The Past, Present, and Future of Notice and Takedown

Hello all, my name is Alex Peele and I have the pleasure of welcoming you to this year's summer legal fellows panel. Today we turn our attention to the past, present, and future of notice and takedown. The notice and takedown requirements of the digital millennium copyright act safe harbor provision have been a longtime fixture of internet policy and have greatly shaped the relationship between online platforms and their users both within the United States and abroad. It was a novel legal mechanism devised for this new thing called the internet written before anyone knew what it would grow into and has been more or less unique within US law since. But now after over 25 years later, Congress is set on adopting this mechanism in new and upcoming legislation governing online platforms like the Take It Down and No Fakes Acts. We hope that today's panel sheds light on why notice and takeown has become the mechanism of choice for legislators and how the history of the DMCA might help us anticipate the impact these new laws will have. Before introducing our fantastic panelists and moderator, I would like to thank my colleagues and summer Wikipdia fellows Daniel Echin, Jack and Nita whose tireless work and commitment made this event possible. Similarly, we want to thank Jim Buati, our legal director, whose guidance was crucial during our fellowship at the Wikip Media Foundation. We'd also like to extend the invite to any of those interested in joining the Wikipdia Foundation's fellowship program. If you're a current law student or recent law school graduate passionate about free knowledge and open source issues, our legal fellowship program can provide incredible in-house experience in internet law through leading research and experimental projects aimed at improving our community. Anyone interested can feel free to reach out to Jim at juati wikimedia.org. Without further ado, I want to welcome our moderator, Stan Adams. Stan is the Wikipdia Foundation's lead public policy specialist for North America. He's responsible for analyzing and responding to legislative and regulatory proposals that could impact Wikipdia projects and the free knowledge movement in North America. This work includes defending existing legal protections that protect volunteer editors from frivolous lawsuits like section 230 and advocating for stronger laws to protect the privacy of the people who read and edit Wikipedia. Stan has spent over a decade working on tech policy issues including digital copyright, free expression, privacy, and AI. Prior to his role at the Wikipdia Foundation, Sans served as general counsel to US Senator John Oaf and as a deputy general counsel at the Center for Democracy and Technology in Washington, DC. Joining Stan on today's panel is Jessica Litman, the John F. Nicole Professor of Law at the University of Michigan Law School, adviser for the American Law Institute for a statement of copyright, a director of the American Trademark Law Society, and the author of Digital Copyright. Ranny Adams, a clinical teaching fellow with the Intellectual Property and Information Policy Clinic at Georgetown Law Center. Paul Giz, co-founder of Profina, a San Francisco based personal data technology company and senior lecturer at Vnius University law faculty in Lithuania, where he teaches courses on human centric data privacy law and copyright X in collaboration with Harvard's Burkeman Klein Center for Internet and Society. and Meredith Rose, the senior policy council at Public Knowledge, a nonprofit organization that promotes freedom of expression and open internet and access to affordable communications tools and creative works. This is an ever evolving field and we definitely also want to hear your thoughts as well. There will be plenty of time for Q&A at the end, so please add your thoughts and questions in the chat. With that, thanks again for being here and we'll hand the conversation to you, Stan. >> Thanks, Alex, and thanks again to our panelists for joining us today. Um, let's do a little bit of scene setting and I'm going to turn the first question to to Jessica with a little bit of intro here. When the DMCA, which was uh enacted in 1998, um became law, uh it included but was not limited to the notice and takedown provisions of section 512. Um it had been the product of years of discussion and negotiation about the future of copyright in the digital millennium. Why was notice and takedown chosen uh as the vehicle to enforce copyright online? And why weren't other policy tools such as the framework codified in section 230 of the Communications Decency Act chosen instead? Oh, you're still muted there, Jessica. The notice and takedown provisions are one moving part in a huge compromise with many moving parts. Um, and it's worth appreciating just how terrified uh, legacy entertainment and information media were about the whole idea of the internet. Um, and the Clinton administration basically promised them that they would make the internet safe uh for movies and music and software and everything else. So, the compromise evolved with a bunch of different pieces. One of them was an investment in copy protection that would, if not prevent, at least make it extremely difficult to gain unauthorized access or make unauthorized copies of any copyrighted material. And uh a second was a an expedited way for copyright owners to get stuff taken down if an unauthorized copy made it into the wild. And initially copyright owners argued that any time an unauthorized copy appeared in any computer connected to the internet that was a copy. it was actionable on a strict liability basis. And the telephone companies that were that error as internet service providers said that's absolutely crazy talk. Uh but the Church of Scientology uh responded to the unauthorized posting of some of its secret church documents by suing everyone who had a copy uh from the Washington Post onto the internet service providers who transmitted the files from Usenet uh to various people's screens. And the judge who decided that case said, "Well, look, I'm not going to hold computer systems liable for stuff they can't know about and can't control. They're just putting the systems in place and people are using them. But if they know, if someone informs them that this is infringing, then I'm comfortable holding them liable for not taking reasonable steps to block it or take it down." And Congress in essence took that compromise and embodied it in the notice and takedown rules. Now, the notice and takedown rules got very complicated uh because of different kinds of safe harbors for different kinds of copies, and Senator John Ashccraftoft insisted on the putback uh provisions that allowed people whose material had been unfairly or wrongly blocked to uh get stuff put back. But the basic judgment was that copyright owners are in the best position to figure out whether their stuff is online online and it isn't authorized. Internet service providers have no way to know whether uh that file is there with permission or without permission. whereas internet service providers are in a better position to take stuff down or block it than anyone else. And so that struck people as an acceptable deal. >> Thank you. Um before I turn to Meredith, I think worth pointing out that there actually is no adjudicatory process involved in sending a notice. Correct. It was just a an allegation of infringement that then >> Right. That's what made it uh easier because before this to get something taken down you had to sue for infringement and that took seven years and $7 million. So in that sense copyright owners were better off because they got an expeditious way of taking stuff down. On the other hand, no one imagined uh today's internet. If you're thinking well copyright owner is going to see a file and send a notice to Google which is going to look at it and says yeah that looks infringing. Uh that just didn't scale. So many many people are relying on automated processes and one of the problems with automated processes is there's no incentive to set them to be less touchy than you might want them to be. So they have pretty much been overincclusive and uh involved the blocking of a whole lot of stuff that is not in fact copyright infringement. Turns out uh risk aversion is hard to code for I suppose. Um and also thank you for you know bringing in church of Scientology. You always know you're in a good panel and that is part of the response to question number one. Um all right turning to you Meredith. uh can you help us elaborate on the mechanics of notice and takeown processes under DMCA? Like how do they work? How do uh the entities handle notices they receive? How do various sizes of entities deal with compliance? Uh how have these online service providers adapted to the requirements of notice and take down beyond just complying with the letter of the law? Um I think Jessica teed that up just a bit, but why don't you fill us in a little more? >> No, she teed it up beautifully. Um, so keep in mind, uh, there's what the law requires and then there are sort of private ordered solutions that are often stacked on top of this. So if we're just looking at the letter of the law, um, essentially it's sort of a back and forth system of notices. Um the reason that this came about in the first place as Jessica mentioned is that there is this potential for secondary liability uh on the behalf of platforms who you know courts essentially will secondary liability is the sort of court- created doctrine where courts will go well you're not the one who did the infringing but you're sort of one order down the chain and we think that there's some sort of secondhand liability that you can pick up from your failure to act in certain situations. So in order to uh sort of cover themselves from that um platforms take advantage of what's called a safe harbor. And the safe harbor essentially works like this. If you uh are a copyright holder or someone authorized to act on their behalf, you will then send a notice to the designated agent um for that website or that platform. Um the designated agents are all listed on a website, I believe, maintained by the copyright office. You have to update your agent every so often. Um and then you have to send a notice which has a list of sort of formalistic requirements. You know, how do you contact me? Here's the work that is um that we believe is being infringed upon. Uh information that will allow the online service provider to be able to identify the infringing activity that you are referring to. This can often become very contentious because sometimes they'll just say um copies of my song and then if you're YouTube, you go, I have several trillion videos on my platform. this isn't helping. Um, but essentially you send that notice and so long as the platform uh I believe the phrase is expeditiously uh acts expeditiously to remove or disable access to the material then they are insulated from potentual secondary copyright infringement liability. Um then there is a second half to it which is implemented less often uh by a lot of platforms. Um, and the idea behind that is that theoretically the platform could be liable to the person who posted that in the first place for taking it down. Now, in order to insulate themselves from that, that's when you have the sort of notice and putback provision. Um, which is the opportunity for the person who uploaded it to go back and say, "No, no, no. This is, you know, I have the rights to upload this, or I believe it's a fair use, or the upload is kosher in some way, shape, or form." Um when the platform gets that counter notice, they have between 10 and 14 days to put that back um they have to notify the person who filed the original complaint and then the person who filed the original complaint has until the end of that window to bring it to court uh and file a lawsuit and notify the service provider that they have done so at which point it all gets punted out to a federal court. uh because that counternotice provision is really about protecting the uh platform from potential liability against their user who uploaded it. That is not as widely adopted. Um certain websites such as Twitch straight up do not have uh counter notice uh provisions or at least have not had them at various provisions in their history. Um and so that is kind of touchandgo. um failure to reinstate you know we get into conversations about risk aversion you know as Jessica and Stan have both mentioned um now again because this is you know this was designed in the late 90s under an assumption that websites and service providers were run by a small team of guys uh you know there was there was a person who was reading the email and analyzing it and thinking about it and investigating it and going hm we should take this down uh that doesn't scale very well and so what a lot of the larger platforms in particular YouTube is kind of the example that most folks think about. They have created these sort of private ordering systems that exist on top of the law um which really kind of prevent people from or or act as a sort of a buffer between the actual user experience of managing copyrighted works and actually invoking the DMCA. Um so YouTube for example has content ID which is a sort of private internal system which uh does a million in one things all of which somewhat poorly. uh which you know is designed to identify uh it's allow it's designed to allow copyright holders to upload samples to upload their work so that it gets fingerprinted and is scanned and they can send uh you know they can demonetize videos they can have takedowns they can strike things sort of back and forth and back and forth um but those are not strictly DMCA systems I think a lot of folks get them conflated because they're dealing with copyright management on these large platforms but going through that is the step often in practical cases before you end up invoking the DMCA. Some folks decide to skip them altogether, especially if you're a very small creator. You do not have access to the full suite of tools that things like content ID allow you to use. Um, and so oftentimes people will just go straight to DMCA takedown notices. Um, but you do end up with this very complicated layered kind of system across these different ecosystems. Um, each of which is often bespoke to the different um, sort of needs of every platform. And in some ways that's a good thing. Um because one of the problems that we have run into a lot uh in the 27 years now that we've been dealing with DMCA takedowns is bad and abusive takedowns. Um there was a statistic I believe from 2017 where uh Amazon's Kindle Direct Publishing uh which again this is 2017 this is before it's kind of at the scope that we know of today um commented that out of all of the takedown notices they get for their self-publishing platform about 50% of them were just attempts by rival authors to knock uh self-published books out of the rankings. So a full half of them were fraudulent um and were specifically designed as sort of a keep your name my name out of your mouth uh kind of you know gaming the competition within the system. And so every kind of platform is going to have a different threat model for the kinds of misuse of notices. Um and you know the ability of platforms to kind of develop their own systems overall is a good thing. um even when it leads to situations like Twitch where there's no sort of putback mechanism that they necessarily honor um because they're not super worried about their users being the ones to hold them liable. >> So I want to get you to say a little bit a little bit more about uh about that piece in particular. Um, in uh for the legal nerds watching, uh, DMCA section 512F is at least part of the anti-abuse mechanisms built in here. Meredith uh, and or Jessica, could you say a little bit more about just like what uh, what has been the fate of 512F? Do people use it and does it work to actually prevent abuse of this existing notice and takeown system? Uh functionally I think the legal term is dead as a door. Um it's uh yeah it has been litigated before. Um it was I believe it was the um Lens v Universal was the uh the Prince dancing baby case uh where for folks who are not familiar with it. This is a very pretty famous case. A woman uploaded a video of her toddler dancing in her kitchen. And in the background, you can hear a couple of bars of uh Let's Get Crazy by Prince. Uh Prince's estate um andor Universal uh took issue with that and had it taken down off of YouTube. Um she had it put back and then we got there was a back and forth. I don't remember the exact sort of litigation posture of it in the early stages, but one of the claims um that uh Lens brought represented by the Electronic Frontier Foundation um was this claim under 512F which essentially uh ostensibly punished sort of misrepresentations and takeown notices. Um, and this became this big litigation question of whether a rights holder was legally obligated to consider things like is this a fair use or could they just sort of fire indiscriminately when they were issuing takedown notices. Um, and the answer was kind of maybe. Uh, Jessica can definitely give a little more gloss on that than I could. I mean, I think the court held that yes, you have to consider fair use, but one of the problems with 512F is it doesn't really have a remedy. So, it was also showed up in a suit that Debold brought way back early uh when a whole bunch of college students at Sworthmore posted some Debold internal documents that pretty much revealed that its voting machines were hackable. uh and so it uh filed DMCA notices to get that stuff off the internet and ultimately uh was found liable under 512F for uh having misused the notice and takedown procedure. But one problem with that is that you don't collect enough in the way of damages under 512F to make any lawyer uh but a real public servant willing uh to take the case. It it's more of a slap on the hands. The other problem is what you mentioned. Most big service providers don't use 512. They use their in-house automated systems and what uh so when Mega Upload uh posted a song, it got taken down by uh I think Universal and Mega Upload said, "No, this doesn't violate any copyright of anyone. It's an original song. you're just upset because your artists are singing it and brought a 512 F action and Universal and Google said, "Whoops. Uh, there was no 512C takedown notice. We didn't take it down using the notice and takeown procedure. we used content ID and that's completely private and there is no remedy uh for abuse of our private content ID system. And so another lovely story is new NY NYU's Angelberg Center uh did a webinar uh about music copyright infringement in which it had two musicologists uh test tell the people in the webinar about their methodology and how they decided that songs in some famous copyright infringement cases were were not infringing. uh the copyright owner uh in those songs used content ID to take down their webinar and there they tried to use the internal uh putback device and were completely unsuccessful. They only got it restored because some of the faculty members of NYU Law School knew some of the people in the general counsel's office uh at Google and made a phone call. But the Google putback device is just completely ineffective for normal human beings who don't have a former student who's a lawyer with the company. >> All right. Thank you both. I think we've done a pretty good job of establishing how it started and how it's going um here in the United States, but let's uh turn our focus uh abroad for just a minute. Uh Paul, I'd like to start with you and then Rian, you may have some followup on this as well. Uh what what legal regimes have jurisdictions outside the United States chosen for intermediary liability? Um, and is there a similar sort of split between requirements for liability protection dealing with copyright versus non-copyright claims? >> Thanks Dan and really happy to be here to discuss this exciting topic. I also came to realize that you know it's almost 30 years since it was adopted. So obviously there is a lot to unpack but I think it's great to start by just noting that DMCA was the first uh historically the first legal act of its kind and it obviously had global effects in shaping the legal frameworks around the world and it set the foundations not only for laws but also for business models hash content ID and also fostering debates among policy makers, lawyers and practicing attorneys as well as academics. So outside of the United States, countries have adopted many framework different frameworks let's say for intermediate liability and they often try to balance uh different policy goals such as harm prevention, free speech or speech protection and innovation. And uh briefly I would like to highlight three possible variations that maybe are worthy of attention uh with regard to frameworks outside of the United States. So the first one is a actual knowledge or awareness approach adopted in countries like Australia, India, Japan and Philippines. Briefly put uh intermediary liability depends on whether the intermediary has actual knowledge over the illegal content. It means that intermediar is required to promptly act typically by removing access or disabling access to the content once they are actually aware of this illegal content. And maybe Japan could be a great example uh of this framework. So in 2001, Japan adopted a provider liability limitation act uh which uh under which uh intermediaries are not liable uh for damages caused by third party information unless it is technically feasible for them to prevent the transmission and they must have either actual or constructive knowledge about the infringement. The second framework uh from foreign jurisdictions could be defined as let's say PR takedown sorry PR notice and takeown system. So in New Zealand and South Africa so far as I know uh there is like a framework which requires intermediaries to promptly remove the content once they are notified and failure to do that means that the intermediary loses the protection and the third one is probably most prominent in Europe it's called mere condiate approach adopted in Europe and now also followed in the United Kingdom but the idea is that interers are seen as passive passive actors, passive conduits and they are exempted from liability as long as they do not initiate, select, modify the content. And uh here the e-commerce directive of 2000 distinguishes between three levels of intermediary uh responsibility mere conduit, catching and hosting. depending on the intermediate involvement liability let's say or duties might increase but uh but the idea is also similar that intermediers have to act expeditiously I think it's worth noting and it's probably a segue into this AI debate but there is a new act uh uh digital services act which follows the path established by the e-commerce directive but also adds additional DSA acts additional obligations on actors, intermediaries, platforms in terms of transparency and reporting about the content they host. And to your second part of the question, Stan, uh there is obviously very clear bifurcation between copyrightable or copyrighted content and then other type of content. Another big Pandora's box is privacy and free speech uh type of infringements. So if I may summarize obviously it's overstatement uh but I think from a legislative perspective we can see that regulators around the world deem copyrighted content as less harmful type of violation than privacy invasions. So in case of privacy violations, privacy invasions, uh platforms are more induced to respond promptly and there is a greater let's say enforcement opportunity for individuals to take down the content. It's different framework probably we can identify similar patterns of notifying the platform but I think privacy rights privacy related violations are more uh important from the regulatory point of view and are worthy of greater protection. It's probably overstatement but I think it's worth making this point. Thanks. >> Thanks Paul. Uh, Rian, anything to to add to that from a international perspective? >> Um, I'm sorry. I I jumped out of the call for a moment, so hopefully I'm not um repeating anything that Paul said, but um I just wanted to add one in kind of uh recent update from the United Kingdom, which I'm sure everyone has heard about in terms of the online safety act um which you know does not pertain to uh copyright infringement, but you know, we're going to be discussing a bit later on about non-copyright um re related liability and um the online safety act concerns intermediary liability in the UK. concerning userenerated um illegal or harmful content and and actually imposes a duty of care on platforms um particularly in relation to protecting children from inappropriate content. So I think that this really um um moves into a platform responsibility for non-copyright um harms which I think has been like quite an interesting and controversial um update from the United Kingdom. Could you say just a few words to compare the duty of care sort of approach compared to the notice the you know providing some point of knowledge through a through a notice system just just briefly outline what that looks like. >> Yeah. So um platforms are really expected to be proactive. So this is including um content moderation. Platforms are really required to implement systems and processes to identify and remove that legal content. So I think that kind of uh creates a burden on um platforms really to take on that responsibility. >> Thank you. Um I guess it might be fair to characterize that as sort of a an assumption of some form of knowledge on the platform's port and then they're already required to take action there rather than having the notice come from >> uh a user for a specific instance of something and then they respond to that. Um so let's I think we'll we'll move sort of away from copyright slightly here uh with the next question. Um this year in particular but um not not constrained to this year uh we've seen an expansion in the sort of interest in and logic of notice and takeown of other kinds of notice based frameworks uh for conditioning intermediary liability protections. For example, the recently enacted Take It Down Act and then the recently introduced No Fakes Act. Uh both pertain to non-copyright material per se, but do involve a a similar notice and takeown system. Uh what does the use of this kind of framework for non-copyright legal claims show about either the success or popularity of the DMCA framework or whatever else it may say about uh that framework? Um, and how will or won't this approach to uh online copyright enforcement translate into non-copyright contexts? Um, Rihanna, do you want to kick us off and then I think this is probably a question that everyone might like to take a piece of. >> Yeah, hopefully I can kind of provide a bit of a background on the act um and then that can kind of start off the conversation. Yes, >> please. Um but yeah, the take take it down Act provides a really helpful example of how notice and takeown framework translates into a a non-copyright context and um it regulates the non-consentual distribution of intimate imagery or NDI um and it was passed by Congress as as you mentioned this year and it's it's the first federal law that's regulating um AI generated content. So, it's quite, you know, um it's it's a first of its kind and and the take it down act is really supposed to fill in gaps in the resources available for victims of NDI um under the DMCA as well. Um notice, you know, the the notice and removal provisions under the DMCA are available for those who have been affected by D NDI, but only where the individual of course owns the copyrights. This is primarily selfies. Um so, the take it down act addresses that gap. um when the individual doesn't own the copyright. So this might be when there is a synthetic depiction of the individual created by artificial intelligence for example. And and the legislation um specifically includes uh a definition um where the deep fake realistically depicts the individual such that the that a reasonable person would believe that the individual is actually depicted. And then maybe we can go into a bit of a conversation later on about that um about concerns over that definition, but the the the act really requires covered platforms to remove NDI within 48 hours. So it it models the um DMCA's section 512 and um it requires platforms to create a process through which victims of NDI can send notice to them about the existence of such material. But the the issue with the take it down act is that it's not it hasn't included any of the DMCA's safeguards. So, you know, we we've spoken about this and and in particular Jessica and Meredith have have gone over um the provisions for the counter notice. Um there's also um individuals must attest under a penalty of perjury that they are authorized to act on behalf of the injured person under the DMCA. Um the DMCA imposes liability on individuals who make knowing and material misrepresentations about the um the nature of the content. But under the uh take it down act we don't we don't have any of these. So there's no requirement for an individual submitting to a test under the penalty of perjury um and instead any any removal notice about an intimate visual depiction may be submitted um by a person claiming to be the depicted individual or their representative. So there's there's no legal consequence for impersonating someone or falsely claiming to act on their behalf. Um there's no not counter notice provision. Um you know we we've gone into a bit of detail about whether or not that's been a success or not, but um uh individuals also cannot claim against platforms under the Take It Down Act for the good faith disabling of access to or removal of content. So there there's no recourse against platforms. um even if that visual depiction is later determined to be um lawfully published and and this this good faith standard is rather vague um and opens up opportunities for misuse. Um also the law requires only a reasonably sufficient identification of the content. Um and really you know I think that under the DMCA this level of ambiguity um would likely invalidate a request. So, you know, in in summary, we're we're seeing the DMCA framework um implemented in new ways in non-copyright context, but it's kind of a far far weaker imitation of of the original. And, you know, we've we've all seen that, you know, even with its guard rails, the DMCA has successfully um been used to remove protected speech. Um and when we don't have any of those safeguards at all, I think this raises some um real concerns about how the take it down act is going to be implemented. >> Indeed. And uh no fakes I think is similarly situated. Um I I do want to move us on to questions from the audience, but I also want to hear from Jessica, Meredith, and Paul on take it down or no fakes before we before we skip ahead. Well, I think what made the notice of takedown regime attractive here was that it's administerable. Indeed, most of the online service providers already have a system in place to do this. so that to the extent that they uh needed to sign on in order for this to get through Congress saying, "Okay, but let's use a system we're familiar with and we have and we've implemented." All of the safeguards in the DMCA were achieved because of people negotiating on behalf of other folks who worried their speech would be suppressed or worried that uh uh there would be wrongful takedowns. And to the extent it's really hard for someone to stand up and say, "Hi, I want to post uh naked selfies of my girlfriend without her permission. please give me uh the safeguards to allow me to do that. Um there wasn't anyone pressing for the kinds of protections uh for legitimate speech that people might want to take down to cause Congress to say, "Okay, we'll incorporate those safeguards into this, too." >> Great point. Uh Meredith, any comment on either of these? I'll note that 48 hour takedown notice uh notice to removal is pretty short. Um >> it's pretty fast. Um, one of the other things that we've seen, you know, and this is part of a general trend is that ever since the DMCA was passed, so for the last, you know, upwards of a quarter century, um, you know, I think content holders would prefer, uh, that we move to what's called a notice and stay down, uh, regime, which is essentially this idea of once I have it taken down, you must now proactively filter so that it never comes up again. uh which is a bit of a pipe dream technologically for all but the absolute largest of these platforms and even then uh everybody's got a story about content ID. It's the most expensive and high-end system out there and it's still kind of a disaster at times. Um and so take it down I believe has a notice and stay down type framework in it. um which is only sort of administerable because there is a um there is sort of a mechanism that is used um for CESAM which is um uh is I don't remember exactly what the acronym stands for child sexual abuse uh imagery uh or material um which is that there's a a sort of there's a technical workaround involving hashing and the Nickmick database that they can they can glom this on to um one of the concerns though as somebody who works in copyright policy is We see a lot of situations where this happens and you had this notice and stated in a much narrower context. Um, which is, you know, again, you're you're dealing with a a semi-nown universe of potential problems and a semi-known universe of files that you need to keep an eye out for. Um, copyright because it is so extensive and applies to anything that is sufficiently expressive and embodied in a tangible medium of expression. Um, you cannot that that is the entire universe essentially. And so when we start seeing um notice and stay down mechanisms, uh one of the things that we try to keep an out eye out for is you know is this going to be used as proof of concept that actually we can do this for the entire universe of copyrighted material and the answer is no. But that doesn't stop people from saying well maybe you need to nerd harder at the problem. >> Um yes, nerding harder is quite often part of the solution implied in many pieces of legislation these days. Um, I will just note before passing it over to Alex for audience Q&A that the no fakes act um also requires a stayown type of thing and uh strongly implies that uh online service providers should build their own hashbased matching system and filter uh to prevent the upload of identical copies of deep fakes I suppose. Um, so lots of uh lots of nerds to be employed in the future should that become law. Uh, over to you Alex for questions from the audience. >> Amazing. Well, on the subject of the no fakes act, uh, let's start with this one. Uh, no fakes includes language that seems to create a right of publicity that is very similar to to a copyright. How well would or wouldn't importing the logic of substantive copyright protection into a right of publicity work? So I can I can hop on that one. Um one of the so the fundamental aims of copyright and rights of publicity are different. Um, so the idea behind a copyright is that you or between the American conception of the copyright, I'm speaking at a very very broad level here, but the American conception of copyright is a policy decision that a creator has put effort into creating a work and in order to allow them to potentially recoup the cost of creating it in terms of their time and their energy, we will give them a limited monopoly for some amount of time. But it's it's designed specifically to incentivize the creation um the production of more works that are then made available to the public in exchange. Um right of publicity and name image likeness rights which are sort of cousin a cousin framework to right of publicity are um in the United States generally only available to folks whose uh likeness has commercial value. So they are available to um you know the high high-profile cases are famous actors, singers um you know sort of folks like that. I do not have rights of publicity. Uh I think most people on this panel um no shade to anybody involved but and most of us on the panel are not going to be recognized on the street by our faces alone. We're not cutting licensing deals for ads. Um, and so the idea behind publicity rights is that if you have a likeness that is valuable and someone goes and uses your likeness uh without your permission to make it appear as though you were endorsing a product that they don't just kind of get away with that scot-free. They do owe you some financial compensation for it. Um, again, this is like grossly oversimplifying it. Um, the problem that happens when you create when you try to scale these publicity rights is that when you're you're creating an all-purpose tool in the process of trying to create something to get these things taken offline, right? So, the question is, well, if you're a famous actor and you want your uh likeness or a deep fake of you removed from YouTube, how do you do that? Um, well, there needs to be some kind of mechanism to do that. How do you restrict that to only people whose likeness is commercially valuable? Do do you have a content ID situation where if you're a super special rights holder? Um, you know, if you're the special boy that uh YouTube decides you get to have access to the special control panel or whatever. Um, no fix creates a sort of general purpose tool that is available to anyone. um including folks like me who I when I if I there's a deep fake of me out there. It's a I have a very different threat model and a very different damage model for what that's going to look like than if it's Brad Pitt. Um I'm not losing out on licensing revenue. I'm worried about somebody deep faking me either into non-consentual intimate imagery, which is things like take it down um or deep faking me into doing something embarrassing that would potentially be like a defamation claim um if I was to pursue it in court. Um, and the other thing that we've learned, again, as at my local angle as somebody who deals with DMCA takedowns, is these things are abused like crazy. Um, and right now we do not have the technology to sufficiently detect whether or not something is an AI deep fake or not. Um, and so the second one of these tools comes out that says, you know, you are now hereby legally uh empowered to demand the removal of something that is a deep fake of you, uh, there's no way to reliably distinguish between deep fake and real footage. And so the very first thing you will see this being used for if it becomes law is taking down embarrassing footage of politicians. It will be used to take down videos um of police brutality or public misconduct. You know with the claim that oh this is absolutely a deep fake of me. Um and no fakes has no no putback mechanism. You know we sort of discussed this in the DMCA. There is a putback mechanism. It's not a particularly robust one and a lot of people ignore it. Um no fix doesn't even have that. It's it's once it's been taken down it needs to go down. needs to stay down. Um, so it's a very very dangerous framework because it's been designed explicitly for um for actors, for celebrities. It's been designed with this economic model and not for the rest of us. Okay, amazing. Uh, moving on to our next question. Um, has there been any legal push back to the automated content recognition systems used in DMCA compliance, which I assume means AI and and automated uh processes. Um, so there's a fair amount of scholarship now, legal scholarship and technical scholarship documenting the problems. Um there isn't really any legal cause of action uh because the notice and takedown in the DMCA is uh styled as a safe harbor. No one has to comply with it. It's just what service providers can do if they want to avoid liability. So failing uh to uh provide any of the elements of it isn't independently actionable. Uh all it is is a shield from liability. So, while there's lots and lots of scholarship documenting tons of problems, uh the only case I'm aware of is the case against uh the case brought by Mega Upload uh for wrongful takedown and that got dismissed because the takedown was not pursuant to uh DMCA notice. All right. Um, so another question. Um, with the new administration in the US utilizing the law somewhat haphazardly and with an eye towards results rather than focusing on the process itself, how do we view triaging takedown requests? For instance, if a site gets to request to take down information related to DEI from the executive branch, but it's not necessarily within the executive branch's jurisdiction. So technically, if the request is for anything other than a copyright related uh infringement claim, it's not considered a valid notice. So this is something that is baked into the DMCA. Um, no, as a practical matter, uh, what actually often ends up happening, as anybody who's followed YouTuber drama can attest. Um, what usually happens is it's framed as a copyrighted claim, a copyright claim, and really there's sort of a secondary motive for that. Um, having said that, websites, especially smaller websites, do have the, you know, a lot of them are still using individual review. Um, you know, we work a lot with the organization for transformative works, which runs archive of our own, the the very large fanfiction archive. Um, they manually review their takedown requests. Uh, they don't get a ton because they're a fanfiction archive. So, people sending DMCA takedowns to fanfiction are kind of wild to make uh, conceptually, but they do get some and they they manually review them. Um, you know, and a lot of them are not valid requests. Um so to that extent on an implementation level you know websites do have some choice in how they want to review this thing. Um and typically if you get political pressure to remove content from a website it tends to come through other avenues rather than a DMCA request. All right. Uh next question. Does the DMCA or case law pertaining to the notice and takedown regime address the issue of enabling the heckler's veto through this regime? I mean, I think that's what the 512F uh cause of action for misrepresentation is trying to do. Um the reason it's been ineffective is largely that it doesn't include the kind of remedies that make it financially feasible to litigate. Uh so it's uh uh an option that few people are motivated to use because it's just too expensive. But that's uh indeed baked into the structure of of the DMCA. Amazing. Um, so next question. How does the fact that AI generated content isn't protected by copyright affect notice and takedown regimes versus no fakes that has a textual uh recognition of AI generated content? >> Maybe I can take this. I think actually there is a very big debate on I think gray zone in terms of what is protected and not protected by copyright and it's a long conversation but I think uh it has been oversimplified by the US copyright office uh by stating which was stating that you know if you use AI tool it's not copyright protected. I think it's a very narrow perspective and it does not reflect how real creators use AI tools. Usually it's a collaboration with multiple tools, traditional digital tools and modern AI tools. So I think uh it's very let's say it's very contested position by US copyright office just to suggest that using AI tools is not necessarily or necessarily leads to no copyright. So that's one big problem and I think the big challenge is that okay so if we have a uh content that is created with the assistance of AI tools uh how do we identify you know the copyrightability how does it fit within the DMCA framework I guess it's a topic for the next h next webinar but in terms of uh I would like to address Meredith's points uh on no fakes and the deep fakes I guess yes notice and take framework is really needed for individuals, not necessarily who are celebrities, but each and every one of us. We need to have tools to be able to request platforms to remove the content that maybe is impersonating us, not only our visual appearance, but also it can be our voice, right? It can be other biological physical traits that we might have. So notice and take down framework or similar framework is needed for all of us. But I would say that we probably need to go one step forward and establish a bit more clearer framework that actually it must be like some sort of opt-in framework so that third parties that are using AI tools to create digital replica of Jessica who I would argue is a celebrity right but also me and myself right if someone is trying to impersonate me I think by default we should see that it's our inherent right to give a ex anti-consent for third parties to create replicas because in Silicon Valley you can imagine there are many companies that are making digital rep replicas not only of celebrities that could be your shopping assistant but also impersonating anyone right any individual. So we need some kind of a new framework uh new contract with technology that people actually should be given the opt-in rights for third parties to create their digital replicas. So it's kind of a new framework we need to think of not only notice and takedown but what happens before that can someone take my voice and and use it >> for different purposes. So I think it's important that that framework not be modeled on copyright. That is the thing about copyright law not only in the US but also in Europe is it's really designed to encourage people to alienate to transfer to sell their rights to someone else. And I think one thing we've learned about right of publicity regimes is as soon as this right is assignable, it start co starts causing all sorts of trouble. So there are for sure companies in California right now who are hiring um actors to do voice work and asking them to sign a uh all rights in those recordings of their voice so that they can use them for AI. I think we also need some opportunity uh to take back assignments that we've made. Um and that it needs to be a right that if we have it isn't something we're encouraged to sell uh when we need the money and then we don't own it anymore. So that requires a kind of thinking that is less based in intellectual property and I think more based in tort uh and in dignitary considerations. >> I will I completely agree. I will just piggyback on I think the reality of why why just contextually sort of in the US in particular why we've seen a lot of these ticket copyright framework comes partly from the fact that the first industries to jump into action in response to this were major entertainment industries. And when you have a hammer kind of everything looks like a nail. Um and also uh the complication posed by section 230 um which is again for non-American I'm grossly oversimplifying. Um section 230 essentially says that platforms generally speaking cannot be held liable for things that their users put on the platform. Um and there's a handful of exceptions to that and one of those is intellectual property. Um, every time Congress tries to punch a new hole and add a new exception to 230, it turns into an absolute circus. Um, and so the political reality is that most folks see that the only way they can get any kind of takeown mechanism. You have to find one of those existing tunnels through section 230 and intellectual property is one of them. And so that is they have been funneling it through that as a matter of political expedience. >> Amazing. And I think we have time for one last question. Um, bit of an open-ended one, but I I I think an interesting one to to end on. Uh, have there been any proposals for alternate regimes to the notice and takeown regime? Uh, well, we've seen proposals for notice and stay down. Um, that's been the that's been the main contender. Um, again, you run into a lot of technological problems. um some of which were foreseen pretty aptly by Congress in the late 90s and many of which were not. Um I think that's been the big one. Um you know again even speaking as somebody who spends a lot of time dealing with notice and takedown abuse I think there is a very begrudging near consensus in a lot of folks on the policy side of this that notice and takeown perhaps not as instantiated right now. I think we've all got complaints about the way that it actually works, but conceptually ends up being kind of the best solution to a otherwise pretty much intractable problem. Um, and the question becomes really how do we reform that really more than how do we toss it and build something entirely different? Also I can add in the age of AI with with this advent of new let's say AI assisted AI generated content clearly there are efforts to create new frameworks that are already AI powered frameworks to manage what gets out and what gets noted what gets uh let's say flagged clearly there are companies working on that I think it will take years to make it into let's say something actionable but uh the technologies are moving so fast it's obvious that we cannot really cope uh with that much AI assisted content with existing frameworks. So we need to come up with new technological layer to let's say organize the content. >> I think Paul's right. I'm going to jump in to close this out here. But before I do, I'll just observe on on Paul's point here. you know, talking about content ID as a hashbased matching system uh sort of built on top of in addition to the DMCA requirements. I wonder if those type of systems will work less well against uh sort of a flood of novel images, we'll call them that, right? You can you can generate easily any number of variations of a of a deep fake that will not have the same hash as each other. Um, and so it could be quite difficult to automate a system uh at least based on that technology. What uh what might fill in that gap uh I think is an open question at the moment. Um so you know tune in again next summer when we're talking about the latest and greatest from uh the AI industry on how they're self-regulating this problem. I'm sure they will all get right on. All right. Thank you so much to our wonderful panelists uh Rihanna, Jessica, Meredith Paul. It's been a pleasure hearing from you all today. Thanks again to our legal fellows who did a wonderful job putting this panel together um and for organizing the whole thing. Um we welcome you back anytime. Um thanks for being with us today. >> Thank you all. Thank you. >> Thank you.