). (Includes in-depth analysis of Robert Morris' Internet Worm.) David Smith, creator of the Melissa virus was also convicted under the computer fraud statute. See e.g. Press Release, U.S. Department of Justice, Creator of "Melissa" Virus Pleads Guilty to State and Federal Charges, Dec. 9, 1999, available at http://www.cybercrime.gov/melissa.htm.

The computer fraud act, 18 U.S.C. § 1030 (a)(5)(A)(i) (1996 & Supp. 2004), imposes liability for knowingly transmitting "a program, information, code, or command, and as a result of such conduct, intentionally causes damage without authorization, to a protected computer."

See also Revlon v. Logisticon Inc., No. 705933 (Cal. Super. Ct. Santa Clara County, complaint filed Oct. 22, 1990)(Defendant allegedly placed disabling device on Plaintiff's computer system. Parties settled under terms of nondisclosure.) (Complaint on file with author or Journal); Clayton X-Ray Co. v. Prof'l Sys. Corp., 812 S.W.2d 565, 567 (Mo. Ct. App. 1991)(Software provider held liable for conversion of purchaser's computer system by placing disabling code in the system.)

⁸⁶Mark F. Grady & Ward Farnsworth, Torts: Cases and Questions, 215-306 (2004).

⁸⁷Lunney v. Prodigy Services Co., 723 N.E. 2d 539, 543 (N.Y. 1999).

⁸⁸See also James v. Meow Media, Inc., 90 F. Supp. 2d 798, 819 (W.D. Ky. 2000) (finding no duty of care from video game manufacturers to victims of violence and stating that imposing such a duty would be detrimental to artistic freedom of speech of media defendants).

⁸⁹868 F.2d 293, 297 (8th Cir. 1989).

⁹⁰Id. at 297. See also Hosp. Computer Sys. v. Staten Island Hosp., 788 F. Supp. 1351, 1361 (D. N.J. 1992) (stating that computer specialists can be held liable for ordinary but not professional negligence).

⁹¹Meiring de Villiers, Computer Viruses and Civil Liability: A Conceptual Framework, 40 Tort Trial & Ins. Prac. L.J. 123 (Fall 2004); Meiring de Villiers, Virus ex Machina Res Ipsa Loquitur, 2003 Stan. Tech. L. Rev. 1.

⁹²Grady, supra note 17, at 298.

⁹⁴Dobbs, supra note 84, at 444.

⁹⁵Grady, supra note 17, at 299.

⁹⁷Id. at 296-98.

¹⁰⁰Grady, supra note 17, at 299.

¹⁰¹Szor, supra note 1, at 98.

¹⁰²Grady, supra note 17, at 301-21.

¹⁰⁷Grady, supra note 17, at 315-16.

¹⁰⁹537 A.2d 468 (Conn. 1988).

¹¹¹Id. at 473. See also Ellis v. D'Angelo, 253 P.2d 675 (Cal. Dist. Ct. App. 1953) (suggesting that a child could be liable for tortious acts if the child could predict the consequences of his act); McGuire v. Almy, 8 N.E.2d 760, 763 (Mass. 1937) ("[W]here an insane person by his act does intentional damage to the person or property of another he is liable for that damage in the same circumstances in which a normal person would be liable."); Mark F. Grady, The Free Radicals of Tort, 11 Sup. Ct. Econ. L. Rev. 189 (2003).

¹¹²173 N.W.2d 619 (Wis. 1970).

¹¹⁵See Grady, supra note 111, at 194.

¹¹⁶Grady, supra note 17, at 306-12.

¹¹⁷The effects of the judgment-proof problem on the deterrence and insurance goals of torts are discussed in the following references: Steven Shavell, The Judgment Proof Problem, 6 Intl. Rev. Law & Econ. 45 (1986); Henry Hansmann & Reinier Kraakman, Toward Unlimited Shareholder Liability for Corporate Torts, 100 Yale L.J. 1879, 1882-83 (2001); John G. Fleming, Report to the Joint Committee of the California Legislature on Tort Liability of the Problems Associated with American Motorcycle Association v. Superior Court, 30 Hastings L.J. 1465, 1470 (1979); William R. Keeton & Even Kwerel, Externalities in Automobile Insurance and the Underinsured Driver Problem, 27 J. Law & Econ. 149, 149-50 (1984); John Summers, Comment, The Case of the Disappearing Defendant: An Economic Analysis, 132 U. Pa. L. Rev. 145 (1983).

¹¹⁸Grady, supra note 17, at 308.

¹¹⁹177 S.E.2d 846 (Ga. Ct. App. 1970).

¹²⁰See Grady, supra note 111, at 190.

¹²¹539 P.2d 36 (Cal. 1975).

¹²²96 Eng Rep 525 (KB 1773).

¹²³See Grady, supra note 111, at 205 for a discussion of the various opinions in Scott v. Shepherd.

¹²⁴105 Eng. Rep. 1023 (K.B. 1816).

¹²⁵Grady, supra note 111, at 202 ("[I]n 1816, the English Court of King's Bench had already decided the first indisputable EFR case, which was Dixon v. Bell.").

¹²⁶105 Eng. Rep. at 1023.

¹²⁷113 Eng. Rep. 1041 (Q.B. 1841).

¹²⁸Id. See also Lane v. Atlantic Works, 111 Mass. 136 (1872) (defendant found liable for encouraging free radicals when carelessly laid iron bars were jostled by child and subsequently fell, injuring plaintiff).

¹²⁹19 Johns. 381 (N.Y. 1822).

¹³⁰Id. See also Grady, supra note 111, at 197-98, 201-02 (providing discussion of Guille).

¹³¹See Weirum v. RKO Gen., Inc., 539 P.2d 36 (Cal. 1975).

¹³²89 N.E. 425 (Ill. 1909).

¹³⁵Id. at 426-427.

¹³⁶Id. at 426. See also, Grady, supra note 111, at 204-05.

¹³⁷Grady, supra note 111, at 205. An analogous situation is presented in Travell v. Bannerman, 75 N.Y.S. 866 (App. Div. 1902), where the harm was caused by children playing with explosive material.

¹³⁸Paul A. Strassman & William Marlow, Risk-Free Access Into the Global Information Infrastructure via Anonymous Re-Mailers, Symposium on the Global Information Infrastructure: Information, Policy & Information Infrastructure, Jan. 28-30, 1996, available at http://www.strassmann.com/pubs/anon-remail.html ("Information terrorism . . . is a unique phenomenon in the history of warfare and crime. For the last two hundred years the theory of warfare has been guided by 'force-exchange' equations in which the outcome was determined by the rate of attrition of each opposing force. In information attacks these equations do not apply because the attacker remains hidden and cannot be retaliated against. Since Biblical times, crimes have been deterred by the prospects of punishment. For that, the criminal had to be apprehended. Yet information crimes have the unique characteristic that apprehension is impossible, since even identification of the criminal is not feasible. Information crimes can be committed easily without leaving any telltale evidence such as fingerprints, traces of poison or bullets.").

¹³⁹Mark D. Rasch, Criminal Law and the Internet, in The Internet and Business: A Lawyer's Guide to the Emerging Legal Issues (Joseph F. Ruh, Jr. ed., The Computer Law Association 1996), available at http://www.cla.org/RuhBook/chp11.htm (anonymity of cyberspace encourages network users to commit offenses they would not otherwise attempt, exacerbated by fact that bounds of acceptable behavior not yet clearly defined in cyberspace).

¹⁴⁰Sarah Gordon, Virus Writers: The End of Innocence, IBM Thomas J. Watson Research Center, http://www.research.ibm.com/antivirus/SciPapers/VB2000SG.htm (last visited Nov. 14, 2005) ("[T]racing a virus author is extremely difficult if the virus writer takes adequate precautions against a possible investigation."); Ian C. Ballon, Alternative Corporate Responses to Internet Data Theft, 471 PLI/Pat. 737, 739 (1997); Mary M. Calkins, Note, They Shoot Trojan Horses, Don't They? An Economic Analysis of Anti-Hacking Regulatory Models, 89 Geo. L.J. 171 (2000); Jelena Mirkovic et. al, Internet Denial-of-service: Attack and Defense Mechanisms 14 (Prentice Hall Professional Technical Reference 2005) ("[V]ery few attackers have been caught and prosecuted. . . . [One] factor is the ease of performing a DoS attack without leaving many traces for investigators to follow. . . . Another type of DoS perpetrator is a sophisticated hacker who uses several means to obscure her identity and create subtle variations in traffic patterns to bypass defenses.").

¹⁴¹See, e.g., Ted Bridis, Microsoft Offers Huge Cash Rewards for Catching Virus Writers, Deseret News, Nov. 6, 2003, at E10, available at 2003 WLNR 13526397 ("Police around the world have been frustrated in their efforts to trace some of the most damaging attacks across the Internet. Hackers easily can erase their digital footprints, crisscross electronic borders and falsify trails to point at innocent computers.").

¹⁴²Rasch, supra note 139. See also Virus Writers Difficult to Find in Cyberspace, BizReport, Sept. 12, 2003, http://www.bizreport.com/news/4917 ("There are many ways for virus writers to disguise themselves, including spreading the programs through unwittingly infected e-mail accounts. The anonymity of the Internet allows you to use any vulnerable machine to launder your identity.").

¹⁴³See supra note 9 for a definition of a denial-of-service attack ("DoS attack").

¹⁴⁴See, e.g., Rik Farrow, Distributed Denial-of-service Attacks (DDoS), IT Architect, Mar. 2000, at 74, available at http://www.itarchitect.com/shared/article/showArticle.jhtml;jsessionid=LXPZARCXE5J42QSNDBECKHSCJUMEKJVN?articleId=8702755&classroom= ("In an ordinary network-based denial-of-service attack, an attacker uses a tool to send packets to the target system, often forcing a reboot. Often the source address of these packets is spoofed, making it difficult to locate the real source of the attack.").

¹⁴⁵L. Jean Camp & Catherine Wolfram, Pricing Security, Econ. of Info. Security 17 (L. Jean Camp & Stephen Lewis, eds., 2004).

¹⁴⁶The virus named "Sobig F," for instance, is programmed to turn a computer into a host that sends out spam e-mail messages, often without the knowledge of the owner. It is widely believed that half a million copies of the virus named AVF were sent by a spammer. Unlike Melissa, the AVF virus does not mail copies of itself out to everyone in the infected computer's address book. Instead, AVF makes the infected computer an intermediary, by opening a backdoor in the infected machine through which spammers can distribute their junk mail.

¹⁴⁷Spam Virus Hijacks Computers, BBC News, June 13, 2003, http://news.bbc.co.uk/2/hi/technology/2987558.stm; Jo Twist, Why People Write Computer Viruses, BBC News, Aug. 23, 2003, http://news.bbc.co.uk/1/hi/technology/3172967.stm; see also Chen & Robert, supra note 11, at 14 ("[A] sociological reason [for continued pervasiveness of worms and viruses] is the lack of accountability for worm writers. The long-held general perception has been that worms and viruses are low risk crimes. It is notoriously difficult to trace a virus or worm to its creator from analysis of the code, unless there are inadvertent clues left in the code.").

¹⁴⁸ICSA Labs 9th Annual Computer Virus Prevalence Survey, supra note 43, at 15 (showing e-mail attachments, as a source of infection, increasing from nine percent in 1996 to eighty-eight percent in 2003, while infection by diskette decreased from seventy percent in 1996 to virtually zero in 2003).

¹⁴⁹See, e.g., Why Anti-Virus Software is Not Enough: The Urgent Need for Server-Based Email Content Checking, Feb. 15, 2003, http://www.secinf.net/anti_virus/. See also Lynne Munro, Protecting Your Computer from Email Viruses and Worms (Oxford University Computing Services rev. 2005).

¹⁵⁰Spammers and Viruses Unite, BBC News, Apr. 30, 2003, http://news.bbc.co.uk/1/hi/technology/2988209.stm. For example, a hijacking program named Proxy-Guru typically arrives as a spam message with an attachment. Opening the attachment triggers the program to forward information about the hijacked account to a Hotmail account. This information then enables a would-be spammer to route mail through the hijacked computer. The source of this spam would be very difficult, if not impossible, to trace, especially if the spammer and the sender of the hijacking program employed anonymity-preserving techniques such as a remailer. Id. See also Jay Lyman, Authorities Investigate Romanian Virus Writer, Linux Insider, Sept. 4, 2003, http://www.linuxinsider.com/story/31500.html (referring to the "difficulty of tracking down virus writers, particularly when they are skilled enough to cover their digital tracks, [so that] few offenders are ever caught."); Noah Levine, Note, Establishing Legal Accountability for Anonymous Communication in Cyberspace, 96 Colum. L. Rev. 1526, 1530-33 (1996).

¹⁵¹A remailed message may still be traceable if, for instance, the anonymous remailer administrators keep a log of the identities of their clients. Not all remailers keep such information, though. See, e.g., Levine, supra note 150, at nn.24-28.

¹⁵²See, e.g., id. at n.50.

¹⁵³See, e.g., Anonymous Remailer, Wikipedia, http://en.wikipedia.org/wiki/Anonymous_remailer (last visited Oct. 29, 2005) ("An anonymous remailer is a server computer which receives messages with embedded instructions on where to send them next, and which forwards them without revealing where they originally came from. There are Cypherpunk anonymous remailers, Mixmaster anonymous remailers, and nym servers, among others, which differ in how they work, in the policies they adopt, and in the type of attack on anonymity of e-mail they can (are intended to) resist.").

¹⁵⁴Microsoft has since created a patch to fix this vulnerability. See United States Computer Emergency Readiness Team, Vulnerability Note VU#842160 (Nov. 3, 2004), http://www.kb.cert.org/vuls/id/842160 (describing details of the vulnerability). The malicious code is commonly referred to as MyDoom{AG, AH, AI} or Bofra. When a user clicked on a malicious e-mail message, Internet Explorer opens and displays an HTML document that exploits the vulnerability to execute the virus. See University of Cambridge, Technical User Support Computing Service, http://www-tus.csx.cam.ac.uk/virus/alerts.html (last visited Oct. 29, 2005).

¹⁵⁵See, e.g., Robert Rossney, How to Keep Your ID a Secret on Usenet, S.F. Chronicle, Mar. 9, 1995, at D3 ("Freed from accountability, people can and do issue all kinds of vileness and stupidity."); Levine, supra note 150, at 1537 ("In each case, anonymity serves to remove, or at least significantly decrease, the deterrent effect of the law against civil or criminal violations."); Tal Z. Zarsky, Thinking Outside the Box: Considering Transparency, Anonymity, and Pseudonymity as Overall Solutions to the Problems of Privacy in the Internet Society, 58 U. Miami L. Rev. 991, 1028 (2004) ("Anonymity adversely affects society by causing the loss of accountability."); Paul A. Strassman & William Marlow, Risk-Free Access Into the Global Information Infrastructure Via Anonymous Re-Mailers, Harvard University, Kennedy School of Government, Symposium on the Global Information Infrastructure: Information, Policy & Information Infrastructure, Cambridge, MA (Jan. 28-30, 1996), available at http://www.strassmann.com/pubs/anon-remail.html (the anonymous remailing service, anon.penet.fi., was frequently used by the Russian (ex-KGB) criminal element; the "double-blind" method of communication it offered is favored for engaging services of cybercriminals and for authorizing payments for criminal acts through a third party).

¹⁵⁶Doug Lichtman & Eric Posner, Holding Internet Service Providers Accountable (John M. Olin Law & Econ., Working Paper No. 217, 2004), available at http://www.law.uchicago.edu/Lawecon/WkngPprs_201-25/217-dgl-eap-isp.pdf ("Sophisticated saboteurs use the Internet's topology to conceal their tracks by routing messages and information through a convoluted path that is difficult for authorities to uncover.");Gordon, supra note 140 (finding no evidence that such prosecutions have alleviated the virus problem, as measured by the rate of creation of new viruses in the wild subsequent to high profile prosecutions). See also Robert Lemos, 'Tis the Season for Computer Viruses, ZDNet UK News, Dec. 13, 1999, http://news.zdnet.co.uk/internet/0,39020369,2075787,00.htm (suggesting that even after the author of the Melissa virus had been apprehended and was expected to be sentenced to a multi-year prison term, the appearance of new viruses on the Internet continued to proliferate, and at an increasing rate).

¹⁵⁷Gordon, supra note 140, at 12 (reference to a survey by A. Briney).

¹⁵⁸Id. at 8 (reference to DefCon survey).

¹⁵⁹Id. at 5 ("Minnesota statute ßß 609.87 to .89 presents an amendment which clearly defines a destructive computer program, and which designates a maximum [prison term of ten] years; however, no cases have been reported. Should we conclude there are no virus problems in Minnesota?"). See also Michael K. Block & Joseph G. Sidak, The Cost of Antitrust Deterrence: Why not Hang a Price-Fixer Now and Then?, 68 Geo. L.J. 1131, 1131-32 (1980); Stevan D. Mitchell & Elizabeth A. Banker, Private Intrusion Response, 11 Harv. J.L. & Tech. 699, 704; Andy McCue, IT Crime Still Going Unreported, IT Week, May 23, 2002, http://www.pcw.co.uk/itweek/analysis/2086802/crime-going-unreported?vnu_lt=pcw_art_related_articles ("What we are seeing is an increase in actual and attempted crimes using technology and particularly the [I]nternet. The number of security breaches reported is only the tip of the iceberg. For every one admitted there might be 100 held within companies.").

¹⁶⁰Madeline Bennett, Crime Laws Lack Coherence, IT Week, May 20, 2002, http://www.computing.co.uk/itweek/news/2084464/crime-laws-lack-coherence?vnu_lt=ctg_art_related_articles (quoting Graham Cluley of anti-virus firm Sophos, "To ensure that virus authors receive sentences that reflect the gravity of their offences, . . . businesses should play their part. 'Viruses can cause great damage, yet businesses are ashamed to report infections. They must take a two-pronged stance: Improve protection on their systems and be prepared to take action against the authors of malicious code.'").

¹⁶¹McCue, supra note 137 (reporting that ninety percent of a security consulting firm's client companies take no action when an attack originates from overseas).

¹⁶²J. Grable, Treating Smallpox with Leeches: Criminal Culpability of Virus Writers and Better Ways to Beat Them at Their Own Game. Computers & the Law Project, University of Buffalo School of Law. See also Sarah Gordon, supra note 140 ("[G]iven the small number of virus writers who have been arrested and tried . . . this lack of arrests is one of the primary indicators used by some to argue that laws are not a good deterrent."). See also BizReport News, Virus Writers Difficult to Find in Cyberspace, Sept. 2003 (reporting that it took 18 days to track down the author of the Blaster worm, even though the author left a clear trail behind, including his alias stitched into the virus code, and references to a web site registered in his name), available at http://www.bizreport.com/print.php?art_id=4917.

¹⁶³Patti Waldmeir, Dark Side of Cybercrime Fight: An International Treaty on Law Enforcement for the Web Poses Unsettling Questions about Civil Liberties, Fin. Times (London), May 10, 2001, at 17; Love Bug Suspect Can Still Face Civil Suits: Philippine Prosecutor, Agence France Presse, Aug. 23, 2000.

¹⁶⁴871 F.Supp. 535 (D. Mass. 1994).

¹⁶⁵No Electronic Theft ("NET") Act, Pub. L. No. 105-147, § 2(b), 111 Stat. 2678 (1997), (codified as amended at 17 U.S.C. § 506(a) (2000)). See also Michael L. Rustad & Thomas H. Koening, Cybertorts and Legal Lag: An Empirical Analysis, 13 S. Cal. Interdisc. L.J. 77, 139 ("The failure of tort law remedies to keep pace with new cyberwrongs is a legal lag, which will persist until the common law catches up with good sense. The growing impact of information technologies 'is already creating a tremendous cultural lag, making nonsense of existing laws.' . . . Courts have yet to develop cybertort remedies for negligent computer security. . . .").

¹⁶⁶Robert Lemos, Microsoft Reveals Web Server Hole, CNET News.com, June 18, 2001, http://news.com.com/Microsoft+reveals+Web+server+hole/2100-1001_3-268608.html.

¹⁶⁷eEye Digital Security Advisory (Aug. 4, 2001), http://www.eeye.com/html/Research/Advisories/AL20010804.html.

¹⁶⁸Baca, supra note 3.

¹⁷⁰Szor, supra note 1, at 367, 410.

¹⁷¹Maine Public Utilities Commission, Docket No. 2000-849, Inquiry Regarding the Entry of Verizon-Maine Into the InterLATA Telephone Market Pursuant To Section 271 of Telecommunications Act of 1996 (2003).

¹⁷²Guess?, Inc., F.T.C. Docket C-4091 (2003) (complaint), available at http://www.ftc.gov/os/caselist/0223260.htm.

¹⁷³Press Release, Federal Trade Commission, Guess Settles FTC Security Charges; Third FTC Case Targets False Claims about Information Security (June 18, 2003), available at http://www.ftc.gov/opa/2003/06/guess.htm. See generally Stewart Baker & Maury Shenk, A Patch in Time Saves Nine: Liability Risks for Unpatched Software, Steptoe & Johnson LLP, Oct. 2003, http://www.steptoe.com/publications/PI3370.pdf.

¹⁷⁴Grady, supra note 111, at 207.

¹⁷⁵259 A.2d 794 (Md. 1970).

¹⁷⁶A.C. 1004 (H.L.) [1970] (appeal taken from Eng.).

¹⁷⁷Cole v. Hous. Auth., 385 N.E.2d 382 (Ill. App. Ct. 1979).

¹⁷⁸Dennis ex rel. Evans v. Timmons, 437 S.E.2d 138 (S.C. Ct. App. 1993).

¹⁷⁹Donehue v. Duvall, 243 N.E.2d 222 (Ill. 1968). The defendants had hauled loads of dirt into their backyard. Children from the neighborhood frequented the pile and threw clods of dirt at each other, and the defendants knew about this. One of the children threw a dirt clod at the five-year-old defendant and injured his eye. The trial court dismissed the complaint, and the Illinois Supreme Court affirmed.

¹⁸⁰See, e.g,. James C. Foster et al., Buffer Overflow Attacks: Detect, Exploit, Prevent 424 (Syngress Publishing 2005) (describing the CodeAssure Vulnerability Knowledgebase troubleshooting system, which is capable of reliably identifying flaws such as buffer and integer overflows in software). See also Joel McNamara, Secrets of Computer Espionage: Tactics and Countermeasures 235 (Wiley 2003) (discussing commercial, sometimes free, scanners, that can be used to probe a system for vulnerabilities).

¹⁸¹Foster, supra note 180, at 8.

¹⁸²See, e.g., Chen, supra note 11, at 13 ("Blaster suggests a trend that the time between discovery of a vulnerability and the appearance of a worm to exploit it is shrinking (to one month in the case of Blaster.)"). See also Universal City Studios, Inc. v. Corley, 273 F.3d 429, 451-52 (2d Cir. 2001) (quoting Universal City Studios, Inc. v. Reimerdes, 111 F. Supp. 2d 294, 331-32 (S.D.N.Y. 2000)).

¹⁸³Crispin Cowan et al., Buffer Overflows: Attacks and Defenses for the Vulnerability of the Decade, Working Paper, Dept. of Computer Science and Engineering, Oregon Graduate Institute of Science & Technology, at 1, available at http://www.cse.ogi.edu/DISC/projects/immunix ("Buffer overflow vulnerabilities particularly dominate in the class of remote penetration attacks because a buffer overflow vulnerability presents the attacker with exactly what they need: The ability to inject and execute attack code. The injected code runs with the privileges of the vulnerable program, and allows the attacker to bootstrap whatever other functionality is needed to control ("own") the host computer.").

¹⁸⁴3 Symantec, Internet Security Threat Report 47 (2003), http://www.securitystats.com/reports/Symantec-Internet_Security_Threat_Report_vIII.20030201.pdf.

¹⁸⁵For a review of publicly available exploits, see Takanen, supra note 33.

¹⁸⁶See, e.g., Nathan P. Smith, Southern Connecticut State University Computer Science Department Stack Smashing Vulnerabilities in the UNIX Operating System (May 7, 1997), http://destroy.net/machines/security/ (thorough academic survey, covering history and terminology, various vulnerabilities and related technologies, as well as solutions); David Litchfield, Exploiting Windows NT 4 Buffer Overruns, http://wwwinfowar.co.uk/mnemonix/ntbufferoverruns.htm (last visited Oct. 31, 2005).

¹⁸⁷See Sun Alert Notification (Feb. 25, 2004), http://sunsolve.sun.com/search/document.do?assetkey=1-26-50104-1.

¹⁸⁸Posting of Ulf Harnhammar to vulndiscuss@vulnwatch.org (Jan. 26, 2003), http://archives.neohapsis.com/archives/vulnwatch/2003-q1/0042.html.

¹⁸⁹http://packetstormsecurity.org/filedesc/hypermail.html.

¹⁹⁰The SpamAssassin spam daemon contains a buffer overflow vulnerability when running in Batched SMTP (BSMTP) mode. See SANS Critical Vulnerability Analysis Vol. 2 No. 4 (Feb. 3, 2003), http://www.sans.org/newsletters/cva/vol2_4.php?printer=Y. See also Szor, supra note 1, at 402 (describing the OpenSSL buffer overflow vulnerability as challenging to exploit); id. at 547 ("Some vulnerabilities are easily exploited by the attackers, while others take months to develop.").

¹⁹¹For statistics and a discussion of the pervasiveness of buffer overflow exploitation, see infra Section IV.C(2)(c).

¹⁹²See Press Release, Federal Trade Commission, supra note 173; Guess?, Inc., supra note 172.

¹⁹³Grady, supra note 17, at 310 ("[T]he defendant, in order to be liable, must negligently provide some special encouragement of wrongdoing that does not exist in the normal background of incitements and opportunities.").

¹⁹⁴Stansbie v. Troman, (1948) 2 K.B. 48 (appeal taken from B.C.).

¹⁹⁵Russo v. Grace Inst., 546 N.Y.S.2d 509 (N.Y. Sup. Ct. 1989), aff'd, 545 N.Y.S.2d 629 (N.Y. App. Div. 1989).

¹⁹⁶See Press Release Office of New York State Attorney General Eliot Spitzer, Attorney General reaches Agreement with Barnes and Noble on Privacy and Security Standards (Apr. 29, 2004), available at http://www.oag.state.ny.us/press/2004/apr/apr29a_04.html.

¹⁹⁷363 P.2d 1 (Cal. 1961).

²⁰¹See supra note 80 and accompanying text.

²⁰²Security vulnerabilities in web-based applications, such as online banking and e-commerce, provide the electronic equivalent of an open door to hackers to access sensitive information in computerized databases. See, e.g., United States of America, In the Matter of GUESS?, Inc. and GUESS.com, Inc. (June 2003), http://www.ftc.gov/os/2003/06/guesscmp.htm (wherein attackers allegedly gained access to sensitive consumer information by conducting a "relatively basic web-based application attack" on the Guess? web site).

²⁰³Vulnerability exploits allow malevolent code to operate with greater speed and efficiency. Nimda, for instance, relied on an input validation exploit known as MIME header parsing, as well as auto-execution capability to spread rapidly. See, e.g., Szor, supra note 1, at 415 ("The ability to infect IIS web servers via an input validation exploit and autoexecute upon users' reading or previewing e-mail played a large role in Nimda's ability to spread so rapidly.").

²⁰⁴Dobbs, supra note 84, at 444; see also Sinram v. Pa. R.R. Co., 61 F.2d 767, 771 (2d Cir. 1932) (Hand, J.) ("[T]he usual test is . . . whether the damage could be foreseen by the actor when he acted; not indeed the precise train of events, but similar damage to the same class of persons.").

²⁰⁵See Weirum v. RKO Gen., Inc., 539 P.2d 36 (Cal. 1975).

²⁰⁶927 P.2d 675, 676 (Utah Ct. App. 1996).

²¹⁰Rupert Goodwins, Playing Silly Buffers: How Bad Programming Lets Viruses In, ZDNet White Paper (Jan. 2004), http://insight.zdnet.co.uk/software/developer/0,39020469,39119117-2,00.htm ("The buffer overflow is the mechanism of choice for the discerning malware merchant. New hardware and software techniques are reducing the incidence of this perennial problem, but it's unlikely ever to go away completely."); see also Cowan, supra note 183 (describing buffer overflows as not the "most common form of security vulnerability" over the last decade, but also the dominant penetration mechanism for anonymous Internet users to gain remote control of a computer or network. Because buffer overflow attacks enable anyone to take total control of a host, they represent one of the most serious classes of security threats.").

²¹¹Cowan, supra note 183 ("If the buffer overflow vulnerability could be effectively eliminated, a very large portion of the most serious security threats would also be eliminated.").

²¹²Foster, supra note 180, at 20; see also Szor, supra note 1, at 538 ("Every month critical vulnerabilities are reported in a wide variety of operating systems and applications. Similarly, the number of computer worms that exploit system vulnerabilities is growing at an alarming rate.").

\²¹³See Donaldson, supra note 77, at 1 ("Despite its lengthy history and simple preventative methods, the buffer overflow continues to be a significant and prominent computer security concern even today. For example, buffer overflow problems are implicated in five of the SANS Top 20 vulnerabilities. The SuSE Linux web site lists twenty-two buffer overflow vulnerabilities that require patching. Of the 44 CERT advisories published between 1997 and 1999, twenty-four were related to buffer overflow issues.").

²¹⁴CERT Advisories, available at http://www.cert.org/advisories/. See also Mark Shaneck, An Overview of Buffer Overflow Vulnerabilities and Internet Worms, CSCI, Dec. 2003, at 1, available at http://www-users.cs.umn.edu/~shaneck/MarkShaneck_BufferOverflows.pdf.

²¹⁵John Leyden, Worms Spread Faster, Blended Threats Grow, The Register, Oct. 2003, at 1, http://www.theregister.co.uk/2003/10/01/worms_spread_faster_blended_threats/.

²¹⁶Among five attacks used in the 1998 Lincoln Labs intrusion detection evaluation, three were essentially social engineering attacks that snooped user credentials and two were buffer overflows. Out of thirteen CERT advisories from 1998, nine involved buffer overflows. A least half of 1999 CERT advisories involved buffer overflows. See, e.g., Fred B. Schneider et al., TRUST IN CYBERSPACE passim (Fred B. Schneider ed., National Academy Press 1999); according to the (informal) Bugtraq security vulnerability survey, approximately two-thirds of respondents felt that buffer overflows are the leading cause of security vulnerability. Cowan, supra note 183; Bugtraq mailing list, available at http://msgs.securepoint.com/bugtraq/; Steve Bellovin, Buffer Overflows and Remote Root Exploits. Personal communication, cited in Crispin Cowan, supra note 183, at 5.

²¹⁷Edmund X. DeJesus, Drowning in Buffer Overflow Vulnerabilities, Security Wire Perspectives, May 24, 2004, http://searchsecurity.techtarget.com/originalContent/0,289142,sid14_gci966412,00.html.

²¹⁸State of Maine Public Utilities Commission, Inquiry Regarding the Entry of Verizon-Maine into the InterLATA (Long Distance) Telephone Market Pursuant to Section 271 of the Telecommunications Act of 1996, Docket No. 2000-849 (Oct. 18, 2000), available at http://www.bakernet.com/ecommerce/verizoncase.doc (wherein WorldCom claims that "the Slammer worm attack was not, as Verizon claims, an unforeseeable event that was beyond Verizon's control"); see also Hamilton v. Microsoft, Superior Court of California (Los Angeles), Case No. (2003) (alleging that the defendant, Microsoft, is aware that its actions increase Internet security risks of the kind that materialized); see also United States of America, In the Matter of GUESS?, Inc. and GUESS.com, Inc., http://www.ftc.gov/os/2003/06/guesscmp.htm (alleging that the subject information security breach was a well-known and foreseeable consequence of the vulnerability that the defendant had failed to fix).

²²²613 So. 2d 1272, 1274 (Ala. 1993).

²²⁴Id. at 1274-75.

²²⁶Grady, supra note 111, at 200.

²²⁷Travell v. Bannerman, 75 N.Y.S. 866 (App. Div. 1902).

²²⁸Donehue v. Duvall, 243 N.E.2d 222 (Ill. 1969).

²²⁹Segerman v. Jones, 259 A.2d 794 (Md. 1970).

²³⁰Symantec Internet Security Threat Report, supra note 66, at 47.

²³¹Root Access, Wikipedia, http://en.wikipedia.org/wiki/Root_access.

²³³Dorothy E. Denning, INFORMATION WARFARE AND SECURITY 214 (ACM Press 1999) (referring to malicious code executed via a buffer overflow as executing "with the privileges of the program it exploits, which is often root").

²³⁴See, e.g., Securiteam Advisory, DosEMU Buffer Overflow Assists in Gaining Root, http://www.securiteam.com/exploits/2GUPVSAQO0.html.

²³⁵See Szor, supra note 1, at 368.

²³⁶Nimda's other attack vectors were infection of Microsoft IIS web servers via a buffer overflow exploit; infection of network shares; and infection via Javascript added to web pages. See, e.g., Chen, supra note 11 (stating that other attack points frequently used in blended attacks include injecting malicious code into .exe files on a target system, creating world readable network shares, making multiple registry changes, and adding script code to html files).

²³⁷See, e.g., Symantec Security Response from 3-3-2003, http://securityresponse.symantec.com/avcenter/security/Content/3.3.2003.html ("A remotely exploitable vulnerability has been discovered in Sendmail. This vulnerability is due to a buffer overflow condition in the SMTP header parsing component. Remote attackers may exploit this vulnerability by connecting to target SMTP servers and transmitting them malformed data."); see also United States Computer Emergency Readiness Team, Technical Cyber Security Alert TA04-260A (Sept. 16, 2004), http://www.us-cert.gov/cas/techalerts/TA04-260A.html (describing a vulnerability in Microsoft's Graphic Device Interface Plus, which may allow remote executability of malicious code on an affected system, by reading an HTML-rendered e-mail message or opening a crafted JPEG image); see also Foundstone Labs Advisory—090502-PCRO, http://www.foundstone.com/advisories (follow the link to the year "2002"; then scroll down and find the advisory from "9/5/2002") (advisory on remotely exploitable buffer overflow vulnerability in PGP); see also DosEMU Buffer Overflow Assists in Gaining Root, http://www.securiteam.com/exploits/2GUPVSAQ0O.html (describing a buffer overflow vulnerability in a Linux application that enables root access to a malicious attacker); see also McAfee Samba Buffer Overflow, IntruShield Security Advisory SA20, http://www.mcafee.com/us/security/resources/sv_20.htm (describing buffer overflow vulnerability in the Samba server, a widely used, open-source UNIX compatible server, and advising that vulnerability allows remote attackers to gain root access). Software vendors, notably Microsoft, typically issue a patch to fix a vulnerability as soon as it is discovered. Users are not always diligent in implementing the patch, hence the continuing exploitability of some vulnerabilities even after corrective action by the vendor.

²³⁸Buffer Overflow in Microsoft Internet Explorer, Computing Services Security Information, UC Davis Computing Services (Nov. 10, 2004), http://www.nuid.ie/computing/security/index.html (also advising on other remotely exploitable vulnerabilities in Microsoft Internet Explorer).

²³⁹Symantec Internet Security Threat Report, supra note 184.

²⁴⁰See, e.g., Microsoft NetDDE Service Unauthenticated Remote Buffer Overflow (Jan. 21, 2005), http://www.ngssoftware.com/advisories/netddefull.txt (reporting a vulnerability in the Microsoft DDE service that allows a remote attacker to execute arbitrary code on a system without authentication). Microsoft has since released an update addressing the vulnerability: http://www.microsoft.com/technet/security/bulletin/MS04-031.mspx; see also Denning, supra note 233, at 214 (describing hackers executing malicious code remotely, "without logging in through an account or password," by exploiting buffer overflow vulnerabilities).

²⁴¹The vulnerability offered several opportunities to an attacker, including exploiting clients via a malicious web server. If a user were to browse an infected web site, the server would invoke a new session, and then, unbeknownst to the user, bring the user back to the web site but via the new session. At this point, a buffer overflow within Internet Explorer would allow the server to run unauthenticated arbitrary code on the client system. See Remotely Exploitable Buffer Overflow in Microsoft MDAC, http://security-protocols.com/modules.php?name=News&file=article&sid=1391.

I. Introduction

II. Malevolent software: Background

III. Liability issues in blended attacks

IV. Free radicals, the buffer overflow and blended attacks

V. Discussion and conclusion

ENDNOTES