The Unending Search for the Optimal Infringement Filter

The authors' proposal would profoundly alter the legal and technical schemata of the Internet: offering immunity only for webhosts that employ the best available method for filtering content prior to publication. Like many technologies, the idea of an optimal infringement filterdescribed as a rule--based, machine--implemented algorithmic approach to identifying and preventing infringement versus the more standards--based approach of the human mind and judicial enforcement-provides an attractive yet deceptive solution to the authors' copyright quandary.

(collectively treated as "Online Service Providers" in the statute ("OSPs")) with immunity from damages as long as they satisfy certain safe harbor definitions. As many scholars have shown, this section was the product of a delicate compromise reached by Congress between the twin forces of innovation and copyright protection: In order for the Web to develop and innovate, Congress recognized that OSPs would need some immunity from liability for copyright infringement by end users.5 At the time, the application of the doctrines of vicarious and contributory copyright infringement to the complex world of the Internet was a relatively new phenomenon. As a result, OSPs sought clarification on their potential liability for activities like providing connections, storing data, caching and indexing content, and the like. 6 The resulting safe harbors, therefore, were motivated out of the recognition that limiting OSP liability would ensure "'that the efficiency of the Internet will continue to improve and that the variety and quality of services on the Internet will continue to expand.'"7 At the same time, however, Congress set forth some levels of responsibility for webhosts and OSPs to follow: They can lose protection, for example, if the webhost fails to remove infringing material expeditiously, fails to reasonably implement a termination policy, or receives a direct financial benefit from the infringement while having the right and ability to control infringing activity.8

As long as the provider meets these qualifications, they do not have to actively police their computer networks and servers for infringement; rather, they can simply wait until they receive a valid notice to act. This places the burden of discovering infringing content squarely on copyright holders or their agents. As one commentator explained, given the automated nature of providing many online services, among other variables, the DMCA "allocates primary responsibility for policing content on the party with the greatest interest in protecting copyrighted work: the right holders themselves."9 As a result of these guidelines, intellectual property owners have undertaken a program of locating and notifying service providers of infringing material, and ISPs have developed a response system that acts to "take down" allegedly infringing material in order to avoid allegations of contributory liability.

More than ten years after this system was designed and agreed to by both the large content owners and online service providers of the day, the The authors argue that this result is far more desirable than the current system for three main reasons: (1) mechanical filters will be more cost--efficient than humans at identifying and removing infringing works, especially before they are published online; (2) their solution will spur much more innovation among filtering technologies; and (3) their solution provides encouragement to webhosts to filter and will not significantly harm social welfare concerns such as free speech or fair use. Further, the authors assure us that the use of a safe harbor ameliorates the legal uncertainties that webhosts currently face. "If a copyright infringement suit is brought against a webhost, all it will have to do to defeat the lawsuit will be to show that it referred users' content to a clearinghouse for filtering at the time of the infringement," the authors explain.15

Despite our initial skepticism, the possibility of an optimal, or even efficient, infringement filter seems like a worthwhile pursuit. However, 14. While such information is not universally publicly available, some public sources do exist. like many such proposals, Helman and Parchomovsky seem to raise more questions than they answer. One of the biggest threshold issues is the authors' promise of protecting innovation as a desirable spillover effect of their solution. "The technological safe harbor we envision," the authors write, "is a dynamic one; it will change with technological progress, creating opportunities for technology companies to develop superior filtering technologies and spurring websites to adopt them."16 The more effective the technology, the authors reason, the more likely it will be adopted by a larger number of websites, particularly if the law requires it.

While we empathize with the desire to encourage innovation, we would respectfully suggest that the authors take into greater consideration the immense costs that their proposal might impose on webhosts upon being deputized to do the work that copyright owners are already doing. While it is true that their regime would provide an emergent, unproven industry with an added incentive to innovate, the web services and content dissemination industries-proven drivers of economic growth17-would unquestionably be hampered if not harmed by the heavy cost of mandatory adoption, implementation, and maintenance. These costs could undermine other incentives and dampen development of a variety of new communication and distribution technologies, essentially "picking a winner" in the contest for R&D resources instead of allowing the technologies and markets to interact in a truly dynamic fashion. Moreover, to the extent it encourages copyright enforcement innovation, it locks it into the single category of filtering, even though it is well known that network technologies often "route around" content filters and inhibitors as an architectural matter.18 Thus, even if there were dynamism in filtering innovations, such technology may well become irrelevant to the overall network in the end. 19 It is hard to

The Unending Search 89 imagine how the benefits of the dynamism the authors promise could compare with those from more general dissemination technologiesunder their solution, the birth of another YouTube might be impossible if filtering is required in order to receive section 512 DMCA protection.20

Even aside from the threshold questions about dynamism, efficiency, and innovation, there remain other questions about their proposal. Below we group these questions into four provocations that identify what we consider crucial inquiries for making a compelling case for online ex ante infringement filters as an appropriate legal solution.

As noted above, Helman and Parchomovsky propose a solution that has great intuitive appeal for those that believe that the current system is not only challenging and costly for copyright owners but also potentially futile. For example, the authors assert that "expenditures involved in policing the Internet on a regular basis and issuing takedown requests may be prohibitive for many content owners, particularly individual authors, and independent studios and publishers" and "[c]ooperation with webhosts may be the only feasible way to enforce the rights of these content owners."21 The authors argue that active monitoring and enforcement by webhosts can complement the efforts of content owners, because webhosts are best situated to disrupt infringing activities and because they have a better ability to deploy automatic filters and block materials before they are posted. Since copyright owners cannot apply automatic filters ex ante, nor can they limit searches to newly uploaded content, they are forced to comb through every type of file, ex post, for infringing content, over and over again, which seems overly onerous and economically wasteful. The authors write, "[i]mposing th[is] duty on content owners makes very little sense as they have no access to the code of hosting sites, and website operators will rightly be reluctant to grant content owners access privileges to their proprietary code or otherwise intervene in the core operation of their business."22 While we agree about the potentially duplicative costs of forcing copyright owners to continually search each and every website for a particular type of file, we are not convinced that their solution offers a more economical alternative when one considers all of the various costs and benefits that their solution might generate. First, from a copyright protection standpoint, it is not clear what the exact economic and societal benefits are from an ex ante regime that removes infringing content at the earliest possible opportunity. While the authors argue that the most profitable window for a copyrighted work occurs at the beginning of its lifetime,23 the ex ante filter is not tied to the date of the work's creation; rather, it is tied to the date of the allegedly infringing publication, which could be days or months or years later, especially in the case of out--of-print or so--called "orphan works" that may lack significant economic value.24 In addition, presumptions of economic harm in copyright cases have recently come under fire, forcing copyright owners to provide courts with strong evidence of irreparable economic harm before gaining the benefits of injunctive relief, a remedy akin to what the authors propose.25 Therefore, while the case for an ex ante injunction for some works might surely be justified, the case for others might be quite weak. To apply such a strong singular economic presumption to all copyrighted works ever created-regardless of how old they are and what their current economic value might be-raises serious questions as to the benefits such an approach offers and the evidence to justify them.

The authors also fail to offer much evidence on whether or not the costs of copyright--owner enforcement are indeed prohibitive; their only citations in support are to a single law review article speculating as to the problem of enforcement cost and examples of the cost of filtering services, not to any direct evidence on the cost of enforcement. Yet the cost to hire humans to search online and send takedown notices is a critical piece of evidence in this debate. For instance, it could actually be quite low, depending on the labor costs involved, and whether the work is being done in the U.S. or abroad. Further, it might not be necessary to take down every infringing file in order to maintain profitability in specific copyright industries, but rather to remove only the files that come up most often in searches on the first few pages of results. Obscure files are likely less damaging to copyright--holder interests, since they would be downloaded less often. Which version of cost and benefit is true? Again, more evidence on this point is needed.26 Moreover, from an innovation standpoint, it also makes sense to contemplate precisely why webhosts should bear the costs of copyright 23. Id. at 1204. 24. See generally U.S. Copyright Office, Orphan Works, http://copyright.gov/orphan (on file with the Columbia Law Review) (last visited Feb. 10, 2012) (providing background information on orphan works problem).

25. See infra note 76 and accompanying text (discussing recent case law establishing injunctions are not automatic in infringement cases).

The cost of the extant regime is exacerbated by the time--sensitive value of intellectual property. As a general rule, copyright content is most valuable immediately after its release. With the passage of time, the value of content diminishes. As a result, by the time offending content is finally removed, it may be virtually valueless. Helman & Parchomovsky, supra note 2 at 1204. Yet the sole citation for these assertions is to Nat'l Basketball Ass'n v. Motorola, Inc., 105 F.3d 841, 853 (2d Cir. 1997), a case about noncopyrightable "hot news" that has been heavily critiqued and is doctrinally inapplicable to copyrighted works. Moreover, there is strong evidence that many copyrighted works have "long tails" that allow them to produce both intrinsic and extrinsic value over time. See Chris Anderson, The Long Tail: Why the Future of Business is Selling Less of More 19-22 (2006) (describing "long tail" of album tracks).

The Unending Search 91 enforcement when their role, since 1998, has been largely to publish without making content--based considerations. Indeed, their immunityand often their profitability-has depended on it, and to shift the burden along the lines suggested would require a more precise examination of the effects on business models and paths towards innovation. For example, in calculating the "cheapest cost avoider," the authors would have to calculate, or at least estimate, the effect of the substantial shift in information costs regarding copyrighted properties that would now be faced by webhosts due to the shift in filtering responsibilities.

Further, requiring webhosts to filter would substantially raise the cost of publication for webhosts without a predictive sense of how they will respond to this changing variable, namely, how the culture and speed of the Internet is likely to change with automated webhost filtering. The original design of the Internet is largely open and easy to build upon-a system that is designed to move content between hosts and clients as quickly, cheaply, and neutrally as possible, leaving other parties to make determinations about legality and appropriateness after they are posted. Yet content filtering by webhosts necessarily introduces an element of selection and delay into the posting of content, potentially changing the nature and speed of web publication generally.27 How will webhosts respond? Consider, for example, Professor Tim Wu's observations when confronted with the idea of ISP filtering by AT&T:

On the technical side, if I were an AT&T engineer asked to implement this plan, I would resign immediately and look for work at Verizon. AT&T's engineers are already trying to manage the feat of getting trillions of packets around the world at light speed. To begin examining those packets for illegal pictures of Britney Spears would be a nuisance, at best, and a threat to the whole Internet, at worst. Imagine if FedEx were forced to examine every parcel for drug paraphernalia: Next--day delivery would soon go up in smoke. Even China's Internet, whose performance suffers greatly from its filtering, doesn't go as far as what AT&T is proposing.28 While Wu is definitively speaking about ISP filtering, as opposed to webhosts, we would respectfully suggest that his observations are salient in this context as well, particularly given that AT&T engages in both webhosting and more traditional ISP activity. Moreover, when one considers the slim margins in the web hosting market due to heavy competition-for example, YouTube offers free hosting for videos-and the predicted costs of filtering (which the authors acknowledge could be up to $1 million per month per webhost in some instances), one must also question whether the benefits of preventing ex ante infringement would outweigh the costs to consumers, especially given that web hosts are likely to pass these costs onto the consumer regardless of whether they have ever uploaded infringing content or not. Indeed, we note Wu's observations because they suggest the need for further scrutiny of a fascinating distinction that the authors make between actively distinguishing ISPs from webhosts. While they aggressively favor webhost filtering, curiously, the authors just as aggressively oppose filtering by ISPs. Ironically, they conclude-just as Wu does-that applying their system to ISPs would "yield considerably more harm than benefit," namely, harm to innovation privacy, censorship, etc.29 Moreover, since ISP filtering involves scrutinizing all internet traffic in real time, they observe, it would degrade the performance of the system. Indeed, the authors themselves even point out that the lack of ISP transparency and the significant costs of ISP filtering suggest that the costs of filtering are likely to be borne by the subscriber, further surrendering users' interests. Despite these concerns, which they lay out in convincing fashion, the authors then take the opposite view-aggressively promoting filtering by webhosts. Yet for us, the most interesting question to examine is why the authors draw such a unilateral distinction between the two. There is very little discussion here of what evidence draws them to reach this distinction, and more information here would be helpful. Many ISPs also provide webhosting services (AT&T, for example)-and at the very least, some more discussion may be necessary to explain why the Article takes a unilateral stance against ISP filtering but would require it in the webhost context. For example, as the authors note, the performance costs that ISPs would incur attempting to filter traffic in real time are significant, yet video sites like YouTube receive over sixty hours of video every minute in real time.30 The Article fails to explain why the impact of filtering on the performance of YouTube (or perhaps more importantly, smaller, less-resourced, sites) is any less of a concern.

The law, too, has at present failed to treat them differently as a policy matter, particularly where the all--important safe harbor is concerned, which suggests that Congress was just as concerned about innovation in the webhosting domain as it was in the ISP realm.31 We note these 31. Indeed, the DMCA and case law all too often conflate the two. See Darrow & Ferrera, supra note 8, at 13 n.90 (reaching this observation). Construing the term "service provider," courts have consistently applied a very broad definition, encompassing webhosting entities and other activities like Aimster, eBay, and Amazon. Id

The Unending Search 93 evidentiary points, not just as a statutory matter, but because they suggest that the authors may be overlooking the kinds of concerns that led Congress to immunize webhosts (along with ISPs) in the DMCA. Innovation happens at many points and from many different entities, and we would respectfully suggest that the authors may wish to consider further the distributive costs that webhosts would face under their solution and how it might affect future business models.

In addition, the authors assert that there is a disincentive to filter or monitor outside the very narrow "red flag" bounds that the DMCA requires.32 According to Helman and Parchomovsky, the current legal regime lacks sufficient predictability and certainty, leaving webhosts in an undecided gray area. Yet, examples of active filtering and monitoring abound in the form of YouTube's Content ID and the existence of filter vendors such as Audible Magic. A recent study reported that "all" major copyright holders and user--generated content sites utilize filtering technology in some form, and one digital fingerprinting company, Audible Magic, offers its technology for free to smaller user--generated content sites.33 If webhosts had disincentives to filter or monitor, then who is hiring these vendors? Further, what accounts for the range of voluntary extralegal measures that webhosts have willingly undertaken to address infringement, like the landmark User Generated Content Principles (UGC), which marshaled agreement from both the content industries and the web services industry? 34 The Principles clearly anticipate the use of content identification technology that eliminates infringing content, and also call for copyright owners to provide user--generated content web sites with copyright reference material in order to establish a match and instructions on how to treat matches when they occur. Under this system, user generated web sites will utilize the reference information, filter and scan for matches, and respond to those matches-all before posting the material online.35 Similarly, in the ISP context, the development of graduated response and copyright alert measures, while such measures are admittedly imperfect, does suggest that much of the authors' concerns are being addressed through the development of voluntary measures to curb infringement online. Consequently, it seems that the authors may overstate the law's effect on social norms, without truly grappling with the variety of solutions-and incentives to filter-that are already firmly in in the context of the § 512(c) safe harbor]"); Hendrickson v. eBay, Inc., 165 F. Supp. 2d 1082, 1088 (C.D. Cal. 2001) ("eBay clearly meets the DMCA's broad definition of online 'service provider.'").

32. Helman & Parchomovsky, supra note 2, at 1202. 33. See Gallo, supra note 6, at 312 & n.239. 34 The User Generated Content Principles were developed by "leading commercial copyright owners . . . and services providing user--uploaded and user--generated audio and video content . . . to foster an online environment that promotes the promises and benefits of UGC Services and protects the rights of Copyright Owners." See User Generated Content Principles, http://www.ugcprinciples.com/ (last visited April 8, 2012).

35. See Gallo, supra note 6 at 295 ("UGCs are essentially required to participate in this matching process by using the reference material to 'filter' or scan user--uploaded content before that content is made available on its services.").

A related assumption is that competition in the marketplace for filter development is suboptimal.36 Yet the authors provide little to no data on this issue; of course, more innovation may often be desirable, but without some empirical sense of the incentives-or disincentives-to innovate in this field, it is hard to say what an optimal level would be. (Indeed, from our end, antipiracy efforts seem to be rapidly growing and improving, not shrinking, under the watchful encouragement of the content industry.)37 Some of these measures are far more creative and innovative than the authors' method would allow; for example, YouTube's Audio ID and Video ID operate by enabling a rights holder to choose whether to block, track, or monetize the posting, offering a wider menu of content management tools to the copyright owner than a filter alone would provide.38 A further, evidentiary point: While we are sympathetic to the idea that having copyright owners constantly troll the Internet for infringing material is costly, we remain less convinced that costs would be lowered by shifting the responsibility to webhosts instead. Both systems involve some duplication: Currently, the authors maintain that content owners have to scan each and every site for infringing material; and under their proposal, each and every webhost would have to scan each and every file prior to posting. The difference between the first and the second scenarios, however, is that there is arguably more collective action in the first instance, because more and more copyrighted works are becoming concentrated in the hands of fewer and fewer parties, and because rights holders face lower information costs than webhosts regarding their copyrighted works. This brings down the cost of monitoring because copyright holders' filtering strategies already bundle more and more works in their efforts.39 Arguably, since there appear to be only a limited number of piracy surveillance vendors, the arguments about preserving competition are still true even in an ex post regime, making it unclear precisely how much would actually be saved by transferring these costs to

The Unending Search 95 webhosts. Further, the authors do not account for the possibility that much of the infringement that content owners care deeply about stems from a small number of websites (e.g., MegaUpload.com), which means that "trolling" the entire Internet may not be necessary at all. Comparatively, the startup costs of creating this new system would be significant, and shifting them to every single webhost seems inefficient without a deeper empirical showing of precisely how much would be saved, relative to the amount that has already been spent by copyright owners to solve the problem.

Even the issue of what would constitute the "best" filter poses questions of evidence and process. Arguing that a best available filter can be a determinable strategy is a bit like arguing that there is one best meal, best book, or best student out there. There are far too many variableswith too many types of copyrighted works and too many degrees of infringement-to presume that a singular standard can emerge and serve as a litmus test for safe harbor protection. For example, the authors' argument rests on the presumption that false negatives (infringing posts that are not blocked) should be weighed equally with false positives (noninfringing posts that are blocked). We are not sure that the two are equal. While it is a difficult thing to weigh the costs of permitting one infringing file to exist versus the costs of censoring one noninfringing file, it is safe to say that the latter involves a potential chilling effect on speech broadly, whereas the former does not. Put differently, a single instance of censorship bears ripple effects that cause greater chilling of speech among noninfringing activity, and this is an externality that is worth considering.

A further question is institutional competence. Should we entrust this determination to an institutional body-the Copyright Office-that is at least arguably susceptible to capture and major public choice concerns?40 Moreover, does the Copyright Office have the expertise to evaluate technological filters? One would imagine that this would require significant engineering, computer science, and informatics expertise, something the Copyright Office has not been delegated or funded to pursue. And what happens if the Office's determination is contested? How many years and how much money will be spent litigating those issues? And what will happen to the safe harbor in the meanwhile? Last, the information costs would be significant under the authors' proposal due to the difficulty of identifying protected material. The authors acknowledge this, but then argue that a database that collects

works and gray areas involving contested noninfringing works will be handled to the reader's satisfaction. To support this conclusion, they rely almost entirely on the presumption that the two most common file types to be filtered will be either the easy noninfringing works that have no matching content or entire "verbatim copies" of works are almost always infringing; thus, filters that detect these attributes will successfully determine the legal status of most files.42

In truth, however, the question of copyright infringement is far more complex than the authors give it credit. For example, in order to determine that a file is infringing, a mechanical filter would have to draw a sufficient legal conclusion concerning at least the following:

• Does the individual or entity that objects to the use of the work have standing under both Article III of the Constitution and Section 501(b) of the Copyright Act to enforce the copyright?44

• For works subject to the 1909 Copyright Act, were all applicable formalities satisfied?45

• Is the uploaded file substantially similar to the protectable expression of the original work of authorship, which excludes at a minimum unprotectable ideas, scènes--à--faire, expression that is merged with ideas, and standards within an industry?46 42. See Helman & Parchomovsky, supra note 2, at 1229-30 (describing categories of blatant copyright infringements, noninfringing posts, and some "hard cases").

43. See, e.g., Cmty. for Creative Non--Violence v. Reid, 490 U.S. 730, 749-50 (1989) (discussing importance of certain copyright ownership); Press Release, Elec. Frontier Found., Righthaven Copyright Troll Lawsuit Dismissed as Sham (June 14, 2011), https://www.eff.org/press/archives/2011/06/14 (on file with the Columbia Law Review) (discussing dismissal of copyright lawsuit by organization that "never owned the copyright in the first place").

44. See, e.g., Silvers v. Sony Pictures, Inc., 402 F.3d 881, 890 (9th Cir. 2005) (holding screenwriter who purchased right to sue for infringement still lacked standing without ownership of at least one exclusive right under 17 U.S.C. § 106); see also Elec. Frontier Found., Diehl v. Crook, available at https://www.eff.org/cases/diehl--v--crook (last visited Feb. 11, 2012) (on file with the Columbia Law Review) (describing settlement of case in which online journalist sued individual for sending "baseless [DMCA] takedown notices claiming to be the copyright holder of an image used in the story").

45. See, e.g., Elec. Frontier Found., JibJab Media v. Ludlow Music, available at https://www.eff.org/cases/jibjab--media--inc--v--ludlow--music--inc (last visited Feb.

Moreover, even the suggestion that filters will somehow "know" enough to make other legal determinations-such as whether or not a work has been published-is suspect. The authors suggest that the filters will be able to rely on the Copyright Office records for such information, or on information submitted by the copyright owners themselves. Yet there is no evidence that these records are accurate enough to presume their validity. The Copyright Office does not verify their accuracy, instead assuming that if publication is contested, it will be resolved in court. Thus, the incentives for copyright owners, even of unpublished works, would be to lie and always say that something is published, as it would directly benefit them in terms of filtering. As Jason Mazzone has shown, copyfraud continues to be a serious problem.62 Ownership of a work can also be highly contested and illegitimate or multiple claims easy to make.63

These complex questions of infringement are exactly the reason Congress places service providers in a neutral role under the DMCA. While the providers do have important technical information about the file itself, they lack numerous facts that determine its legal status. Maintaining the copyright owner or her agent as the policing party is one way to allocate responsibility more closely to those most likely to know or discover these facts. Until and unless filters can somehow outperform copyright owners, it seems ill advised to us to shift this responsibility.

Procedural due process has been the subject of several articles that are critical of the current safe harbor regime's notice--and--takedown provisions.64 Thus, it is worth considering whether the authors' solution would provide more or less process. In his seminal article on procedural due process, Some Kind of Hearing, Judge Henry Friendly lays out eleven elements of a fair hearing. These include an unbiased tribunal; notice of the proposed action and the grounds for it; an opportunity to present reasons why the proposed actions should not be taken; the right to call witnesses; the right to know the evidence against one; the right to have a decision based only on the evidence presented; the right to counsel; the making of a record; a written statement of reasons; public attendance; and judicial review.65 While Judge Friendly goes on to argue that not every one of these criteria is required for every hearing, they provide a useful framework for evaluating the impact of an optimal filter on due process concerns. Under these criteria, the authors' proposal does not appear to fare well. While one might argue that a mechanical filter is an unbiased tribunal and will only make a decision based on the evidence presented, there appears to be little in the authors' proposal to account for the other nine elements.

The best available technology standard also appears to lack any accountability for mistakes or knowing misrepresentations, either substantive or procedural. Who is responsible for violations of Judge Friendly's list? Who is responsible for suppressing the noninfringing creative expressions that might be flagged as false positives by this best filter? Under the current DMCA regime, Congress saw fit to include a counterbalancing mechanism for false and fraudulent takedowns, § 512(f), which allows suits by users or service providers against those who knowingly send notices with material misrepresentations in them.66 And

First Amendment principles in copyright law.74 Yet there is no evidence that any filter, let alone the best available, is currently capable of detecting when these principles should come into play. The authors presume that the "best filtering technology" will actually work in most instances. But if it doesn't, for some reason (for example, due to countermeasures taken to evade filtering), then the law has effectively made a decision to grant a subsidy to companies that adopt suboptimal and perhaps even counterproductive technologies. It is hard to see how this system ensures more efficiency given this risk.75 Moreover, recent case law regarding intellectual property infringement and injunctions has established that injunctions, both preliminary and permanent, are not automatic in either patent or copyright cases even when infringement has been found.76 Yet one of the main benefits proposed by the authors of filtering is that the machines will block publication ex ante-a remedy that appears increasingly difficult to achieve in federal court these days. In light of these concerns, it is surprising that the authors insist so vehemently on ex ante filtering instead of ex post compensation. Given the strength of these precedents, monetary compensation would be the likely outcome from a federal court action and thus, potentially a better model for any technological approximation of law instead of blocking or banning content prior to publication. In fact, this appears to be how YouTube's Content ID has increasingly approached the filtering issue.77

Without a compelling justification for explaining why ex ante filtering is preferable to ex post takedowns, the authors' argument seems to seriously overlook the substantial prior restraint risks of their proposal. Given that ex ante prevention of publication is the quintessential example of a First Amendment prior restraint, there are additional strong reasons to resist this approach. Among them is the question of timeliness of publication. The authors acknowledge that their filter model may be imperfect and there may be cases where mistakes are made. However, they argue, users should be able to counternotice and raise objections to being filtered and thus, be able to exercise their First Amendment rights at some future time if they are truly noninfringing.

However, our First Amendment cases show that the doctrine is quite time--sensitive, raising questions about the utility of counternotification.78 Given that so many online political commentaries depend on reuse and remix of original video, government--mandated filters would almost certainly raise serious concerns.79 Imagine, for example, that Fox News regularly submits each segment of the O'Reilly Factor as a separate copyrighted work to all of the best available filters immediately upon airing. The night before the 2012 Presidential Election, Mr. O'Reilly misrepresents the position of Barack Obama on a critical policy issue. In response, The Colbert Report and The Daily Show attack O'Reilly's misrepresentation by showing a clip of his program followed by other video clips that disprove his assertion. Under current copyright law, this type of commentary and criticism is classic fair use and nothing short of an eleventh hour temporary restraining order-which would likely be denied-would prevent it from airing on cable and satellite networks all over the country. However, for the online versions of shows, even the best filter would most likely block the clips ex ante from being published online because they used wholesale verbatim copies of the O'Reilly segment, an outcome explicitly prohibited by the Arizona decision. Any subsequent counternotification or appeal could take days if not weeks to resolve, rendering the critiques irrelevant in terms of influencing voters in the election.

At the very least, rewriting the DMCA to require content filtering from webhosts, without a prior showing of infringement would surely invoke the kind of constitutional scrutiny that the law originally sought to avoid. Again, while the authors spend some pages discussing this issue and acknowledging it as a concern, they ultimately delegate the tough questions and their consequences to the technology itself to resolve.

Given the questions we raise about evidence, infringement, due process, and about the impact on social welfare from such filters, especially in the arena of free speech and the First Amendment, the Helman and Parchomovsky proposal seems premature and potentially ill--