It was first published by The Harvard Law Review, and he's now expanding it into a book. Zittrain is Professor of Internet Governance and Regulation, Oxford University, and Jack N. & Lillian R. Berkman Visiting Professor of Entrepreneurial Legal Studies at Harvard Law School. I will let him explain his request in his own words:

I've written before on the Microsoft antitrust case and on SCO (see http://papers.ssrn.com/sol3/papers.cfm?abstract_id=174110 and http://papers.ssrn.com/sol3/papers.cfm?abstract_id=529862 respectively). I've just finished a new paper on the future of the Net, in which I extol its open qualities but fear that a focus on an open Internet can too often exclude worrying about an open PC -- which I define in a broader fashion than the divide between free and proprietary software typically contemplates.

I think it's critically important that users retain general purpose PCs, even some with proprietary OSes, instead of "information appliances." I fear these appliances, like TiVo, can come to predominate -- or that the PC itself will morph towards becoming one, with new gatekeepers determining what code will or won't run on them, rather than the users themselves.

As I shift from the paper towards a book on the subject, I'd like to know where the argument could use improvement or elaboration. Any comments appreciated, and I'd be happy to try to respond to them or to questions.

...JZ"

I think he'd particularly like your input on the technical aspects of the paper. He is aware that not everyone will agree with every point he makes, but in true Open Source fashion, he's interested in hearing what you have to say, knowing that Groklaw readers have a high level of computer skills, for the most part, and that thoughtful comments can be expected.

The paper's downloadable as a PDF at SSRN, with free registration. Or, it's available at Harvard Law Review. (More papers by Professor Zittrain can be found here.) I've done a text version, with the author's permission, and with a lot of help from Carlo Graziani's bot, some new scripts from Mathfox, and some OCR and HTML help from jnc. I could not have done this without their help. The pagination is to match the PDF, as opposed to the Harvard Law Review's page numbering. Note that this article is copyrighted, and thus is not to be mirrored elsewhere. If you click on the footnote numbers, you can jump back and forth easily. Also, the various headings are linked.

******************************

THE GENERATIVE INTERNET

TABLE OF CONTENTS

A. Generative Technologies Defined ........7

1. Capacity for Leverage ........8
2. Adaptability..................8
3. Ease of Mastery................8
4. Accessibility ..................9
5. Generativity Revisited............9

B. The Generative PC ........9
C. The Generative Internet ........14
D. The Generative Grid ...........21

A. Generative Equilibrium ........23
B. Generativity as Vulnerability: The Cybersecurity Fulcrum .............30

1. A Threat Unanswered and Unrealized...........30
2. The PC/Network Grid and a Now-Realized Threat......35

A. Elements of a Postdiluvian Internet ......40

1. Information Appliances ............41
2. The Appliancized PC...............44

B. Implications of a Postdiluvian Internet: More Regulability, Less Generativity ..........48

A. Refining Principles of Internet Design and Governance .........56

1. Superseding the End-to-End Argument........56
2. Reframing the Internet Governance Debate .........59
3. Recognizing Interests in Tension with Generativity.......61
4. “Dual Machine” Solutions .........63

B. Policy Interventions ........64

1. Enabling Greater Opportunity for Direct Liability .......64
2. Establishing a Choice Between Generativity and Responsibility .......65

THE GENERATIVE INTERNET
Jonathan L. Zittrain^*

The generative capacity for unrelated and unaccredited audiences to build and distribute code and content through the Internet to its tens of millions of attached personal computers has ignited growth and innovation in information technology and has facilitated new creative endeavors. It has also given rise to regulatory and entrepreneurial backlashes. A further backlash among consumers is developing in response to security threats that exploit the openness of the Internet and of PCs to third-party contribution. A shift in consumer priorities from generativity to stability will compel undesirable responses from regulators and markets and, if unaddressed, could prove decisive in closing today’s open computing environments. This Article explains why PC openness is as important as network openness, as well as why today’s open network might give rise to unduly closed endpoints. It argues that the Internet is better conceptualized as a generative grid that includes both PCs and networks rather than as an open network indifferent to the configuration of its endpoints. Applying this framework, the Article explores ways — some of them bound to be unpopular among advocates of an open Internet represented by uncompromising end-to-end neutrality — in which the Internet can be made to satisfy genuine and pressing security concerns while retaining the most important generative aspects of today’s networked technology.

I. INTRODUCTION

2

Both mirroring and contributing to the Internet’s explosive growth has been the growth of the personal computer (PC): an affordable, multifunctional device that is readily available at retail outlets and easily reconfigurable by its users for any number of purposes. The audience writing software for PCs, including software that is Internet capable, is itself massive and varied.⁶ This diverse audience has driven the variety of applications powering the rapid technological innovation to which we have become accustomed, in turn driving innovation in expressive endeavors such as politics, entertainment, journalism, and education.⁷

Though the openness of the Internet and the PC to third-party contribution is now so persistent and universal as to seem inevitable, these

3

Internet technologists have for too long ignored Internet and PC security threats that might precipitate an unprecedented intervention into the way today’s Internet and PCs work. These security threats are more pervasive than the particular vulnerabilities arising from bugs in a particular operating system (OS): so long as OSs permit consumers to run third-party code — the sine qua non of a PC OS — users can execute malicious code and thereby endanger their own work and that of others on the network. A drumbeat of trust-related concerns plausibly could fuel a gradual but fundamental shift in consumer demand toward increased stability in computing platforms. Such a shift would interact with existing regulatory and commercial pressures — each grounded in advancing economically rational and often legally protected interests — to create solutions that make it not merely possible, but likely, that the personal computing and networking environment of tomorrow will be sadly hobbled, bearing little resemblance to the one that most of the world enjoys today.

The most plausible path along which the Internet might develop is one that finds greater stability by imposing greater constraint on, if not outright elimination of, the capacity of upstart innovators to demonstrate and deploy their genius to large audiences. Financial transactions over such an Internet will be more trustworthy, but the range of its users’ business models will be narrow. This Internet’s attached consumer hardware will be easier to understand and will be easier to use for purposes that the hardware manufacturers preconceive, but it will be off limits to amateur tinkering. Had such a state of affairs evolved by the Internet’s twenty-fifth anniversary in 1994, many of its formerly unusual and now central uses would never have developed because the software underlying those uses would have lacked a platform for exposure to, and acceptance by, a critical mass of users.

Those who treasure the Internet’s generative features must assess the prospects of either satisfying or frustrating the growing forces against those features so that a radically different technology configuration need not come about. Precisely because the future is uncertain, those who care about openness and the innovation that today’s Internet and PC facilitate should not sacrifice the good to the perfect — or the future to the present — by seeking simply to maintain a tenuous technological status quo in the face of inexorable pressure to change. Rather, we should establish the principles that will blunt the most un-

4

Although these matters are of central importance to cyberlaw, they have generally remained out of focus in our field’s evolving literature, which has struggled to identify precisely what is so valuable about today’s Internet and which has focused on network openness to the exclusion of endpoint openness. Recognition of the importance and precariousness of the generative Internet has significant implications for, among other issues, the case that Internet service providers (ISPs) should unswervingly observe end-to-end neutrality, the proper focus of efforts under the rubric of “Internet governance,” the multifaceted debate over digital rights management (DRM), and prospects for Internet
censorship and filtering by authoritarian regimes.

Scholars such as Professors Yochai Benkler, Mark Lemley, and Lawrence Lessig have crystallized concern about keeping the Internet “open,” translating a decades-old technical end-to-end argument concerning simplicity in network protocol design into a claim that ISPs should refrain from discriminating against particular sources or types of data.⁸ There is much merit to an open Internet, but this argument is really a proxy for something deeper: a generative networked grid. Those who make paramount “network neutrality” derived from end-to-end theory confuse means and ends, focusing on “network” without regard to a particular network policy’s influence on the design of network endpoints such as PCs. As a result of this focus, political advocates of end-to-end are too often myopic; many writers seem to presume current PC and OS architecture to be fixed in an “open” position. If they can be persuaded to see a larger picture, they may agree to some compromises at the network level. If complete fidelity to end-to-end network neutrality persists, our PCs may be replaced by information appliances or may undergo a transformation from open platforms to gated communities to prisons, creating a consumer information environment that betrays the very principles that animate end-to-end theory.

The much-touted differences between free and proprietary PC OSs may not capture what is most important to the Internet’s future. Proprietary systems can remain “open,” as many do, by permitting unaffiliated third parties to write superseding programs and permitting PC

5

The focus on the management of domain names among those participating in dialogues about Internet governance is equally unfortunate. Too much scholarly effort has been devoted to the question of institutional governance of this small and shrinking aspect of the Internet landscape.⁹ When juxtaposed with the generativity problem, domain names matter little. Cyberlaw’s challenge ought to be to find ways of regulating — though not necessarily through direct state action — which code can and cannot be readily disseminated and run upon the generative grid of Internet and PCs, lest consumer sentiment and preexisting regulatory pressures prematurely and tragically terminate the grand experiment that is the Internet today.

New software tools might enable collective regulation if they can inform Internet users about the nature of new code by drawing on knowledge from the Internet itself, generated by the experiences of others who have run such code. Collective regulation might also entail new OS designs that chart a middle path between a locked-down PC on the one hand and an utterly open one on the other, such as an OS that permits users to shift their computers between “safe” and “experimental” modes, classifying new or otherwise suspect code as suited for only the experimental zone of the machine.

In other words, to Professor Lessig and those more categorically libertarian than he, this Article seeks to explain why drawing a bright line against nearly any form of increased Internet regulability is no longer tenable. Those concerned about preserving flexibility in Internet behavior and coding should participate meaningfully in debates about changes to network and PC architecture, helping to ameliorate rather than ignore the problems caused by such flexibility that, if left unchecked, will result in unfortunate, hamhanded changes to the way mainstream consumers experience cyberspace.

An understanding of the value of generativity, the risks of its excesses, and the possible means to reconcile the two, stands to reinvigorate the debate over DRM systems. In particular, this understanding promises to show how, despite no discernable movement toward the

6

Finally, some methods that could be plausibly employed in the Western world to make PCs more tame for stability’s sake can greatly affect the prospects for Internet censorship and surveillance by authoritarian regimes in other parts of the world. If PC technology is inaptly refined or replaced to create new points of control in the distribution of new code, such technology is horizontally portable to these regimes — regimes that currently effect their censorship through leaky network blockages rather than PC lockdown or replacement.

II. A MAPPING OF GENERATIVE TECHNOLOGIES

7

A. Generative Technologies Defined

1. Capacity for Leverage. — Our world teems with useful objects, natural and artificial, tangible and intangible; leverage describes the extent to which these objects enable valuable accomplishments that otherwise would be either impossible or not worth the effort to achieve. Examples of devices or systems that have a high capacity for leverage include a lever itself (with respect to lifting physical objects), a band saw (cutting them), an airplane (getting from one place to another), a piece of paper (hosting written language), or an alphabet (constructing words). A generative technology makes difficult jobs easier. The more effort a device or technology saves — compare a sharp knife to a dull one — the more generative it is. The greater the variety of accomplishments it enables — compare a sharp Swiss Army knife to a sharp regular knife — the more generative it is.¹¹

2. Adaptability. — Adaptability refers to both the breadth of a technology’s use without change and the readiness with which it might be modified to broaden its range of uses. A given instrumentality may be highly leveraging, yet suited only to a limited range of applications. For example, although a plowshare can enable the planting of a variety of seeds, planting is its essential purpose. Its comparative leverage quickly vanishes when devoted to other tasks, such as holding doors open, and it is not readily modifiable for new purposes. The same goes for swords (presuming they are indeed difficult to beat into plowshares), guns, chairs, band saws, and even airplanes. In contrast, paper obtained for writing can instead (or additionally) be used to wrap fish. A technology that offers hundreds of different additional kinds of uses is more adaptable and, all else equal, more generative than a technology that offers fewer kinds of uses. Adaptability in a tool better permits leverage for previously unforeseen purposes.

3. Ease of Mastery. — A technology’s ease of mastery reflects how easy it is for broad audiences both to adopt and to adapt it: how much skill is necessary to make use of its leverage for tasks they care about, regardless of whether the technology was designed with those tasks in mind. An airplane is neither easy to fly nor easy to modify for new purposes, not only because it is not inherently adaptable to purposes other than transportation, but also because of the skill required to make whatever transformative modifications might be possible and

8

4. Accessibility. — The more readily people can come to use and control a technology, along with what information might be required to master it, the more accessible the technology is. Barriers to accessibility can include the sheer expense of producing (and therefore consuming) the technology, taxes and regulations imposed on its adoption or use (for example, to serve a government interest directly or to help private parties monopolize the technology through intellectual property law), and the use of secrecy and obfuscation by its producers to maintain scarcity without necessarily relying upon an affirmative intellectual property interest.¹²

By this reckoning of accessibility, paper and plowshares are highly accessible, planes hardly at all, and cars somewhere in the middle. It might be easy to learn how to drive a car, but cars are expensive, and the privilege of driving, once earned by demonstrating driving skill, is revocable by the government. Moreover, given the nature of cars and driving, such revocation is not prohibitively expensive to enforce effectively.

5. Generativity Revisited. — As defined by these four criteria, generativity increases with the ability of users to generate new, valuable uses that are easy to distribute and are in turn sources of further innovation. It is difficult to identify in 2006 a technology bundle more generative than the PC and the Internet to which it attaches.

B. The Generative PC

9

The essence — and genius — of software standing alone is that it allows unmodified hardware to execute a new algorithm, obviating the difficult and expensive process of embedding that algorithm in hardware.¹⁴ PCs carry on research computing’s tradition of separating hardware and software by allowing the user to input new code. Such code can be loaded and run even once the machine is solely in the consumer’s hands. Thus, the manufacturer can write new software after the computer leaves the factory, and a consumer needs to know merely how to load the cassette, diskette, or cartridge containing the software to enjoy its benefits. (In today’s networked environment, the consumer need not take any action at all for such reprogramming to take place.) Further, software is comparatively easy for the manufacturer to develop because PCs carry on another sensible tradition of their institutional forbears: they make use of OSs.¹⁵ OSs provide a higher level of abstraction at which the programmer can implement his or her algorithms, allowing a programmer to take shortcuts when creating software.¹⁶ The ability to reprogram using external inputs provides adaptability; OSs provide leverage and ease of mastery.

Most significant, PCs were and are accessible. They were designed to run software not written by the PC manufacturer or OS publisher, including software written by those with whom these manufacturers had no special arrangements.¹⁷ Early PC manufacturers not only published the documentation necessary to write software that would run on their OSs, but also bundled high-level programming languages along with their computers so that users themselves could learn to

10

That the resulting PC was one that its own users could — and did — program is significant. The PC’s truly transformative potential is fully understood, however, as a product of its overall generativity.

PCs were genuinely adaptable to any number of undertakings by people with very different backgrounds and goals. The early OSs lacked capabilities we take for granted today, such as multitasking, but they made it possible for programmers to write useful software with modest amounts of effort and understanding.²⁰

Users who wrote their own software and thought it suited for general use could hang out the shingle in the software business or simply share the software with others. A market in third-party software developed and saw a range of players, from commercial software publishers employing thousands of people, to collaborative software projects that made their outputs freely available, to hobbyists who showcased and sold their wares by advertising in computer user magazines or through local computer user groups.²¹ Such a range of developers enhanced the variety of applications that were written not only because accessibility arguably increased the sheer number of people coding, but also because people coded for different reasons. While

11

This variety is important because the usual mechanisms that firms use to gauge and serve market demand for particular applications may not provide enough insight to generate every valuable application. It might seem obvious, for example, that a spreadsheet would be of great use to someone stuck with an adding machine, but the differences between the new application and the old method may prevent firms from recognizing the former as better serving other tasks. The PC, however, allowed multiple paths for the creation of new applications: Firms could attempt to gauge market need and then write and market an application to meet that need. Alternatively, people with needs could commission firms to write software. Finally, some could simply write the software themselves.

This configuration — a varied audience of software writers, most unaffiliated with the manufacturers of the PCs for which they wrote software — turned out to benefit the manufacturers, too, because greater availability of software enhanced the value of PCs and their OSs. Although many users were tinkerers, many more users sought PCs to accomplish particular purposes nearly out of the box; PC manufacturers could tout to the latter the instant uses of their machines provided by the growing body of available software.

Adapting a metaphor from Internet architecture, PC architecture can be understood as shaped like an hourglass.²² PC OSs sit in the narrow middle of the hourglass; most evolved slowly because proprietary OSs were closed to significant outside innovation. At the broad bottom of the PC hourglass lies cheap, commoditized hardware. At the broad top of the hourglass sits the application layer, in which most PC innovation takes place. Indeed, the computer and its OS are products, not services, and although the product life might be extended through upgrades, such as the addition of new physical memory chips or OS upgrades, the true value of the PC lies in its availability and stability as a platform for further innovation — running applications from a variety of sources, including the users themselves, and thus lev-

12

Despite the scholarly and commercial attention paid to the debate between free and proprietary OSs,²⁴ the generativity of the PC depends little on whether its OS is free and open or proprietary and closed — that is, whether the OS may be modified by end users and third-party software developers. A closed OS like Windows can be and has been a highly generative technology. Windows is generative, for instance, because its application programming interfaces enable a programmer to rework nearly any part of the PC’s functionality and give external developers ready access to the PC’s hardware inputs and outputs, including scanners, cameras, printers, monitors, and mass storage. Therefore, nearly any desired change to the way Windows works can be imitated by code placed at the application level.²⁵ The qualities of an OS that are most important have to do with making computing power, and the myriad tasks it can accomplish, available for varied audiences to adapt, share, and use through the ready creation and exchange of application software. If this metric is used, debates over whether the OS source code itself should be modifiable are secondary.

Closed but generative OSs — largely Microsoft DOS and then Microsoft Windows — dominate the latter half of the story of the PC’s rise, and their early applications outperformed their appliance-based counterparts. For example, despite the head start from IBM’s production of a paper tape–based word processing unit in World War II,

13

The technology and market structures that account for the highly generative PC have endured despite the roller coaster of hardware and software producer fragmentation and consolidation that make PC industry history otherwise heterogeneous.³⁰ These structures and their implications can, and plausibly will, change in the near future. To understand these potential changes, their implications, and why they should be viewed with concern and addressed by action designed to minimize their untoward effects, it is important first to understand roughly analogous changes to the structure of the Internet that are presently afoot. As the remainder of this Part explains, the line between the PC and the Internet has become vanishingly thin, and it is no longer helpful to consider them separately when thinking about the future of consumer information technology and its relation to innovation and creativity.

C. The Generative Internet

14

How did this state of affairs come to pass? Unlike, say, FedEx, whose wildly successful offline package delivery network depended initially on the financial support of venture capitalists to create an efficient physical infrastructure, those individuals thinking about the Internet in the 1960s and 1970s planned a network that would cobble together existing networks and then wring as much use as possible from them.³⁶

The network’s design is perhaps best crystallized in a seminal 1984 paper entitled End-to-End Arguments in System Design.³⁷ As this paper describes, the Internet’s framers intended an hourglass design, with a simple set of narrow and slow-changing protocols in the middle, resting on an open stable of physical carriers at the bottom and any number of applications written by third parties on the top. The Internet hourglass, despite having been conceived by an utterly different group of architects from those who designed or structured the market

15

The authors of End-to-End Arguments describe, as an engineering matter, why it is better to keep the basic network operating protocols simple — and thus to implement desired features at, or at least near, the endpoints of the networks.³⁸ Such features as error correction in data transmission are best executed by client-side applications that check whether data has reached its destination intact rather than by the routers in the middle of the chain that pass data along.³⁹ Ideas that entail changing the way the routers on the Internet work — ideas that try to make them “smarter” and more discerning about the data they choose to pass — challenge end-to-end philosophy.

The design of the Internet also reflects both the resource limitations and intellectual interests of its creators, who were primarily academic researchers and moonlighting corporate engineers. These individuals did not command the vast resources needed to implement a global network and had little interest in exercising control over the network or its users’ behavior.⁴⁰ Energy spent running the network was seen as a burden; the engineers preferred to design elegant and efficient protocols whose success was measured precisely by their ability to run without effort. Keeping options open for growth and future development was seen as sensible,⁴¹ and abuse of the network was of little worry because the people using it were the very people designing it — a culturally homogenous set of people bound by their desire to see the network work.

Internet designers recognized the absence of persistent connections between any two nodes of the network,⁴² and they wanted to allow additions to the network that neither taxed a central hub nor required centrally managed adjustments to overall network topography.⁴³ These constraints inspired the development of a stateless protocol — one that did not presume continuous connections or centralized knowledge of those connections — and packet switching,⁴⁴ which broke all

16

A flexible, robust platform for innovation from all corners sprang from these technical and social underpinnings. Two further historical developments assured that an easy-to-master Internet would also be extraordinarily accessible. First, the early Internet consisted of nodes primarily at university computer science departments, U.S. government research units, and select technology companies with an interest in cutting-edge network research.⁴⁶ These institutions collaborated on advances in bandwidth management and tools for researchers to use for communication and discussion.⁴⁷ But consumer applications were nowhere to be found until the Internet began accepting commercial interconnections without requiring academic or government research justifications, and the population at large was solicited to join.⁴⁸ This historical development — the withering away of the norms against commercial use and broad interconnection that had been reflected in a National Science Foundation admonishment that its contribution to the functioning Internet backbone be used for noncommercial purposes⁴⁹ — greatly increased the Internet’s generativity. It opened development of networked technologies to a broad, commercially driven audience that individual companies running proprietary services did not think to invite and that the original designers of the Internet would not have thought to include in the design process.

A second historical development is easily overlooked because it may in retrospect seem inevitable: the dominance of the Internet as the network to which PCs connected, rather than the emergence of proprietary networks analogous to the information appliances that PCs themselves beat. The first large-scale networking of consumer PCs

17

The proprietary services could be leveraged for certain tasks, and their technologies were adaptable to many purposes and easy to master, but consumers’ and outsiders’ inability to tinker easily with the services limited their generativity.⁵⁴ They were more like early video game consoles⁵⁵ than PCs: capable of multiple uses, through the development of individual “cartridges” approved by central management, yet slow to evolve because potential audiences of developers were slowed or shut out by centralized control over the network’s services.

The computers first attached to the Internet were mainframes and minicomputers of the sort typically found within university computer science departments,⁵⁶ and early desktop access to the Internet came through specialized nonconsumer workstations, many running variants

18

A single hobbyist took advantage of PC generativity and produced and distributed the missing technological link. Peter Tattam, an employee in the psychology department of the University of Tasmania, wrote Trumpet Winsock, a program that allowed owners of PCs running Microsoft Windows to forge a point-to-point Internet connection with the servers run by the nascent ISP industry.⁵⁹ Ready consumer access to Internet-enabled applications such as Winsock, coupled with the development of graphical World Wide Web protocols and the PC browsers to support them — all initially noncommercial ventures — marked the beginning of the end of proprietary information services and peer-to-peer telephone-networked environments like electronic bulletin boards. After recognizing the popularity of Tattam’s software, Microsoft bundled the functionality of Winsock with later versions of Windows 95.⁶⁰

As PC users found themselves increasingly able to access the Internet, proprietary network operators cum content providers scrambled to reorient their business models away from corralled content and toward accessibility to the wider Internet.⁶¹ These online service providers quickly became mere ISPs, with their users branching out to the thriving Internet for programs and services.⁶² Services like CompuServe’s “Electronic Mall” — an e-commerce service allowing outside vendors, through arrangements with CompuServe, to sell products to subscrib-

19

The greater generativity of the Internet compared to that of proprietary networked content providers created a network externality: as more information consumers found their way to the Internet, there was more reason for would-be sources of information to set up shop there, in turn attracting more consumers. Dispersed third parties could and did write clients and servers for instant messaging,⁶⁴ web browsing,⁶⁵ e-mail exchange,⁶⁶ and Internet searching.⁶⁷ Furthermore, the Internet remained broadly accessible: anyone could cheaply sign on, immediately becoming a node on the Internet equal to all others in information exchange capacity, limited only by bandwidth. Today, due largely to the happenstance of default settings on consumer wireless routers, Internet access might be free simply because one is sitting on a park bench near the apartment of a broadband subscriber who has “gone wireless.”⁶⁸

The resulting Internet is a network that no one in particular owns and that anyone can join. Of course, joining requires the acquiescence of at least one current Internet participant, but if one is turned away at one place, there are innumerable others to court, and commercial ISPs provide service at commoditized rates.⁶⁹ Those who want to offer services on the Internet or connect a formerly self-contained local network — such as a school that wants to link its networked computers

20

D. The Generative Grid

21

Significantly, the last several years have witnessed a proliferation of PCs hosting broadband Internet connections.⁷⁵ The generative PC has become intertwined with the generative Internet, and the whole is now greater than the sum of its parts. A critical mass of always-on computers means that processing power for many tasks, ranging from difficult mathematical computations to rapid transmission of otherwise prohibitively large files, can be distributed among hundreds, thousands, or millions of PCs.⁷⁶ Similarly, it means that much of the information that once needed to reside on a user’s PC to remain conveniently accessible — documents, e-mail, photos, and the like — can instead be stored somewhere on the Internet.⁷⁷ So, too, can the programs that a user might care to run.

This still-emerging “generative grid” expands the boundaries of leverage, adaptability, and accessibility for information technology. It also raises the ante for the project of cyberlaw because the slope of this innovative curve may nonetheless soon be constrained by some of the very factors that have made it so steep. Such constraints may arise because generativity is vulnerability in the current order: the fact that tens of millions of machines are connected to networks that can convey reprogramming in a matter of seconds means that those computers stand exposed to near-instantaneous change. This kind of generativity keeps publishers vulnerable to the latest tools of intellectual property infringement, crafted ever more cleverly to evade monitoring and control, and available for installation within moments everywhere. It also opens PCs to the prospect of mass infection by a computer virus that exploits either user ignorance or a security vulnerability that allows code from the network to run on a PC without approval by its owner. Shoring up these vulnerabilities will require substantial changes in some aspects of the grid, and such changes are sure to affect the current level of generativity. Faced with the prospect of such changes, we

22

III. GENERATIVE DISCONTENT

A. Generative Equilibrium

23

In the late 1990s, Professor Lessig and others argued that the debate was too narrow, pioneering yet a third strand of scholarship. Professor Lessig argued that a fundamental insight justifying cyberlaw as a distinct field is the way in which technology — as much as the law itself — can subtly but profoundly affect people’s behavior: “Code is law.”⁸¹ He maintained that the real projects of cyberlaw are both to correct the common misperception that the Internet is permanently unregulable and to reframe doctrinal debates as broader policy-oriented ones, asking what level of regulability would be appropriate to build into digital architectures. For his part, Professor Lessig argued that policymakers should typically refrain from using the powers of regulation through code — powers that they have often failed, in the first instance, to realize they possess.⁸²

The notion that code is law undergirds a powerful theory, but some of its most troubling empirical predictions about the use of code to restrict individual freedom have not yet come to pass. Work by professors such as Julie Cohen and Pamela Samuelson has echoed Professor Lessig’s fears about the ways technology can unduly constrain individ-

24

Trusted systems are systems that can be trusted by outsiders against the people who use them.⁸⁴ In the consumer information technology context, such systems are typically described as “copyright management” or “rights management” systems, although such terminology is loaded. As critics have been quick to point out, the protections afforded by these systems need not bear any particular relationship to the rights granted under, say, U.S. copyright law.⁸⁵ Rather, the possible technological restrictions on what a user may do are determined by the architects themselves and thus may (and often do) prohibit many otherwise legal uses. An electronic book accessed through a rights management system might, for example, have a limitation on the number of times it can be printed out, and should the user figure out how to print it without regard to the limitation, no fair use defense would be available. Similarly, libraries that subscribe to electronic material delivered through copyright management systems may find themselves technologically incapable of lending out that material the way a traditional library lends out a book, even though the act of lending is a privilege — a defense to copyright infringement for unlawful “distribution” — under the first sale doctrine.⁸⁶

The debate over technologically enforced “private” copyright schemes in the United States grew after the Digital Millennium Copy-

25

These are the central fears of the “code is law” theory as applied to the Internet and PCs: technical barriers prevent people from making use of intellectual works, and nearly any destruction of those barriers, even to enable lawful use, is unlawful. But these fears have largely remained hypothetical. To be sure, since the DMCA’s passage a number of unobtrusive rights management schemes have entered wide circulation.⁹⁰ Yet only a handful of cases have been brought against those who have cracked such schemes,⁹¹ and there has been little if any decrease in consumers’ capacity to copy digital works, whether for fair use or for wholesale piracy. Programmers have historically been able to crack nearly every PC protection scheme, and those less technically inclined can go to the trouble of generating a fresh copy of a digital work without having to crack the scheme, exploiting the “analog

26

Thus, the mid-1990s’ fears of digital lockdown through trusted systems may seem premature or unfounded. As a practical matter, any scheme designed to protect content finds itself rapidly hacked, and the hack (or the content protected) in turn finds itself shared with technically unsophisticated PC owners. Alternatively, the analog hole can be used to create a new, unprotected master copy of protected content. The fact remains that so long as code can be freely written by anyone and easily distributed to run on PC platforms, trusted systems can serve as no more than speed bumps or velvet ropes — barriers that the public circumvents should they become more than mere inconveniences. Apple’s iTunes Music Store is a good example of this phenomenon: music tracks purchased through iTunes are encrypted with Apple’s proprietary scheme,⁹⁴ and there are some limitations on their use that, although unnoticeable to most consumers, are designed to prevent the tracks from immediately circulating on peer-to-peer networks.⁹⁵ But the scheme is easily circumvented by taking music purchased from the store, burning it onto a standard audio CD, and then re-ripping the CD into an unprotected format, such as MP3.⁹⁶

An important claim endures from the third “code is law” strand of cyberlaw scholarship. Professors Samuelson, Lessig, and Cohen were right to raise alarm about such possibilities as losing the ability to read anonymously, to lend a copyrighted work to a friend, and to make fair use of materials that are encrypted for mere viewing. However, the second strand of scholarship, which includes the empirical claims

27

The first strand, which advocates incremental doctrinal adaptation to new technologies, explains how Internet regulators have indeed acted with a light touch even when they might have had more heavy-handed options. As I discuss elsewhere, the history of Internet regulation in the Western world has been tentative and restrained, with a focus by agile Internet regulators on gatekeeping regimes, effected through discrete third parties within a sovereign’s jurisdiction or through revisions to technical architectures.⁹⁷ Although Internet regulators are powerful in that they can shape, apply, and enforce the authority of major governments, many have nevertheless proven willing to abstain from major intervention.

This lack of intervention has persisted even as the mainstream adoption of the Internet has increased the scale of interests that Internet uses threaten. Indeed, until 2001, the din of awe and celebration surrounding the Internet’s success, including the run-up in stock market valuations led by dot-coms, drowned out many objections to and discussion about Internet use and reform — who would want to disturb a goose laying golden eggs? Further, many viewed the Internet as immutable and thought the problems it generated should be dealt with piecemeal, without considering changes to the way the network functioned — a form of “is-ism” that Professor Lessig’s third strand of cyberlaw scholarship famously challenged in the late 1990s even as he advocated that the Internet’s technology remain essentially as is, unfettered by regulation.⁹⁸

By 2002 the dot-com hangover was in full sway, serving as a backdrop to a more jaundiced mainstream view of the value of the Internet. The uptake of consumer broadband, coupled with innovative but disruptive applications like the file-sharing service Napster — launched by amateurs but funded to the tune of millions of dollars during the boom — inspired fear and skepticism rather than envy and

28

By incorporating the trajectory of disruption by the generative Internet followed by comparatively mild reaction by regulators, the three formerly competing strands of cyberlaw scholarship can be reconciled to explain the path of the Internet: Locked-down PCs are possible but undesirable. Regulators have been unwilling to take intrusive measures to lock down the generative PC and have not tried to trim the Internet’s corresponding generative features to any significant degree. The Internet’s growth has made the underlying problems of concern to regulators more acute. Despite their unwillingness to act to bring about change, regulators would welcome and even encourage a PC/Internet grid that is less exceptional and more regulable.

29

B. Generativity as Vulnerability: The Cybersecurity Fulcrum

Consumers hold the key to the balance of power for tomorrow’s Internet. They are powerful because they drive markets and because many vote. If they remain satisfied with the Internet’s generative characteristics — continuing to buy the hardware and software that make it possible and to subscribe to ISPs that offer unfiltered access to the Internet at large — then the regulatory and industry forces that might otherwise wish to constrain the Internet will remain in check, maintaining a generative equilibrium. If, in contrast, consumers change their view — not simply tolerating a locked-down Internet but embracing it outright — then the balance among competing interests that has favored generative innovation will tilt away from it. Such a shift is and will continue to be propelled by an underappreciated phenomenon: the security vulnerabilities inherent in a generative Internet populated by mainstream consumers possessing powerful PCs.

The openness, decentralization, and parity of Internet nodes described in Part II are conventionally understood to contribute greatly to network security because a network without a center will have no central point of breakage. But these characteristics also create vulnerabilities, especially when united with both the flaws in the machines connecting to the network and the users’ lack of technical sophistication. These vulnerabilities have been individually noted but collectively ignored. A decentralized Internet does not lend itself well to collective action, and its vulnerabilities have too readily been viewed as important but not urgent. Dealing with such vulnerabilities collectively can no longer be forestalled, even though a locus of responsibility for action is difficult to find within such a diffuse configuration.

1. A Threat Unanswered and Unrealized. — On the evening of November 2, 1988, a Cornell University graduate student named Robert Tappan Morris transmitted a small piece of software over the Internet from Cornell to MIT.¹⁰² At the time, few PCs were attached to the Internet; rather, it was the province of mainframes, minicom-

30

The sole purpose of the software — called variously a “virus” or, because it could transmit itself independently, a “worm” — was to propagate, attempting to crack each newly discovered computer by guessing user passwords or exploiting a known security flaw.¹⁰⁵ It succeeded. By the following morning, a large number of machines on the Internet had ground to a near halt. Each had been running the same flavor of vulnerable Unix software that the University of California at Berkeley had made freely available.¹⁰⁶ Estimates of the number of computers affected ranged from 1000 to 6000.¹⁰⁷

Within a day or two, puzzled network administrators reverse engineered the worm.¹⁰⁸ Its methods of propagation were discovered, and the Berkeley computer science department coordinated patches for vulnerable Unix systems.¹⁰⁹ These patches were made available over the Internet, at least to those users who had not elected to disconnect their machines to prevent reinfection.¹¹⁰ The worm was traced to Morris, who apologized; a later criminal prosecution for the act resulted in his receiving three years’ probation, 400 hours of community service, and a $10,050 fine.¹¹¹

The Morris worm was a watershed event because it was the first virus spread within an always-on network linking always-on computers. Despite the dramatically smaller and demographically dissimi-

31

The computers of the 1988 Internet could be compromised because they were general-purpose machines, running OSs with which outsiders could become familiar and for which these outsiders could write executable code. They were powerful enough to run multiple programs and host multiple users simultaneously.¹¹² They were generative. When first compromised, they were able to meet the demands of both the worm and their regular users, thereby buying time for the worm’s propagation before human intervention was made necessary.¹¹³ The OS and other software running on the machines were not perfect: they contained flaws that rendered them more generatively accessible than their designers intended.¹¹⁴ More important, even without such flaws, the machines were designed to be operated at a distance and to receive and run software sent from a distance. They were powered and attached to a network continuously, even when not in use by their owners. And the users of these machines were lackadaisical about implementing available fixes to known software vulnerabilities — and often utterly, mechanistically predictable in the passwords they chose to protect entry into their computer accounts.¹¹⁵

Each of these facts contributing to the vulnerabilities of networked life remains true today, and the striking feature is how few recurrences of truly disruptive security incidents there have been since 1988. Remarkably, a network designed for communication among academic and government researchers scaled beautifully as hundreds of millions of new users signed on during the 1990s, a feat all the more impressive when one considers how technologically unsavvy most new users were in comparison to the 1988 crowd. However heedless network administrators in 1988 were of good security hygiene, mainstream consumers of the 1990s were far worse. Few knew how to manage or code their PCs, much less how to apply patches rigorously or observe good password security.

32

The worm did not literally “infect” the Internet itself because it did not reprogram the Internet’s various distributed routing mechanisms and corresponding links. Instead, the burden on the network was simply increased traffic, as the worm impelled computers to communicate with each other while it attempted to spread ad infinitum.¹¹⁷ The computers actually infected were managed by disparate groups and individuals who answered to no particular authority for their use or for failing to secure them against compromise.¹¹⁸ These features led those individuals overseeing network protocols and operation to conclude that the worm was not their problem.

Engineers concerned with the Internet protocols instead responded to the Morris worm incident by suggesting that the appropriate solution was more professionalism among coders. The Internet Engineering Task Force — the far-flung, unincorporated group of engineers who work on Internet standards and who have defined its protocols through a series of formal “request for comments” documents, or RFCs — published informational RFC 1135 as a postmortem on the worm incident.¹¹⁹ After describing the technical details of the worm, the RFC focused on “computer ethics” and the need to instill and enforce ethical standards as new people — mostly young computer scientists such as Morris — signed on to the Internet.¹²⁰

The state of play in 1988, then, was to acknowledge the gravity of the worm incident while avoiding structural overreaction to it. The most concrete outcomes were a criminal prosecution of Morris, an apparent consensus within the Internet technical community regarding individual responsibility for the problem, and the Department of Defense’s creation of the CERT Coordination Center to monitor the overall health of the Internet and dispatch news of security threats¹²¹ —

33

Compared to the computers connected to the 1988 Internet, the computers connected to the proprietary consumer networks of the 1980s described in Part II were not as fundamentally vulnerable because those networks could be much more readily purged of particular security flaws without sacrificing the very foundations of their existence.¹²² For example, although it would have been resource intensive, CompuServe could have attempted to scan its network traffic at designated gateways to prevent the spread of a worm. More important, no worm could have spread through CompuServe in the manner of Morris’s because the computers attached to CompuServe were configured as mere “dumb terminals”: they exchanged data, not executable files, with CompuServe and therefore did not run software from the CompuServe network. Moreover, the mainframe computers at CompuServe with which those dumb terminals communicated were designed to ensure that the line between users and programmers was observed and digitally enforced.¹²³

Like CompuServe, the U.S. long-distance telephone network of the 1970s was intended to convey data — in the form of telephone conversations — rather than code between consumers. In the early 1970s, several consumers who were curious about the workings of the telephone network discovered that telephone lines used a tone at a frequency of 2600 Hertz to indicate that they were idle.¹²⁴ As fortune would have it, a toy whistle packaged as a prize in boxes of Cap’n Crunch cereal could, when one hole was covered, generate exactly that tone.¹²⁵ People in the know could dial toll-free numbers from their home phones, blow the whistle to clear but not disconnect the line, and then dial a new, non–toll free number, which would be connected without incurring a charge.¹²⁶ When this imaginative scheme came to light, AT&T reconfigured the network so that the 2600-Hertz tone no longer controlled it.¹²⁷ Indeed, the entire notion of in-band signaling was eliminated so that controlling the network required something other than generating sound into the telephone mouthpiece. To accomplish this reconfiguration, AT&T separated the data and executa-

34

An understanding of AT&T’s long-distance telephone network and its proprietary information service counterparts, then, reveals both the benefit and the bane of yesterday’s Internet and today’s generative grid: the Internet’s very channels of communication are also channels of control. What makes the Internet so generative is that it can transmit both executable code to PCs and data for PC users. This duality allows one user to access her news feeds while another user creates a program that aggregates hundreds of news feeds and then provides that program to millions of other users. If one separated data from the executable code as AT&T or CompuServe did, it would incapacitate the generative power of the Internet.

CompuServe’s mainframes were the servers, and the consumer computers were the clients. On AT&T’s network, both the caller and the call’s recipient were “clients,” with their link orchestrated by a centralized system of switches — the servers. On the Internet of both yesterday and today, a server is definable only circularly: it is a computer that has attached itself to the Internet and made itself available as a host to others — no more, no less.¹²⁸ A generative PC can be used as a client or it can be turned into a website that others access. Lines between consumer and programmer are self-drawn and self-enforced.

Given this configuration, it is not surprising that there was little momentum for collective action after the Morris worm scare. The decentralized, nonproprietary ownership of the Internet and its computers made it difficult to implement any structural revisions to the way they worked. And more important, it was simply not clear what curative steps would not entail drastic and purpose-altering changes to the very fabric of the Internet: the notion was so wildly out of proportion to the level of the perceived threat that it was not even broached.

2. The PC/Network Grid and a Now-Realized Threat. — In the absence of significant reform, then, how did the Internet largely dodge further bullets? A thorough answer draws on a number of factors, each instructive for the situation in which the grid finds itself today. First, the computer scientists were right that the ethos of the time frowned upon destructive hacking.¹²⁹ Even Morris’s worm arguably

35

Second, network operations centers at universities and other institutions became full-time facilities, with professional administrators who more consistently heeded the admonitions to update their patches regularly and to scout for security breaches. Administrators carried beepers and were prepared to intervene quickly in the case of a system intrusion. And even though mainstream consumers began connecting unsecured PCs to the Internet in earnest by the mid-1990s, their machines generally used temporary dial-up connections, greatly reducing the amount of time per day during which their computers were exposed to security threats and might contribute to the problem.

Last, there was no commercial incentive to write viruses — they were generally written for fun. Thus, there was no reason that substantial resources would be invested in creating viruses or in making them especially virulent.¹³²

Each of these factors that created a natural security bulwark has been attenuating as powerful networked PCs reach consumers’ hands. Universal ethics have become infeasible as the network has become ubiquitous: truly anyone who can find a computer and a connection is allowed online. Additionally, consumers have been transitioning to

36

CERT Coordination Center statistics reflect this sea change. The organization began documenting the number of attacks against Internet-connected systems — called “incidents” — at its founding in 1988. As Figure 1 shows, the increase in incidents since 1997 has been exponential, roughly doubling each year through 2003.

37

CERT announced in 2004 that it would no longer keep track of this figure because attacks have become so commonplace and widespread as to be indistinguishable from one another.¹³⁹

At the time of the Morris worm, there were an estimated 60,000 distinct computers, or “hosts,” on the Internet.¹⁴⁰ In 2005, the count was over 353 million,¹⁴¹ and worldwide PCs in use were estimated at almost 900 million.¹⁴² This massive set of always-on, powerful PCs, many with high-speed Internet connections and run by unskilled users, is a phenomenon new to the twenty-first century. Today’s viruses are highly and nearly instantly communicable, capable of sweeping through a substantial worldwide population in a matter of hours.¹⁴³

38

Well-crafted worms and viruses routinely infect swaths of Internet- connected PCs. In 2004, for example, the Sasser worm infected more than half a million computers in three days.¹⁴⁴ The Slammer worm in January 2003 infected 90% of a particular kind of Microsoft server — hindering hundreds of thousands of computers — within fifteen minutes.¹⁴⁵ The SoBig.F virus was released in August 2003 and quickly accounted for over 70% of all e-mail in the world, causing 23.2 million virus-laden e-mails to arrive on America Online’s doorstep alone.¹⁴⁶ SoBig.F receded as a threat, in part because its author designed it to expire a few weeks later.¹⁴⁷ If any of these pieces of malware¹⁴⁸ had been truly “mal” — for example, programmed to propagate indefinitely, erasing hard drives, transposing numbers randomly inside spreadsheets, or adding profanity at random intervals to Word documents — no security system would have stood in the way of major compromise. Although these examples all occurred on PCs running Microsoft Windows, the fundamental problem arises from generativity, not from a particular vendor’s security flaw. Whether running MacOS, Windows, or a flavor of UNIX, a PC that is flexible enough to have its code rewritten with the uninformed assent of its user is a PC that can be compromised.

Combine one well-written worm of the sort that can evade firewalls and antivirus software with one truly malicious worm writer, and worrisome low-level annoyances could spike to more acute effects: check-in unavailable at some airline counters; no overnight deliveries or other forms of package and letter distribution; payroll software unable to generate paychecks for millions of workers; or vital records held in

39

Government regulators, network designers, and network operators have encouraged, or at least tolerated, a generative Internet and a generative PC. They have responded with a light touch to a rising number of problems occasioned by the many possible uses of information technology, proceeding carefully with secondary liability regimes or architectural mandates in an attempt to eliminate perceived harmful activities. As these activities increase — frustrating legitimate interests both on and off the Internet — such restraint will be tested at precisely the time that the emerging PC/Internet grid is amenable to behavior-controlling interventions. Part IV thus turns to that regulation-friendly grid and how it has come about after the consummately uncontrolled growth described in Part II.

IV. A POSTDILUVIAN INTERNET

This Part describes fundamental changes to network and PC architecture that I believe are not only possible, but are also plausible, driven or enabled quite naturally by the collective, if unconcerted, pressures of the powerful regulators and soon-to-be security-conscious consumers described in Part III. Together, these changes stand to negate many of the generative features celebrated in Part II. After Part IV describes this emerging postdiluvian Internet, Part V describes what might be done about it.

A. Elements of a Postdiluvian Internet

40

There are already several Internet-aware mainstream information appliances. A flagship example is TiVo, a digital video recorder that connects to a user’s cable, satellite, or antenna feed to record television programs.¹⁴⁹ TiVo also connects to a phone line or Internet connection to download program information daily.¹⁵⁰ It is thus both a product and a service. Consumers who have TiVo nearly uniformly love it, and many say they would not want to watch television without it.¹⁵¹

The designers of TiVo did not write its software from scratch; they implemented it on top of the highly generative GNU/Linux OS.¹⁵² Most TiVo owners do not realize that they have purchased a form of PC. There is no sign that an OS — in the generative sense of something open to third-party programs — is present. Users interact with TiVo using a remote control,¹⁵³ and the limitations on TiVo’s use are not easy to circumvent.¹⁵⁴ Such limitations include the inability to create standardized digital output: TiVo’s recorded programs cannot be copied over to a PC without TiVo’s own DRM-restricted software.¹⁵⁵

TiVo works as reliably as a traditional appliance because its makers know, with much greater certainty than most PC manufacturers, the uses to which it can be put. This certainty is good for TiVo and its partner hardware manufacturers¹⁵⁶ because it gives TiVo complete control over the combined product/service that it provides. It is good for consumers because they find TiVo useful and trustworthy. It is also satisfying to regulators because TiVo was designed to prevent contributions to the peer-to-peer problem. As discussed earlier, TiVo’s recorded shows may only be saved to a standard VCR or DVD, or in a copy-protected PC format, rather than, as would be trivial for the

41

TiVo heralds growth in Internet-aware appliances that exploit the generativity of OSs and networks but that are not themselves generative to consumers or other end users. This lack of generativity does not alone bespeak a major shift in the fundamentals of the Internet/PC grid, though it does perhaps make for some lost collective potential. Consumers who might otherwise buy a television-aware PC — and who might find themselves pleasantly surprised that it can be adapted to uses that they did not contemplate at the time of purchase — instead will buy TiVos. In turn, there is less incentive for some coders to

42

Similar patterns may be found for other new information appliances. For example, smartphones are mobile cellular telephones that are designed to surf the Internet and handle nontelephonic tasks like taking pictures and maintaining calendars and to-do lists.¹⁶⁵ Some smartphones like the Palm Treo are based on general-purpose hand-held computers and then add telephonic functionality;¹⁶⁶ they can run applications from third parties, possibly to the chagrin of cellular operators whose service is subject to disruption should Treos be compromised by malicious code. Others, including some phones by Cingular, run a version of Windows,¹⁶⁷ but are configured by the cellular carriers who sell them to run only specially “signed” software: customers cannot simply double-click their way to running software not approved by the cellular carrier. Beyond smartphones, some information appliances are more closely related to the PC. For example, the Xbox is a powerful video game console produced by Microsoft. As a general-purpose device it has capacity for non-gaming applications, but, unlike a PC running Windows, it is generatively far less accessible: third-party hardware and software add-ons must be licensed by Mi-

43

2. The Appliancized PC. — The PC is heading in the direction of these information appliances. The first step that OS makers have taken as a natural response to threats from viruses and worms is to caution users before they run unfamiliar code for the first time. Users have found that they can compromise their PCs simply by visiting the wrong webpage, by clicking on executable e-mail attachments, or by downloading malicious software. Microsoft Windows presents a security warning when a user tries to run “unknown” software, defined as software without a digital certificate recognized by Microsoft.¹⁶⁹ In the most recent version of Windows (Windows XP), when a user attempts to run such an unknown program, or when one tries to execute automatically — perhaps when the user visits a webpage — the user is presented with the warning: “The publisher could not be verified. Are you sure you want to run this software?”

Users in many situations will not know how to answer this question reasonably. Unless the user was not expecting to run any software at that instant, she knows only that she wants to run the software so long as it works. How will the consumer know that it works unless she tries it? Frequently, she will assume that the computer ought to know the answer better than she does and find herself saying no to perfectly good software.

The consumer-choice solution is thus no panacea for the unintentional downloading of harmful applications. Consumers confused enough to click on a virus-laden e-mail will not likely be deterred by a warning box, especially if that warning box appears frequently when innocuous but unsigned software is run. Further, a signature alone means nothing: one could sign software with the equivalent of “Donald Duck.”¹⁷⁰ The user is simply not in the best position to determine what software is good and what software is bad. While not so effective at solving the fundamental generativity-as-vulnerability problem,

44

For those who buy their own computers and wish to operate them without a constant infusion of viruses, the next step of security, beyond a generic and ineffective warning, requires specific advice: does the OS maker think this software should run? Software publishers can readily offer an architecture through which to answer such a question. Microsoft has long maintained a digital signature program for such software as device drivers, which govern the interaction between the OS and certain external PC devices like printers, scanners, and cameras.¹⁷¹ Third-party vendors can write their own drivers and leave them unsigned, they can sign them on their own authority, or they can submit them to Microsoft for approval. In the third case, Microsoft tests the drivers and, if they pass Microsoft’s test, signs them as approved for use.¹⁷²

In some respects, this sort of testing and approval is a positive development for consumers. Too often, PC users find their machines consumed by viruses and are baffled that such infection is not covered by the PC manufacturer’s or OS vendor’s warranties. The OS vendor can address this concern by promising some form of assistance with OS problems, so long as the consumer sets the computer not to run unsigned or unapproved software. An OS maker like Microsoft can also benefit because it is uniquely positioned to offer this value-added service, one that gives it first-order gatekeeping ability over every piece of software running on its machines. As it does with an information appliance like the Xbox,¹⁷³ Microsoft could decide to charge a fee to approve third-party software were it to revisit something akin to a brief, informally stated, and quickly disavowed possibility of collecting a “vig,” or small tax, on “every transaction over the Internet that uses Microsoft’s technology.”¹⁷⁴ Alternatively, the OS maker could offer its approval for free, still benefiting as kingmaker and gaining helpful influence over the PC experiences a user is likely to have by assuring some minimum quality control. An independent software maker might chafe at having to obtain such approval, but it could always

45

To understand the full implications of these potential solutions — and why they are troubling — it helps to juxtapose this developing architecture with another new feature in OSs that is possible now that networking is so ubiquitous: automatic updating.¹⁷⁵ This new feature, which appears in the latest Microsoft and Apple OSs and in many individual pieces of software, takes account of the fact that more and more PCs have always-on broadband Internet connections. From the moment the computer is first connected to the Internet, the feature is enabled for some software. For others, including Windows XP, the feature is off, but the computer prompts the user to turn it on.¹⁷⁶ With automatic updating, the computer regularly checks — typically daily — for updates from the OS publisher and from the makers of any software installed on the PC. At first blush, this function is innocuous enough; it takes advantage of the networkability and adaptability of the PC to obtain the most recent security patches as they become available. Because it does not rely on consumer vigilance, this development solves some of the consumer maintenance problems noted as early as 1988, during the Morris worm incident, when many computer flaws went “unpatched.”

So far Apple and Microsoft install automatically only security- related updates that they deem “critical”; updates to the “regular” functioning of the OS or auxiliary applications still require the consumer’s approval. Many other makers of stand-alone software use automatic updating far more liberally, and there is no technical limit on what changes to a PC they can effect. They might not only update themselves, but also use the feature to download a small patch to other vendors’ software, to install entirely new software, to upgrade the OS, or to eliminate installed software. Thus, the security benefits of automatic updating may well fail to justify the new vulnerabilities it creates, especially for producers of more obscure pieces of software whose update servers might be more easily compromised by third parties than the “bunkerized” versions run by OS makers. But whether or not it addresses security concerns, automatic updating opens the door to an utter transformation of the way the Internet grid works.

46

Automatic updating works in concert with appliancization, allowing manufacturers to see when their software has been hacked or altered — and to shut down or reinstall the original OS when they have. Exactly this happened with the Hughes DirecTV satellite receiver information appliance. Just before the Super Bowl in 2001, consumers who had hacked their DirecTV receivers to receive smartcard access found their receivers suddenly no longer working: the satellite had broadcast not only programming for people to watch, but programming for the receiver to obey.¹⁷⁸ The receiver checked for particular hacks and, if they were found, self-destructed, rendering the affected cards entirely useless. By some reports, the last few computer bytes of the hacked smartcards were rewritten to read “Game Over.”¹⁷⁹

Automatically updating software on PCs is becoming more common at the same time as the Internet itself becomes a host for highly controlled software. The emergence of the PC/Internet grid makes it easier for applications to be developed to run on remote servers rather than on the PC itself. A PC or information appliance equipped with only a web browser can now access a range of services — a development variously known as application streaming, web services, and Web 2.0.¹⁸⁰ On one view, this development is generatively neutral, merely shifting the locus of generative software writing to a server on the Internet rather than on the PC itself — perhaps even avoiding the generative damage caused by PC lockdown. This shift, however, undermines distributed PC processing power for novel peer-to-peer ap-

47

B. Implications of a Postdiluvian Internet:
More Regulability, Less Generativity

From the regulators’ point of view, automatic updating presents new gatekeeping opportunities. Updates can be and have been used by manufacturers not only to add functionality, but also to take it away, at times apparently because of legal concerns. For example, an earlier version of Apple’s iTunes software permitted users to stream their music on the Internet, without permitting them to copy each oth-

48

Professor Randal Picker sees automatic updating as transforming the information technology landscape and suggests that regulators should indeed exploit it.¹⁸⁹ For example, in the copyright context, he believes that a software author who builds automatic updating into a product that could facilitate infringement ought to have “a duty of on-going design to reduce noninfringing use.”¹⁹⁰ For those who fail to build in automatic updating in the first instance, Professor Picker suggests a “hard use test” designed to make it legally risky to release potentially infringing software without retaining programming control.¹⁹¹ Professor Picker’s argument is straightforward enough: once it becomes easy to revise distributed products to make them less harmful (in the eyes of regulators), why not encourage such revisions? But Professor Picker fails to take into account the generative loss from compelling software originators to retain exclusive control.

A current example is illustrative: MP3 players, including the iPod, are increasingly being used for radio-like broadcasts. Through so-called “podcasting,” an owner of an MP3 player can lawfully download, for listening purposes, a number of selected programs from

49

Worse, Professor Picker’s solution would be difficult to apply to group-written open-source software.¹⁹³ A regulatory judgment in favor of software as service would, if not carefully crafted, punish decentralized development processes, which in turn might reduce or eliminate entire categories of information technology innovation.

Furthermore, the logic of Professor Picker’s argument for imposing gatekeeping responsibilities need not stop at a software author’s own products. Consider the consequences if OS makers were held responsible for all applications running on their systems. For example, DeCSS, a program that decrypts DVDs, has been found to be an illegal circumvention tool under section 1201 of the DMCA.¹⁹⁴ Under threat of liability or direct government regulation, it would take little technical effort for Microsoft, using exactly the technologies that antivirus vendors use to screen for and eliminate malware, to send an update to its OS customers that would prevent DeCSS from running. Indeed, any vendor of antivirus software with automatically updating definitions could be so pressured.

To be sure, the potential for new uses does not always militate against trying to eliminate real infringements. But the advent of software as service shifts generative power away from the broad audiences at the edges of information technology deployment and toward the center — toward professional software authors constrained by vendors, regulators, and private litigants who can influence them.¹⁹⁵ This shift will be unfortunate if it tends to frustrate the vast creativity and energy of “edge” contributors of the sort described in Part II.¹⁹⁶

50

The “third strand” cyberlaw scholars, who first raised concerns about an era of trusted systems in the mid-1990s, foresaw software that would refuse to abide by a user’s desires if those desires exceeded the permissions built into the technology. Such software exists, but as Part III explains, its controls are structurally weak when implemented on generative PCs. So long as the user can run unrestricted software and can use an open network to obtain cracked copies of the locked-down content, trusted systems provide thin protection. For instance, Microsoft’s Windows Media Player contains a powerful rights management system, but it will still play content stripped of protective tags. Even if it should cease to do so — instead playing only certified content — users could still install third-party players that ignore such restrictions.

United States lawmakers recognized this loophole in 2002, proposing the Consumer Broadband and Digital Television Promotion Act¹⁹⁷ (CBDTPA). The Act would have set in motion a process by which technology makers would, among themselves, develop standards for flagging digital content restrictions and then be compelled to write software that would respect those restrictions.¹⁹⁸ It would apply to any “digital media device,” meaning: [A]ny hardware or software that — (A) reproduces copyrighted works in digital form; (B) converts copyrighted works in digital form into a form whereby the images and sounds are visible or audible; or (C) retrieves or accesses copyrighted works in digital form and transfers or makes available for transfer such works to hardware or software described in subparagraph (B).¹⁹⁹ If technology makers could not agree on a standard, the FCC was to set one for them.²⁰⁰

51

The security- and market-driven phenomena that this Part describes point to a technology configuration that already accomplishes many of the goals of the CBDTPA. Unlike the CBDTPA, these phenomena are not fiats to be grafted onto a vibrant, resistant marketplace. They are the marketplace, representing a sum across the technology and publishing industries, governments, and consumers. In essence, they point to a license to code that is issued by mainstream software makers but can be shaped by governments.

Such a license may not be hard to obtain. Like a driver’s license or a cosmetologist’s license that vouches for the basic training and identity of the holder, a license to code would exist, at first, simply to require a verifiable identity behind the production of software. It could be held by a software author to indicate permission to make new works and could also apply to each item of software itself so that regulators could revoke the license of individual works. Further, the creator could be identified easily and held accountable for creating a virus or for negligently allowing his or her identity to be stolen.

New software writers might find users skeptical of running their software at first. Like new traders on the auction site eBay, new producers would have to prove their worth among daring consumers, slowly building a reputation for trustworthiness. Alternatively, new producers could submit the software for testing, such as the sort of testing that Microsoft requires for those who want their device drivers to be signed or that some smartphone makers require before allowing programs to run on their devices.

Should approved software later turn out to enable undesirable behavior, a government could demand that an OS maker treat it as a virus and revoke its license — or even the license of its author. Different governments might make different judgments about the software and thus could ask OS makers to block the offending software only on PCs in their respective jurisdictions. In a country like China, a movement

52

So far, this Part has discussed reasons why consumers, OS makers, and regulators might appreciate a world in which the PC is more locked down in certain ways — ways that strike at the heart of its generativity. To be sure, there is no basis on which to insist flatly that any tradeoff between regulability and generativity should favor the latter. But this is a false dichotomy if we can make the grid more secure without sacrificing its essential generative characteristics. Making progress on the security problem is difficult because the distributed nature of the Internet and individual ownership of PCs do not induce participants to internalize their negative security externalities. As Part V discusses, ISPs are not held economically accountable when their subscribers’ computers fall victim to viruses. Similarly, individual users may not care if their compromised machines cause trouble for other, faraway users. Locking down the PC, although attractive from a regulatory point of view, is undesirable because of its effect on innovation: technical innovation will slow as third parties are squeezed out of the development cycle, and intellectual and artistic innovation will slow as some of the technical innovations forestalled are quite possibly ones that would enhance valuable expressive activities.

The existence of a widely distributed, multipurpose PC inviting third-party contribution permits innovation arbitrage, whereby firms that are slow to invent or improve information appliances find themselves leapfrogged by entrepreneurial PC software programmers with less invested in the status quo. For example, the free Internet telephony services offered by such PC applications as Skype and audio-enabled instant messenger programs serve as benchmarks for appliancized telephone services that connect to the Internet through wi-fi

53

V. HOW TO TEMPER A POSTDILUVIAN INTERNET

The emergence of broadband has meant that these applications can run either locally or remotely and still be instantly accessible. Part III of this Article describes how disruptive uses of Internet and PC generativity, in particular cybersecurity problems, stand to raise alarm and ignite a significant reaction. Part IV sketches how, paradoxically, the

54

The worst aspects of this future ought to be blunted even though they will benefit important stakeholders: The postdiluvian Internet creates new opportunities for regulators. It accords with, or is identical to, the plans of many technology industry firms. In addition, it is genuinely premised on new conveniences for consumers, allowing their desired activities to take place more reliably, though at the expense of unquantifiable future opportunities.

To evaluate the different paths information technology might take, we must bear in mind key contributors to its success: those who are creative and are inspired to express that creativity, whether through producing new code or code-enabled art. Amateurs, who produced applications that others overlooked, played a vital role in the rise of the Internet and the PC that Part II chronicles. In this sense, of course, “amateurs” are those who do what they do because they love to do it. The availability of tools through which amateurs could express creativity meant that code was written by parties other than those who had chosen lives of professional codewriting. Today, thanks to networked information technology and the recursively generative code produced in large part by amateurs, art can be produced and shared by people other than professional artists, citizens can engage in far-ranging dialogues with others whom they would not otherwise encounter, and people can work together from the four corners of the globe to produce intellectual projects of social and economic significance.²⁰⁵

The most important opportunities for such creativity ought to be retained as the Internet evolves. But this will require those who support creative communities to make an important concession. They will have to face the reality that a free and open Internet, including open PCs distributed among tens of millions of consumers, is simply not possible unless the most pressing demands of countervailing regulatory forces are satisfied. It is now an opportune time for thoughtful

55

This Part sketches ways in which current thinking by cyberlaw scholars on these issues is perhaps too constrained. It also describes some specific projects that could help solve some of the Internet’s most pressing problems with as little constriction of its generative capacities as possible.

A. Refining Principles of Internet Design and Governance

56

According to end-to-end theory, placing control and intelligence at the edges of a network maximizes network flexibility and user choice.²⁰⁹ The political implication of this view — that end-to-end design preserves user freedom — depends on an increasingly unreliable presumption: whoever runs a machine at a given network endpoint can readily choose how the machine will work. For example, in response to a network teeming with viruses and spam, network engineers suggest more bandwidth (to make invisible the transmission of “deadweights” like viruses and spam) and better protection at user endpoints, rather than interventions by ISPs closer to the middle of the network.²¹⁰ But consumers are not well positioned to maintain their machines painstakingly against attack, leading them to prefer the locked-down PCs described in Part IV. Those who favor end-to-end principles because they favor generativity must realize that failure to take action at the network level may close some parts of the grid because consumers may demand, and PC manufacturers may provide, locked-down endpoint environments that promise security and stability with minimum user upkeep. Some may embrace a categorical end-to-end approach anyway: even in a world of locked-down PCs, there will no doubt remain non-mainstream generative computing platforms for professional technical audiences. But this view is too narrow. We ought to see the possibilities and benefits of PC generativity made available to everyone, including the millions of people who obtain PCs for current rather than future uses, but who end up delighted at the new uses to which they can put their machines.

Put simply, complete fidelity to end-to-end may cause users to embrace the digital equivalent of gated communities. Gated communities offer safety and stability to residents and a manager to complain to when something goes wrong. But from a generative standpoint, digital gated communities are prisons. Their confinement is less than obvious because what they block is generative possibility: the ability of outsiders to offer code and services to users, giving users and producers an opportunity to influence the future without a regulator’s permission.

57

Thus, strict loyalty to end-to-end neutrality should give way to a new generativity principle, a rule that asks that modifications to the PC/Internet grid be made when they will do the least harm to its generative possibilities. Under such a principle, for example, it may be preferable in the medium term to screen out viruses through ISP-operated network gateways rather than through constantly updated PCs.²¹¹ Although such network screening theoretically opens the door to additional filtering that may be undesirable, this risk should be balanced against the very real risks to generativity inherent in PCs operated as services rather than products.

This generativity principle suggests at least two ways in which we might fundamentally reconceptualize the map of cyberspace. First, we must bridge the divide between those concerned with network connectivity and protocols, on the one hand, and those concerned with PC design, on the other — a divide that end-to-end unfortunately encourages. Such modularity in stakeholder competence and purview was originally a useful and natural extension of the Internet’s hourglass architecture: it meant that network experts did not have to be PC experts and vice versa, just as the OS-application divide in the corresponding PC hourglass means that application developers need not know the ins and outs of PC peripherals and networking. But this division of responsibilities, which works so well for technical design, is crippling our ability to think through the trajectory of applied information technology. Now that the PC and the Internet are so inextricably intertwined, it is not enough for network engineers to worry only about network openness and assume that the endpoints can take care of themselves. It is abundantly clear that endpoints simply cannot.

Second, “middle” and “endpoint” are no longer subtle enough to capture the important features of the Internet/PC landscape. It remains correct that from a network standpoint, protocol designs and the ISPs that implement them are the “middle” of the network, as distinct from PC “endpoints.” But the true import of a vernacular of “middle” and “endpoint” for policy purposes relates to individuals’ power to control their experiences on the network. If consumers can no longer exercise meaningful control over their PC endpoints, instead ceding such control to government or corporate authority such as an OS maker or a handful of security vendors, then the PCs become driven by a “middle” and their identities as endpoints diminish. Even today, consumers might not want or have the ability to fine-tune their PCs, and the taxonomy of generativity would say that such fine-tuning is not possible because the PCs are not easy for a mass audience to master even though they remain leveraged and adaptable. But there are a variety of methods by which PCs can compensate for the difficulty of mastery, only some of which require centralized control. For example, users might be able to choose from an array of proxies — perhaps Microsoft, Ralph Nader, or a public interest organization — for guidance on decisions about PC configuration. Now that the network’s endpoints are controllable by faraway entities, abandoning the end-to-end debate’s simplistic divide between middle and endpoint will enable us to identify and respond better to the emerging threats to the Internet’s generativity.

2. Reframing the Internet Governance Debate. — Since the mid-1990s, an intense but misguided debate has raged over Internet governance. Those who care about the Internet’s future are unduly occupied with domain names and the Internet Corporation for Assigned Names and Numbers (ICANN), the nonprofit organization chartered to administer top-level domain name policy. The existence of ICANN has proved a focus for debates about Internet governance for the circular reason that ICANN is an organization administering a particular — and quite limited — part of Internet functionality. The issues at stake in domain name assignment are real, but the focus on such disputes and on the governance of ICANN is myopic in relation to the threats posed by undue restriction of the Internet’s generativity.

At the other extreme is the overbroad focus of the World Summit on the Information Society, a United Nations project that took place from 2003 to 2005.²¹² The World Summit included narrow debates about domain name management, but it also covered the larger dilemma of the global digital divide: how to ensure Internet access to as many people and peoples as possible.²¹³ This cause is truly worthy,

58

Just as Internet architects should be encouraged to apply network design principles to PCs as PCs develop, groups engaged with issues of Internet governance should take note of the ways in which the Internet/PC grid is developing and grapple directly with how to maintain generativity. To proceed assuming that the primary challenge for Internet governance is narrowly one of managing administrative functions or is broadly one of deploying the network to additional people is to overlook the most important questions facing the Internet’s future.

Part IV argues that OS makers or security firms may block the deployment of individual PC applications on behalf of PC users who crave security, creating broader bottlenecks to application deployment by anyone other than centralized kingmakers. The puzzle, then, is how to avoid these bottlenecks, whether coming from government or from private code-filtering schemes, while conceding that PC users can no longer be expected to exercise meaningful choice about code without help. A worthy Internet governance project to retain consumer choice without creating a new bottleneck could take the form of a grassroots campaign or public interest organization with participation from Internet architects. This project could set up a technical architecture to label applications and fragments of executable code, coupled with an organization to apply such labels nondiscriminatorily. Alternatively, the project could establish a distributed architecture by which the individual decisions about whether to run a given application, and the subsequent results, could serve as advice to others contemplating whether to run such code. The history of the Internet is seasoned with successful organizations devoted to such ends even though ideological views and regulatory agendas are often embedded in technical decisions.²¹⁴ Public interest “underwriters’ laboratories” for the Internet would reduce consumer demand for evaluations by OS makers or ISPs. Precisely because the lines separating viruses, spyware, poorly written software, and flat rent extraction by software authors are so blurry, they are best adjudicated using the sorts of quasi-public mechanisms that have served Internet development in the past. The alternative is to see such power accrete in a handful of private firms with incentives to become gatekeepers for purposes other than security.

60

3. Recognizing Interests in Tension with Generativity. — Those who have made the broad case for Internet freedom — who believe that nearly any form of control should be resisted — ought to be prepared to make concessions. Not only are many of the interests that greater control seeks to protect indeed legitimate, but an Internet and PCs entirely open to new and potentially dangerous applications at the click of a mouse are also simply not suited to widespread consumer use. If the inevitable reaction to such threats is to be stopped, its underlying policy concerns must in part be met.

There was a time in the Internet’s development during which it made sense to eschew modalities of control because the network was experimental and because the harm that could be brought about by its misuse was limited. The network, after all, only carried bits. One might compare it to a hypothetical highly generative children’s chemistry set, which would be adaptable and could be leveraged: it would contain chemicals that could accomplish a variety of tasks, with small

61

At an abstract level, then, one might ask: for a given technology, how much generativity can be maintained in the face of concessions to its threats? Knowing only that one technology is in the field of computer or information science and another is in the field of, say, physics, might be enough to imply that there ought to be room for more generativity for the former than for the latter because the risks of harm — particularly physical harm — from misuse or abuse of the former are structurally likely to be lower. The worst examples of the harm caused by wayward uses of applications and networks — including uses for invasion of privacy, financial scams, defamation, and copyright infringement — are less physical than that which could be caused by wayward uses of fertilizer, explosives, and petri dish cultures.²¹⁷ With generative technologies in the physical sciences, on the one hand, the good applications may be much more identifiable ex ante and the harmful ones more diffuse, suggesting there is less to be gained from maintaining universal generativity. With the Internet, on the other hand, truly harmful applications are the known exceptions to be carved out from the universe of potential applications the technology enables. At the very least, then, if individual harms can be largely identified and then prevented or rectified by narrow interventions, these interventions would be preferable to undermining the generative nature of the PC and the Internet.

Now that the Internet is no longer experimental, an intuition that “it’s only bits” is not so compelling. The degree of openness suited to, say, 1988 — when an inculcation of “professional ethics” among Internet users was thought to be the appropriate way to deal with virus writers²¹⁸ — does not suit the present day.

Some of the parties best suited to solve the problems of viruses, privacy invasion, spam, and copyright infringement may be sitting on

62

4. “Dual Machine” Solutions. — Part IV suggests that consumers, rightly fearful of security vulnerabilities latent in the generative Internet/PC grid, will demand a future in which locked-down information appliances predominate over generative PCs. One may seek the best of both worlds, however, by creating both generativity and security within a single device. To accomplish this compromise, we might build PCs with physical switches on the keyboard — switching be-tween “red” and “green.”²¹⁹ A PC switched to red mode would be akin to today’s PCs: it would be capable of running any software it encountered. This mode would maximize user choice, allowing participation in unanticipated applications, such as PC-to-PC telephony, whose value in part depends on uptake by other users. Such a configuration would retain a structural vulnerability to worms and viruses, however. Hence the availability of green mode, by which the computer’s processor would be directed to a different OS and different data within the same hardware. In green mode, the computer might run only approved or vetted software — less interesting, but much more reliable. The consumer could then switch between the two modes, attempting to ensure that valuable or sensitive data is created and stored in green mode and leaving red mode for experimentation. A crude division such as this has the benefit of being eminently understandable to the consumer — just as a driver can understand putting a sport utility vehicle into all-wheel drive for off-roading — while retaining much of the leverage and adaptability of today’s PC.

But such PCs give rise to new problems. For example, ISPs might offer a lower rate for connecting a green PC and a higher rate for a red one — presuming the green to be less burdensome for customer service and less amenable to network abuse. Corporate environments might offer only green PCs and thus limit the audience for available innovation. Or the green PC might be so restrictively conceived that most users would find it unpalatable and would thus continue to choose between traditional PCs and vendor-specific information appliances. Even to hypothesize a green PC is to ask that some way be found to determine which software is suitable for use on an open PC and which is not. A PC running multiple virtual machines²²⁰ is a promising ave-

63

B. Policy Interventions

One way to reduce pressure on institutional and technological gatekeepers is to make direct responsibility more feasible. Forthcoming piecemeal solutions to problems such as spam take this approach. ISPs are working with makers of major PC e-mail applications to provide for forms of sender authentication.²²¹ A given domain can, using public key encryption tools, authenticate that it is indeed the source of e-mail attributed to it. With Microsoft’s Sender ID or something like it, e-mail purporting — but not proving — to be from a user at yahoo.com can be filtered as spam so easily that it will no longer be worthwhile to send. This regime will hold ISPs more accountable for the e-mail they permit their networks to originate because they will find themselves shunned by other ISPs if they permit excessive anonymous spam. This opportunity for more direct liability reduces the pressure on those processing incoming e-mail — both the designated recipients and their ISPs — to resort to spam filtration heuristics that may unintentionally block legitimate e-mail.²²²

The same principle can be applied to individuals’ uses of the Internet that are said to harm legally protected interests. From the point of view of generativity, music industry lawsuits against individual file sharers inflict little damage on the network and the PC themselves, even if they are bad policy because the underlying substantive law demarcating the protected interest is itself ill-advised — as I believe it is. The Internet’s future may be brighter if technical processes are re-

64

As discussed in section II.C, many Internet consumers have embraced wi-fi, and the wireless routers default to sharing the connection with anyone nearby who has a PC configured with a wireless antenna. Consumers may not intend to open their networks, but they carry generative benefits for those nearby without their own Internet access.²²³ Usage by others does not typically impede the original consumer’s enjoyment of broadband, but should outsiders use that connection, say, to send viruses or to pirate copyrighted files, the original consumer could be blamed when the Internet connection is traced.²²⁴ As such examples arise and become well-known, consumers will seek to cut off others’ access to their surplus network resources, and the manufacturers of wireless routers will change the default to “closed.” If, however, genuine individual identity can be confirmed in appropriate circumstances, wi-fi sharing need not be impeded: each user will be held responsible for his or her own actions and no more. But insofar as technologically guaranteed anonymity is retained, more drastic means to eliminate individual wrongdoing through gatekeeper intervention wait in the wings.

2. Establishing a Choice Between Generativity and Responsibility.

— To the extent that those who use generative platforms to invade legally protected interests can be held directly liable, maintainers of technology platforms — ISPs and newly service-like OS makers — should be encouraged to keep their platforms generative, rather than narrowing their offerings to facilitate regulatory control as Professor Picker suggests.²²⁵

In turn, the more service-oriented and less generative the platform, the more legal responsibility we should impose on the technology provider to guarantee a functioning system. If a TiVo unit were not to operate as promised — suppose it simply crashed and failed to record any television programs — the law of warranty would quickly come into play. If the TiVo unit were new enough, the company would make good on a repair or replacement. Yet this simple exchange rarely takes place after the purchase of a new computer. Suppose a new PC stops functioning: after a week of using it to surf the Internet and write articles, the consumer turns it on and sees only a blue error

65

Hardware and OS makers are right that they ought to bear very little responsibility for this all-too-common problem because it is not clear that either the hardware or the OS is at fault. The mishmash of software found on even a two-week-old Internet-exposed PC thus precludes any obvious responsibility of a particular hardware or software manufacturer when problems arise.

Part IV argues that TiVo and the PC are converging. To the extent that PC OSs seek to screen what programs can run on them, the law should hold OS makers responsible for problems that arise just as TiVo and cellular phone manufacturers take responsibility for issues that arise with their controlled technologies. If the OS remains open to new applications by third parties, the maker’s responsibility should be duly lessened. Such a regime permits technology vendors to produce closed platforms but encourages them to produce generative platforms by scaling liabilities accordingly. This tracks the intuition behind secondary theories of liability: technology makers may shape their technologies largely as they please, but the configurations they choose then inform their duties and liabilities.

This Part sketches a modest route by which the operation of law might appropriately further generativity. There are no doubt others, such as shaping consumer protection law to ensure that a shift from product to service does not permit a technology vendor to upset settled consumer expectations through a supposedly routine automatic product update that in fact substantially changes the benefit of the consumer’s bargain. Each of these kinds of interventions is grounded in recognition that the law already influences the development of technology in manifold ways and thus can be adjusted to take generativity into account as an important end among other policy priorities.

VI. CONCLUSION

66

The challenge facing those interested in a vibrant global Internet is to maintain the core of that experimentalist spirit in the face of growing pressures. One path leads to a velvet divorce, creating two separate Internets with distinct audiences: a return to the quiet backwater for the original experimentalist Internet that would restart the generative cycle among researchers and hackers distinct from consumers who live with a new, controlled Internet experience. Two Internets would consign the existing grid to an appliancized fate, in which little new happens as existing technology players incrementally modify existing applications without the competitive pressure of grid-enabled innovation arbitrage.

The alternative paths that this Article advocates try to maintain the fundamental generativity of the existing grid while taking seriously the problems that fuel enemies of the Internet free-for-all. It requires charting an intermediate course to make the grid more secure — and to make some activities to which regulators object more regulable — in order to continue to enable the rapid deployment of the sort of amateur programming that has made the Internet such a stunning success.

Crippling generative accessibility and adaptability by transforming the PC into an information appliance is undesirable. So, too, are ham-fisted clamps by ISPs upon network traffic in an effort to beat back viruses and other PC security threats, even as complete fidelity to end-to-end neutrality may on balance harm the generative information technology environment. Some limits are inevitable, and this Article attempts to point to ways in which these limits might be most judiciously applied. The key is to set such limits through thoughtful adjustments to accessibility that do not themselves spawn new centralized gatekeepers. The right interventions will preserve the public’s ability to adopt new technologies from all corners, creating rough accountability for those who wish to introduce software to the world and for individuals who put that software to certain uses, while enabling those who maintain generative technologies — the Internet architects, ISPs, and OS publishers — to keep those technologies open and to ensure that those who wish to contribute to the global information grid can do so without having to occupy the privileged perches of established firms or powerful governments.

67

¹⁰⁰ 125 S. Ct. 2764 (2005). I explore in detail the history and trajectory of regulatory interventions to prevent defamation and copyright infringement in Zittrain, supra note 97.

¹⁰¹ See OpenNet Initiative, Documenting Internet Content Filtering Worldwide, http://www. opennet.net/modules.php?op=modload&name=Archive&file=index&req=viewarticle&artid=1 (last visited Apr. 9, 2006); Jonathan Zittrain & Benjamin Edelman, Documentation of Internet Filtering Worldwide, http://cyber.law.harvard.edu/filtering (last visited Apr. 9, 2006).

¹⁰² Bob Sullivan, Remembering the Net Crash of ’88, MSNBC, Nov. 2, 1998, http://www. msnbc.com/news/209745.asp?cp1=1.

¹⁰³ See id. (noting that only 60,000 nodes were connected to the Internet and that Morris’s software affected primarily universities and research centers); see also PATRICIA WALLACE, THE INTERNET IN THE WORKPLACE 35–37 (2004) (discussing changes in computer configuration in the workplace and contrasting PCs with a mainframe setup).

¹⁰⁴ See JOYCE K. REYNOLDS, THE HELMINTHIASIS OF THE INTERNET 1–2 (1989), available at http://www.faqs.org/ftp/rfc/pdf/rfc1135.txt.pdf; Sullivan, supra note 102. For more on how worms behave, see Eugene H. Spafford, Crisis and Aftermath, 32 COMM. ACM 678, 678–84 (1989).

¹⁰⁵ See U.S. GEN. ACCOUNTING OFFICE, GAO/IMTEC-89-57, COMPUTER SECURITY: VIRUS HIGHLIGHTS NEED FOR IMPROVED INTERNET MANAGEMENT 8 & n.1, 13–14 (1989) [hereinafter GAO REPORT], available at ftp://coast.cs.purdue.edu/pub/doc/morris_worm/GAO-rpt.txt; Sullivan, supra note 102.

¹⁰⁶ See REYNOLDS, supra note 104, at 1; John Markoff, Computer Invasion: ‘Back Door’ Ajar, N.Y. TIMES, Nov. 7, 1988, at B10.

¹⁰⁷ See GAO REPORT, supra note 105, at 8, 17; Markoff, supra note 106.

¹⁰⁸ REYNOLDS, supra note 104, at 3–4.

¹⁰⁹ Id.; Sullivan, supra note 102.

¹¹⁰ Spafford, supra note 104, at 678; Sullivan, supra note 102.

¹¹¹ United States v. Morris, 928 F.2d 504, 505–06 (2d Cir. 1991). For a general discussion of United States v. Morris, see Susan M. Mello, Comment, Administering the Antidote to Computer Viruses: A Comment on United States v. Morris, 19 RUTGERS COMPUTER & TECH. L.J. 259 (1993).

¹¹² See WALLACE, supra note 103, at 36.

¹¹³ Morris reportedly considered writing a “worm killer” to remove worms from computers once he realized the damage that the worm was causing. He did not, however, because he felt that he had already caused enough damage. See Brief for Appellant at 18–19, Morris (No. 90-1336), 1990 WL 10029997.

¹¹⁴ See Spafford, supra note 104, at 678–81.

¹¹⁵ See id. at 680.

¹¹⁶ See GAO REPORT, supra note 105, at 28.

¹¹⁷ See REYNOLDS, supra note 104, at 2–3.

¹¹⁸ See GAO REPORT, supra note 105, at 19–21 (noting that “each host site is responsible for establishing security measures adequate to meet its needs”).

¹¹⁹ See REYNOLDS, supra note 104.

¹²⁰ See id. at 5–8.

¹²¹ On the origin of CERT, see HOWARD F. LIPSON, CERT COORDINATION CTR., SPECIAL REPORT CMU/SEI-2002-SR-009, TRACKING AND TRACING CYBER-ATTACKS: TECHNICAL CHALLENGES AND GLOBAL POLICY ISSUES 5 (2002), available at http://www.cert.org/archive/pdf/02sr009.pdf; and CERT, Frequently Asked Questions, http://www.cert.org/faq/ cert_faq.html (last visited Apr. 9, 2006).

¹²² See supra section II.C, pp. 1987–94.

¹²³ See Harmon, supra note 53 (describing CompuServe as “an information service”).

¹²⁴ See Ron Rosenbaum, Secrets of the Little Blue Box, ESQUIRE, Oct. 1971, at 116, 119. For an account about the individual who claimed to be the original “phone phreaker,” see James Daly, John Draper, FORBES, June 3, 1996, at 138.

¹²⁵ See Rosenbaum, supra note 124, at 120.

¹²⁶ See id. at 119–20.

¹²⁷ See Amy Harmon, Defining the Ethics of Hacking, L.A. TIMES, Aug. 12, 1994, at A1.

¹²⁸ See COMING OF AGE, supra note 22, at 107–24 (describing the Internet as “a set of independent networks interlinked to provide the appearance of a single, uniformed network” and explaining the architecture of the Internet).

¹²⁹ Increases in computer crime have received attention from the hacker community and have influenced hackers’ behavior. See Harmon, supra note 127 (chronicling one self-described hacker’s efforts to “stay[] on the right side of the blurry line that separates hacking from criminal behavior”).

¹³⁰ See Steve Lohr, A Virus Got You Down? Who You Gonna Call?, N.Y. TIMES, Aug. 12, 1996, at D1 (stating that most viruses are not deliberately destructive).

¹³¹ Estimates of the cost of the Morris attack vary widely. Compare Kevin Commins, Insurers Plan Computer Virus Coverage, J. COM. & COM., June 8, 1989, at 1A ($1.2 billion), with Panel Speculates on Rogue Hacker’s Motives, ST. LOUIS POST-DISPATCH, Apr. 2, 1989, at 10A ($96 million). The estimates likely vary because they are based on relatively soft variables such as estimates of staff time, the number of computers affected, and productivity loss.

¹³² For an analysis of hackers, see CLIFFORD STOLL, THE CUCKOO’S EGG: TRACKING A SPY THROUGH THE MAZE OF COMPUTER ESPIONAGE (1989). The “hacker ethic” is commonly defined by “the belief that system-cracking for fun and exploration is ethically acceptable as long as the hacker commits no theft, vandalism, or breach of confidentiality.” Wikipedia, Hacker Ethic, http://en.wikipedia.org/wiki/Hacker_ethic (last visited Apr. 9, 2005); see also PEKKA HIMANEN, THE HACKER ETHIC AND THE SPIRIT OF THE INFORMATION AGE (2001) (providing an in-depth exploration of the hacker ethic).

¹³³ In fact, more U.S. households now connect to the Internet using broadband than using dial-up. See Press Release, FCC, FCC Report Shows Strongest Ever American Broadband Market 1 (Sept. 9, 2004), available at http://hraunfoss.fcc.gov/edocs_public/attachmatch/DOC-251959A2.pdf.

¹³⁴ See LUKE DUDNEY, SANS INST., INTERNET SERVICE PROVIDERS: THE LITTLE MAN’S FIREWALL (2004), available at http://www.sans.org/rr/whitepapers/casestudies/1340.php (discussing port blocking, packet blocking, and other methods that ISPs could employ to prevent the spread of computer viruses, and running an experiment to assess the extent of the “open proxy” problem).

¹³⁵ Id. at 4.

¹³⁶ Id. at 5.

¹³⁷ See id. at 4.

¹³⁸ See CERT Coordination Center, CERT/CC Statistics 1988–2005, http://www.cert.org/ stats#incidents (last visited Apr. 9, 2006).

¹³⁹ See id. Other studies have noted the exploding number of incidents of application attacks as a threat, as websites increasingly link webpages to company databases. See, e.g., BEE WARE SAS, THE RISK OF APPLICATION ATTACKS SECURING WEB APPLICATIONS 1–2 (2005), available at http://www.securitydocs.com/pdf/2839.PDF.

¹⁴⁰ See Sullivan, supra note 102; Internet Sys. Consortium, ISC Domain Survey: Number of Internet Hosts, http://www.isc.org/index.pl?/ops/ds/host-count-history.php (last visited Apr. 9, 2006).

¹⁴¹ See Internet Sys. Consortium, supra note 140.

¹⁴² See Press Release, Computer Indus. Almanac Inc., Mobile PCs In-Use Surpass 200M (June 20, 2005), available at http://www.c-i-a.com/pr0605.htm.

¹⁴³ See, e.g., Associated Press, A New Computer Virus Is Making the Rounds, N.Y. TIMES, June 20, 2000, at C6 (describing the Love Bug virus that struck in May 2000).

¹⁴⁴ See Worm Brings Down PC’s and Networks, N.Y. TIMES, May 4, 2004, at C10.

¹⁴⁵ John Schwartz, Rampant Epidemics of Powerful Malicious Software, N.Y. TIMES, Dec. 1, 2003, at C19; John Schwartz, Worm Hits Microsoft, Which Ignored Own Advice, N.Y. TIMES, Jan. 28, 2003, at C4.

¹⁴⁶ Brendan I. Koerner, In Computer Security, a Bigger Reason To Squirm, N.Y. TIMES, Sept. 7, 2003, § 3, at 4; Sobig Is Biggest Virus of All, BBC NEWS, Aug. 21, 2003, http://news.bbc.co.uk/ 2/hi/technology/3169573.stm.

¹⁴⁷ Amy Harmon, As Digital Vandals Disrupt the Internet, A Call for Oversight, N.Y. TIMES, Sept. 1, 2003, at A1; Koerner, supra note 146.

¹⁴⁸ See generally Wikipedia, Malware, http://en.wikipedia.org/wiki/Malware (last visited Apr. 9, 2006) (defining malware as “software designed to infiltrate or damage a computer system”).

¹⁴⁹ See TiVo, What Is TiVo?, http://www.tivo.com/1.0.asp (last visited Apr. 9, 2006).

¹⁵⁰ See Wikipedia, TiVo, http://en.wikipedia.org/wiki/TiVo (last visited Apr. 9, 2006).

¹⁵¹ See, e.g., Eric A. Taub, How Do I Love Thee, TiVo?, N.Y. TIMES, Mar. 18, 2004, at G1.

¹⁵² See Peter Wayner, Whose Intellectual Property Is It Anyway? The Open Source War, N.Y. TIMES, Aug. 24, 2000, at G8; see also TiVo, TiVo — GNU/Linux Source Code, http:// www.tivo.com/linux/linux.asp (last visited Apr. 9, 2006) (making TiVo’s source code publicly available).

¹⁵³ See Katie Hafner, Now Preening on the Coffee Table, N.Y. TIMES, Feb. 19, 2004, at G1.

¹⁵⁴ See, e.g., Paul Festa, TiVo Hacks Flourish, CNET NEWS.COM, Nov. 11, 2004, http:// news.com.com/TiVo+hacks+flourish/2100-1041_3-5447461.html.

¹⁵⁵ TiVoToGo allows users to transfer recordings from their TiVos very slowly in a copy-protected format to their own PCs and to copy them to PC DVDs. See David Pogue, TiVo Adds Portability to the Mix, N.Y. TIMES, Jan. 6, 2005, at G1.

¹⁵⁶ Philips, Sony, and others make TiVo-compliant boxes under licenses from TiVo. See Laurie J. Flynn, Networks See Threat in New Video Recorder, N.Y. TIMES, Nov. 5, 2001, at C4.

¹⁵⁷ TiVo can, however, share its programming with other TiVos in a household. See Katie Dean, TiVo Breaks into Home Networks, WIRED NEWS, June 10, 2004, http://www.wired.com/ news/digiwood/0,1412,63776,00.html.

¹⁵⁸ See, e.g., TiVo Inc., Definitive Proxy Statement (Form DEF 14A), at 7, 13–14 (May 31, 2005) (noting that NBC is entitled to nominate one of TiVo’s directors and indicating that subsidiaries of NBC own more than four percent of TiVo’s stock).

¹⁵⁹ See Digital Broad. Content Prot., Report and Order and Further Notice of Proposed Rule-making, 30 Commc’ns Reg. (P & F) FCC 03-273, at 1189, 1194–97 (Nov. 4, 2003); see also Press Release, FCC, FCC Adopts Anti-Piracy Protection for Digital TV (Nov. 4, 2003), available at http://hraunfoss.fcc.gov/edocs_public/attachmatch/DOC-240759A1.pdf.

¹⁶⁰ However, TiVo users have discovered a clandestine thirty-second skip feature embedded in their TiVos. See O’Reilly, The 30-Second Skip, http://hacks.oreilly.com/pub/h/491 (last visited Apr. 9, 2006).

¹⁶¹ See Paramount Pictures Corp. v. Replay TV, 298 F. Supp. 2d 921, 923 (C.D. Cal. 2004).

¹⁶² See id.

¹⁶³ See Dan Tynan, Winners and Losers 2005, PC WORLD, Dec. 27, 2005, http://www.pcworld. com/news/article/0,aid,123923,00.asp (describing TiVo’s compatibility with a DRM system provided by Macrovision).

¹⁶⁴ See TiVo Home Media Engine Software Development Kit, http://tivohme.sourceforge.net (last visited Apr. 9, 2006).

¹⁶⁵ See Simon Romero, Wireless Wanderer: A Field Study, N.Y. TIMES, May 24, 2001, at G1.

¹⁶⁶ See Press Release, palmOne, Inc., Treo 600 from palmOne Now Available for Cingular Wireless GSM/GPRS Customers (Nov. 13, 2003), available at http://www.palm.com/us/company/ pr/2003/111303.html; Update: palmOne Finally Takes Wraps Off Treo 650, SMARTPHONETODAY, Oct. 25, 2004, http://www.smartphonetoday.com/articles/2004/10/2004-10-25-palmOne-Finally-Takes.html.

¹⁶⁷ See, e.g., Thomas J. Fitzgerald, Music for Your Cellphone, N.Y. TIMES, July 7, 2005, at C9; Windows Mobile, http://www.microsoft.com/windowsmobile/devices/default.mspx (last visited Apr. 9, 2006).

¹⁶⁸ See Ina Fried, Digging Profits Out of Xbox, CNET NEWS.COM, Aug. 10, 2005, http:// news.com.com/Digging+profits+out+of+Xbox/2100-1043_3-5827110.html (describing the Xbox licensing scheme); Wikipedia, Xbox 360, http://en.wikipedia.org/wiki/Xbox_360#Dashboard (last visited Apr. 9, 2006) (describing non-gaming applications). This licensing structure may exist in part because the Xbox hardware is sold at a loss. See Fried, supra.

¹⁶⁹ Microsoft recognizes a variety of certificate authorities such as VeriSign, which in turn can accept signatures from authors of code. See, e.g., Press Release, Microsoft Corp., Microsoft and VeriSign Announce .NET Alliance (July 10, 2001), http://www.microsoft.com/presspass/press/ 2001/Jul01/07-10VeriSignPR.mspx. See generally PageBox, Trusted Sites, http://www.pagebox. net/java/java-trusted.html (last visited Apr. 9, 2006) (explaining the technology behind trusted website support).

¹⁷⁰ MSDN, Introduction to Code Signing, http://msdn.microsoft.com/library/default.asp?url=/ workshop/security/authcode/intro_authenticode.asp (last visited Apr. 9, 2006) (describing the minimal credentials that a software publisher must present to be eligible for certification).

¹⁷¹ See Microsoft, Digital Signature Benefits for Windows Users, http://www.microsoft.com/ winlogo/benefits/signature-benefits.mspx (last visited Apr. 9, 2006).

¹⁷² See id. (describing Microsoft’s “Certified for Windows” program).

¹⁷³ See supra note 168 and accompanying text.

¹⁷⁴ See David Bank, Microsoft Moves To Rule On-Line Sales, WALL ST. J., June 5, 1997, at B1.

¹⁷⁵ See, e.g., Microsoft, Use Microsoft Update To Help Keep Your Computer Current, http:// www.microsoft.com/athome/security/update/msupdate_keep_current.mspx (last visited Apr. 9, 2006).

¹⁷⁶ In Windows XP Service Pack 2, this feature is automatically turned on. Should the user disable the feature, frequent warnings indicate that the computer is not fully protected due to its inability to receive updates. See Scott Spanbauer, Internet Tips: Tweak Windows XP SP2 Security to Your Advantage, PC WORLD, Oct. 2004, at 166, 167–68, available at http://www.pcworld.com/howto/article/0,aid,117422,00.asp.

¹⁷⁷ This phenomenon is in use by PC applications as well as OSs. See, e.g., Ban Hits Half-Life 2 Pirates Hard, BBC NEWS, Nov. 25, 2004, http://news.bbc.co.uk/2/hi/technology/4041289.stm; Peter Cohen, iTunes Update Disables Internet Playlist Sharing, MACWORLD, May 27, 2003, http://www.macworld.com/news/2003/05/27/itunes/.

¹⁷⁸ See Mark Ward, Toasting the Crackers, BBC NEWS, Jan. 26, 2001, http://news.bbc.co.uk/1/ hi/sci/tech/1138550.stm; May Wong, DirecTV Fires Back at Hackers, MIAMI HERALD, Jan. 27, 2001, at 1C.

¹⁷⁹ See, e.g., Ward, supra note 178.

¹⁸⁰ See John Battelle, Op-Ed., Building a Better Boom, N.Y. TIMES, Nov. 18, 2005, at A29; Tim O’Reilly, What Is Web 2.0, O’REILLY NETWORK, Sept. 30, 2005, http://www.oreillynet.com/ pub/a/oreilly/tim/news/2005/09/30/what-is-web-20.html.

¹⁸¹ For a close examination of the range and benefits of such applications as SETI@home, see Yochai Benkler, Sharing Nicely: On Shareable Goods and the Emergence of Sharing as a Modal-ity of Economic Production, 114 YALE L.J. 273, 291–95 (2004).

¹⁸² For a discussion of concerns that hosted services can create new bottlenecks in data ownership, see the conversation about network effects and service levels in Read/WriteWeb, http:// www.readwriteweb.com (Nov. 15, 2004).

¹⁸³ See Damon Darlin, A Journey to a Thousand Maps Begins with an Open Code, N.Y. TIMES, Oct. 20, 2005, at C9 (describing Google Maps Mania, which catalogues information-rich maps built using Google Maps).

¹⁸⁴ See Google, Google Maps API Terms of Use, http://www.google.com/apis/maps/terms.html (last visited Apr. 9, 2006) (stating that Google reserves the right to modify or discontinue Google Maps at any time and for any reason).

¹⁸⁵ Cohen, supra note 177.

¹⁸⁶ Id.

¹⁸⁷ 464 U.S. 417 (1984). In Sony, the Supreme Court declined to impose contributory liability on VCR manufacturers for enabling VCR users to infringe copyrights. See id. at 419–21.

¹⁸⁸ See A & M Records, Inc. v. Napster, Inc., 114 F. Supp. 2d 896, 927 (N.D. Cal. 2000), aff’d in part and rev’d in part, 239 F.3d 1004 (9th Cir. 2001); see also Napster Filter Welcomed by Music Industry, CNN.COM, Mar. 2, 2001, http://archives.cnn.com/2001/LAW/03/02/napster.hearing.04 (analyzing the aftermath of the Napster litigation).

¹⁸⁹ See Randal C. Picker, Rewinding Sony: The Evolving Product, Phoning Home and the Duty of Ongoing Design, 55 CASE W. RES. L. REV. 749, 766–68 (2005) (arguing that the Sony test should be replaced by one that turns on whether the producer remains able to distribute updates to consumers).

¹⁹⁰ Id. at 752.

¹⁹¹ Id. at 767.

¹⁹² See Cyrus Farivar, New Food for IPods: Audio by Subscription, N.Y. TIMES, Oct. 28, 2004, at G5; Wikipedia, Podcasting, http://en.wikipedia.org/wiki/Podcasting (last visited Apr. 9, 2006).

¹⁹³ See generally Zittrain, supra note 24, at 272–73 (describing the open development model for software development).

¹⁹⁴ See Universal City Studios, Inc. v. Reimerdes, 111 F. Supp. 2d 294, 316–19 (S.D.N.Y. 2000); see also ELEC. FRONTIER FOUND., supra note 88, at 6 (discussing Reimerdes and other DMCA litigation).

¹⁹⁵ For a discussion of the further prospect that law enforcement authorities could issue orders to software makers to use automatic updating to retrieve information from users’ PCs, see Zittrain, supra note 77, at 89.

¹⁹⁶ See Benkler, supra note 181, at 356–58 (describing the value and power of non-market-mediated relationships to produce socially and economically useful results).

¹⁹⁷ S. 2048, 107th Cong. (2002). The CBDTPA was similar in substance to the Security Systems Standards and Certification Act, which was drafted in mid-2001 but never introduced in Congress. See Security Systems Standards and Certification Act, Aug. 6, 2001, http://cryptome.org/sssca.htm.

¹⁹⁸ See S. 2048 § 3.

¹⁹⁹ Id. § 9(3).

²⁰⁰ See id. § 3(c).

²⁰¹ Maria Trombly et al., China’s Bet on Linux, CIO, Oct. 15, 2005, at 21, 21 available at http:// www.cio.com/archive/101505/tl_opensource.html.

²⁰² See, e.g., Lawrence Lessig, The Limits in Open Code: Regulatory Standards and the Future of the Net, 14 BERKELEY TECH. L.J. 759, 768–69 (1999) (suggesting that free software can provide a check on government power because it is less regulable than software generated and maintained by a single firm).

²⁰³ Skype was founded by the makers of the Kazaa filesharing program in August 2003 with $250,000 in seed funding. Its goal was to produce software for PC-to-PC voice communication. See Constance Loizos, Draper Cashes in on Billion-Dollar Skype, PRIVATE EQUITY WEEK, Sept. 19, 2005, http://www.privateequityweek.com/pew/freearticles/1122124881310.html. Skype received an additional $1 to $2 million in November 2003 and $18.8 million more in March 2004. Skype was purchased by eBay in September 2005 for at least $2.5 billion. See id.; Press Release, Skype, eBay Completes Acquisition of Skype (Oct. 14, 2005), http://www.skype.com/ company/news/2005/skype_ebaycompletesacquisition.html. In partnership with Netgear, Skype plans to release a stand-alone wi-fi-based mobile phone in mid-2006. See Press Release, NETGEAR, Inc., NETGEAR, Skype To Connect on Family of Innovative Products Including World’s First Skype WiFi Mobile Phone (Jan. 4, 2006), http://www.netgear.com/pressroom/press_ releasesdetail.php?id=305.

²⁰⁴ Skype represents a good example of code that can evade censorship — its adoption by end users is notoriously difficult for ISPs to stop — and code that could be the vector for the “biggest botnet ever,” given the fact that its code is largely encrypted, its adoption widespread, and its communication with other Skype programs continuous and opaque. See PHILIPPE BIONDI & FABRICE DESCLAUX, SILVER NEEDLE IN THE SKYPE 3–4, 112–13 (2006), available at http:// www.blackhat.com/presentations/bh-europe-06/bh-eu-06-biondi/bh-eu-06-biondi-up.pdf.

²⁰⁵ See Yochai Benkler, Coase’s Penguin, or, Linux and The Nature of the Firm, 112 YALE L.J. 369, 371 (2002) (“A new model of production has taken root, one that should not be there, at least according to our most widely held beliefs about economic behavior. [We] resist the idea that thousands of volunteers could collaborate on a complex economic project. It certainly should not be that these volunteers will beat the largest and best-financed business enterprises in the world at their own game. And yet, this is precisely what is happening in the software industry.”); Julie E. Cohen, The Place of the User in Copyright Law, 74 FORDHAM L. REV. 347, 348–49, 370–73 (2005) (rejecting the simplified notion of the “romantic user” who frequently contributes to debates and transforms others’ work and asserting the possibility of generative contribution by the “situated user” engaged in “consumption, communication, self-development, and creative play”); Dan Hunter & F. Gregory Lastowka, Amateur-to-Amateur, 46 WM. & MARY L. REV. 951 (2004) (describing how the Internet facilitates amateur contribution).

²⁰⁶ See Saltzer et al., supra note 37.

²⁰⁷ See Marjory S. Blumenthal, End-to-End and Subsequent Paradigms, 2002 LAW REV. MICH. ST. U.-DETROIT C.L. 709, 717 (remarking that end-to-end arguments “interact[] with economics, public policy, and advocacy dynamically to shape access to communication and information and to influence innovation”).

²⁰⁸ For articles noting the centrality of end-to-end to the debate, see id., which describes end-to-end as the current paradigm for understanding the Internet, and Lawrence Lessig, The Architecture of Innovation, 51 DUKE L.J. 1783 (2002), which argues that end-to-end establishes the Internet as a commons. For debate about the perceived values at stake in end-to-end arguments, see Yochai Benkler, e2e Map (Stanford Program in Law, Sci. & Tech., Conference Summary, The Policy Implications of End-to-End, 2000), http://cyberlaw.stanford.edu/e2e/e2e_map.html. For arguments for the preservation of end-to-end neutrality in network implementation, see Written Ex Parte of Professor Mark A. Lemley and Professor Lawrence Lessig, In re Application for Consent to the Transfer of Control of Licenses MediaOne Group, Inc. to AT&T Corp., No. 99-251 (F.C.C. 1999), available at http://cyber.law.harvard.edu/works/lessig/cable/fcc/fcc.html; Lemley & Lessig, supra note 8; David D. Clark & Marjory S. Blumenthal, Rethinking the Design of the Internet: The End to End Arguments vs. the Brave New World (Stanford Program in Law, Sci. & Tech., Conference Paper, The Policy Implications of End-to-End, 2000), available at http:// cyberlaw.stanford.edu/e2e/papers/TPRC-Clark-Blumenthal.pdf, which describes the benefits of end-to-end and how those benefits are in tension with security concerns; Paul A. David, The Beginnings and Prospective Ending of “End-to-End”: An Evolutionary Perspective on the Internet’s Architecture 26 (Stanford Econ. Dept., Working Paper No. 01-012, 2001), available at http://www-econ.stanford.edu/faculty/workp/swp01012.pdf, which argues that end-to-end openness is a public good, the potential loss to society of which must be calculated when more extensive security solutions are considered; and David P. Reed et al., Active Networking and End-to-End Arguments (Stanford Program in Law, Science & Tech., Conference Paper, The Policy Implications of End-to-End, 2000), http://cyberlaw.stanford.edu/e2e/papers/Saltzer_Clark_Reed_ActiveNetworkinge2e. html, which argues for the preservation of end-to-end and using end-to-end openness as an organizing principle against which to measure programmability and active networking.

²⁰⁹ See supra pp. 1988–89.

²¹⁰ See Saul Hansell, Spam Fighters Turn to Identifying Legitimate E-Mail, N.Y. TIMES, Oct. 6, 2003, at C1 (discussing authentication and other possible solutions for limiting spam); Yakov Shafranovich, 2004: The Year That Promised Email Authentication, CIRCLEID, Dec. 27, 2004, http://www.circleid.com/posts/2004_the_year_that_promised_email_authentication (discussing various e-mail authentication proposals to limit spam on the receiving end); see also Saul Hansell, 4 Rivals Near Agreement on Ways To Fight Spam, N.Y. TIMES, June 23, 2004, at C1 (discussing approaches toward authentication proposed by major ISPs).

²¹¹ See, e.g., DUDNEY, supra note 134 (providing a case study of traffic filtering by ISPs).

²¹² See World Summit on the Info. Soc’y, Overview, http://www.itu.int/wsis/basic/about.html (last visited Apr. 9, 2006). For commentary on the World Summit, see, for example, John Markoff, Control the Internet? A Futile Pursuit, Some Say, N.Y. TIMES, Nov. 14, 2005, at C4; and Victoria Shannon, Other Nations Hope To Loosen U.S. Grip on Internet, N.Y. TIMES, Nov. 15, 2005, at C14.

²¹³ See World Summit on the Info. Soc’y, Why a Summit on the Information Society, http:// www.itu.int/wsis/basic/why.html (last visited Apr. 9, 2006); see also Jennifer L. Schenker, U.N. Meeting Debates Software for Poor Nations, N.Y. TIMES, Dec. 11, 2003, at C4 (reporting that the World Summit representatives portrayed “open-source, or free-to-share, software [as] crucial for the developing world because it would permit poorer countries to develop their own technology instead of having to import it”).

²¹⁴ See, e.g., Anti-Spyware Coal., http://www.antispywarecoalition.org (last visited Apr. 9, 2006) (association of information technology companies that focuses on combating spyware); StopBadware.org, http://www.stopbadware.org (last visited Apr. 9, 2006) (nonprofit academic project aimed at fighting malicious software); see also The WildList Org. Int’l Home Page, http://www. wildlist.org (last visited Apr. 9, 2006) (grassroots organization devoted to disseminating information about viruses).

²¹⁵ See L. Jean Camp & Allan Friedman, Good Neighbors Can Make Good Fences: A Peer-to-Peer User Security System (Univ. of Mich. Sch. of Info., Conference Paper, Telecommunications Policy and Research Conference, Sept. 24, 2005), available at http://web.si.umich.edu/tprc/papers/2005/453/tprc_GoodNeighbors.pdf; Alla Genkina & L. Jean Camp, Re-Embedding Existing Social Networks into Online Experiences To Aid in Trust Assessment (Apr. 1, 2005), available at http:// ssrn.com/id=707139; L. Jean Camp et al., Net Trust, http://www.ljean.com/netTrust.html (last visited Apr. 9, 2006).

²¹⁶ Indeed, such a scenario need not be hypothetical. See, e.g., Ken Silverstein, The Radioactive Boy Scout, HARPER’S MAG., Nov. 1998, at 59 (recounting the story of a child who created a nuclear reactor in his backyard shed).

²¹⁷ Some scholars, however, rebuke the notion that nonphysical harm is always less injurious than physical harm. See, e.g., Eugene Volokh, Crime-Facilitating Speech, 57 STAN. L. REV. 1095, 1217 (2005) (commenting on the significance of speech that facilitates grave nonphysical harm and suggesting, therefore, that it ought to enjoy no First Amendment protection).

²¹⁸ See supra p. 2006.

²¹⁹ For a preliminary sketch of such a division, see BUTLER LAMPSON, ACCOUNTABILITY AND FREEDOM (2005), available at http://www.ics.uci.edu/~cybrtrst/Posters/Lampson.pdf.

²²⁰ A virtual machine is a self-contained operating environment that isolates an application from the entire computer on which it runs, denying the application access to other compartments of the system. See Wikipedia, Virtual Machine, http://en.wikipedia.org/wiki/Virtual_machine (last visited Apr. 9, 2006).

²²¹ See sources cited supra note 210. One example of such an authentication system is Microsoft’s Sender ID. See Microsoft, Sender ID, http://www.microsoft.com/mscorp/safety/ technologies/senderid/default.mspx (last visited Apr. 9, 2006).

²²² See generally David R. Johnson et al., The Accountable Internet: Peer Production of Internet Governance, 9 VA. J.L. & TECH. 9 (2004), http://www.vjolt.net/vol9/issue3/v9i3_a09-Palfrey. pdf (discussing the imperfections of filtration).

²²³ See Yochai Benkler, Some Economics of Wireless Communications, 16 HARV. J.L. & TECH. 25 (2002) (suggesting that open wireless networks will be more efficient at optimizing wireless communications capacity than spectrum property rights will be).

²²⁴ See Michel Marriott, Hey Neighbor, Stop Piggybacking on My Wireless, N.Y. TIMES, Mar. 5, 2006, at A1 (explaining some of the dangers of open wireless networks).

²²⁵ See supra p. 2022.

²²⁶ See Wikipedia, Blue Screen of Death, http://en.wikipedia.org/wiki/Blue_screen_of_death (last visited Apr. 9, 2006).

© Copyright 2006 Jonathan Zittrain