California's AI Transparency Act Takes Effect

Senator Josh Becker 00:00

Well, I thank everyone for being here. 3 years ago, I read a story in the paper which was about a family in San Francisco. They got a call from their son, and he'd gotten into a car accident, obviously very distressed, and he put them on the phone with his lawyer. They needed money to be wired right away, which they did. And it turns out, after they wired the money, they found out it wasn't their son at all. It was AI. And that started really a 3-year journey for me to address this issue and say we have to help people fundamentally answer the question, was something AI or not. And it's taken 3 years, but we're now at the point of being able to announce major progress here in California and also working with European partners around the world. With recent advancements, AI content, images, audio, video have proliferated, and that is going to continue. I can say, I can predict that will continue at a rapid clip. And while there are some benefits to AI and some good educational material will come out of that, there's also - AI is being used for election disinformation here and around the world. It's being used for scams. It's being used for abusive deepfakes. And I am committed to fighting for AI transparency for Californians and for citizens around the world. At midnight on August 1st, California started implementing the nation's first comprehensive AI transparency law. This is the beginning of a foundational change of how content is represented on the internet. Most people won't notice dramatic changes overnight, but underneath the surface, AI-generated images, video, and audio will have embedded content: a nutritional label that's embedded now in the content that helps people understand where that content came from, who created it, and has it been manipulated using artificial intelligence. The goal is to give Californians the information that they need to decide what to trust. We've spent the last few years making sure this bill is technically feasible, that the companies can implement it and will be expected to implement it and comply. And they're listening. OpenAI just sent out a press release on Friday and said they are marking both audio and video. They're already marking images. Google is complying. We expect the others to comply as well. Adobe is enabling this compliance as well with their technology. And as AI-generated content becomes increasingly realistic, consumers deserve the ability to distinguish what is AI and what is not. Again, the goal is to protect people from scammers, protect against election disinformation, even help journalists determine, you know, what - verify what is true, what is not true. Transparency is essential for our new economy. Again, AI-generated scams to some - I've obviously focused a lot on scams, well, with the DELETE Act, which also went into effect this weekend. It was a very exciting weekend for me.

Eddan Katz 03:18

Thank you.

Senator Josh Becker 03:19

Criminals can clone voices, as I mentioned that story earlier, fabricate convincing videos. There was a story about a video, someone got a call to what they thought was a Zoom meeting and got scammed out of millions of dollars. They can impersonate family members or public officials in ways that simply weren't possible a few years ago. Our laws need to keep pace in order to protect our citizens. Under this law, every public prosecutor in the state will be empowered to pursue enforcement. And this is also why I authored SB 1000 this year, which strengthens the original law. The original law had a threshold, you had to have a million users or more before you had to comply. SB 1000 removes that threshold. Anyone who puts an AI system on the market in California, so I don't care if this is a Silicon Valley developer or a developer out of China, if the AI system is on the market in California, they are expected to comply. And then on January 1st, thanks to a bill authored by my colleague Buffy Wicks, which I co-authored. All online platforms, all social media sites will be required to surface that embedded information. So we're giving them a number of months before they - again, the social media sites will read the embedded information and have to surface that to visitors. One thing that's very exciting, and we'll hear about this a little bit more, we've worked closely with international policymakers, including officials that are developing the European Union's AI Code of Practice. Practice.

Senator Josh Becker 04:48

We think this is one of the first times that a European Union regulation, certainly around AI, came into effect the same day as a California regulation, and that happened this weekend. In the absence of federal leadership, California must step up and help shape AI governance. Washington State has passed a nearly identical law based on my bill. New York State is debating one right now, which will hopefully follow shortly after. But this work isn't done. I'll continue to advance the frontier of AI regulation, working with our partners and working with the European Union. And we need legislation that moves with the agility that the companies in my district are moving. With that, I'll introduce a couple folks. A couple of them will speak here today and the rest be available for questions. We have Berhan Yambelu from Harden AI. This is a UK firm focused on helping companies comply with this bill. We have David Evan Harris, we'll hear from in a moment, from the California Initiative for Technology and Democracy, and the chancellor's public scholar at UC Berkeley. And then Eddan Katz, head of policy and government affairs for Encypher, which is one of the authors of C2PA, one of the leading standards that we're relying on, on all this. With that, I'd like to turn it over to David Evan Harris. Thank you for being here.

David Evan Harris 06:07

Thank you so much, Senator Becker, for the invitation. Without knowing what is real and what is fake, we cannot have a functioning democracy or society. For Californians and for the world, this bill is a critical step forward in ensuring transparency of AI-generated content online. I teach about AI ethics and policy at UC Berkeley, but today I'm here in my capacity as an advisor to the California Initiative for Technology and Democracy, or CITED. With the CITED team, I've been working since late 2023 to advance legislation about AI transparency and provenance in the state of California, along with my incredible team of colleagues including Ken and John who are here today. Um, we owe an enormous debt of gratitude to Senator Becker, Tom Kemp, the executive director of the California Privacy Protection Agency, who was an integral part of pushing forward this law, and to Assemblymember Buffy Wicks, with whom we worked very closely on AB 853, also part of the California AI Transparency Act, or CAITA. I also owe a big debt of gratitude to Ian Higgins in Senator Becker's office, who has gone very, very deep into the technical details of everything that we're doing here today. And yes, this is a historic day. In all the research that I put together, I could not find any other instances of a California law and an EU law coming into force on the same day in coordination. And this is not a patchwork. This is the opposite of a patchwork. This is two of the most competent legislative bodies, regulators in the world, coming together to put forth a solution to a critically important problem. While Washington is abdicating responsibility for this, California and the EU are stepping forward. I also want to make clear that regulating tech does not mean hindering innovation. This is something that we often hear from critics in Silicon Valley, but it's simply not the case. Throughout this process over the last 3 years that we've been working on this, we have seen time and again companies stepping forward to support this legislation. Legislation like this creates an environment, a set of guidelines that create a market themselves for companies to step. And we've, we already heard about Google, Adobe that have worked hard on this, but there have also been a series of smaller companies, TruePic, Steg AI, Digimarc, and now Hardin AI and Encypher that have made the long journey to hot and sweaty Sacramento in person to express their support for this work. And that is evidence that this is building a market for these technologies that will solve such critical problem. So thank you again to Senator Becker and his staff for being here, and I'll hand the floor back to you.

Senator Josh Becker 08:57

Thank you.

David Evan Harris 08:57

Thank you.

Senator Josh Becker 08:58

Thank you. And again, thank you to the CITED team for unbelievable dedicated and difficult work over the last 3 years. And now I'd like to bring up Eddan Katz, as I mentioned, the head of public policy for Encypher. Please join us.

Eddan Katz 09:14

Thanks very much. An honor to be here. I'm Eddan Katz, head of policy and government affairs at Encypher, we build technology that lets anyone confirm where a piece of content came from, who or what made it, and whether it's been changed since. We do that with an open industry standard called C2PA. Think of it as, has been mentioned before, a tamper-proof nutrition label for digital content, cryptographically sealed so it can't be faked. That matters today because as of yesterday, CAITA is in force. For the first time, the public can identify AI-generated content from chatbots right on the websites and platforms they already use and verify authentic journalist-produced content on the sites that support it. And we've made that verification 100% free with no account required. Encypher checks the image with Content Credentials. It checks for the signature, the source ingredients, and the asset hash, this verification runs across everything: images, video, audio, text, and documents like PDF and Word. To get here, we've been integrating this technology with several of the world's largest news media companies to help build, rebuild trust in authentic journalism, and with leading AI companies for compliance with California's new law and the EU AI Act. As aligned with the EU AI Act, by streamlining the marking of content, we have set down the substrate for the global governance of AI. Transparency shouldn't be a premium feature. Starting today, it's free for everyone.

David Evan Harris 10:50

Thank you.

Senator Josh Becker 10:52

Thank you. Thank you, Eddan. I love that, uh, that visual of setting down the substrate, um, and I was also going to pick up on David's comment of the weather. Things are certainly heating up in Sacramento when it comes to AI regulation. We know the technology is advancing rapidly. Of course, we saw a couple of instances very recently of - I don't know if you call it AI breakout or what the proper terminology will be for hacking other systems. And clearly, we're gonna all have to work incredibly hard to bring the regulatory regime into focus. But one of the, again, one of the fundamental steps that I said I'd work on many, a number of years ago, is was something created by AI? How can we help people distinguish AI-created images, video, and audio from non-AI generated. And that's what really, what we're here to accomplish today with our EU colleagues. It's a great, a great moment. So with that, again, also join my thanks, we mentioned, say it again, also with Tom Kemp, who before he was head of the privacy agency worked on this legislation with me, as well as the Delete Act, which we've had 400,000 Californians sign up. So I think privacy and some of this AI regulation actually go hand in hand. So it's been an exciting weekend. With that, any questions? Any questions? Could you explain the enforcement mechanism and are you concerned at all that with, you know, things moving slowly through citations and litigation and things like that, are you concerned at all that the technology would evolve faster than Yeah, so the question is, are we concerned - no, I'll answer the second part first. Are we concerned about technology evolving kind of beyond the capacities of regulator, move faster regulation? That's one of the things I'm proud of for this bill. We created it to be adaptable, essentially, not to lock in to one standard, although we have C2PA right now, which is a tremendous standard for all the reasons that were mentioned. But we kind of built it to keep up with the technology. So that part I'm not concerned about. The embedded disclosures are tamper evident. It doesn't mean that someone can't tamper with them, but we can tell if somebody tampered with them. So that I feel good about. I also feel good about the regulation. I think our attorney general here, Rob Bonta, has been very aggressive. One of the things, we had the same questions around the DELETE Act and would data brokers register, And he went after them, and we've had hundreds of thousands of dollars of fines for data brokers. And there are substantial fines in this bill as well. So we expect the Attorney General to be able to go out after these companies if they don't comply. We will look in the future about, we're looking at other enforcement, additional enforcement mechanisms. For example, right now it's a flat amount, and should we tie that to the size of the company? I think the EU actually has some interesting enforcement pieces there that we can probably learn from. So, I think some of that will be evolving.

Senator Josh Becker 14:31

I'll comment that and then maybe invite And actually, maybe we'll do that and then maybe invite you guys to come in on the technical question. First, on the question that you raised, you know, we heard there's a big - at the beginning of the year, that was all the buzz, AI preemption. And there was an executive order. Not much has happened since then. So I think that I would see us continue to move forward. I don't see the federal government coming in anytime soon. I don't think a lot will happen this year. End of the year, and I think, you know, we'll be way on our way and the companies are already complying. So I think we're past that. I do get concerned about that on the privacy area and, and something we're going to have to, you know, continue to watch. Any comments on that? Yeah.

David Evan Harris 15:18

So I testified before the United States Senate about a year and a half ago on this issue, and I was very pleased to find senators from both sides of the aisle very excited about regulating in this space, and we have now years of proposals, over 100 different proposals for legislation in the United States Congress on various aspects of AI. A number of those cover this very issue that we're talking about today of AI transparency and provenance. The first big effort at blocking states from making laws was actually stopped by Marsha Blackburn, a Republican from Tennessee, known for their extreme liberal views, right? So she wanted to protect the rights of musicians. Tennessee is the home of the best-named AI law in the country, the Elvis Act, protecting musicians from impersonation. And I don't think we have to look too hard to see that this is actually one of the most popular issues to legislate on in Washington. And as these efforts have advanced, actually, we've seen the people responsible for them in the White House leave, and we've seen a huge shift in the White House's posture about this, especially since this recent episode with Claude and Mythos and the White House saying, actually, maybe we are gonna try to control AI, maybe this isn't the laissez-faire vision that Elon Musk and David Sachs had when they got started at the beginning of this administration.

Senator Josh Becker 16:42

Yeah, I'm gonna add a word to that, and also to the question maybe that was raised by CalMatters, which was around, you know, can the technology - can the enforcement technology keep up with the, you know, anyone trying to work around it? Maybe you want to -

Eddan Katz 17:01

Yeah, I'll start with answering that, and I think it's important to note that this type of transparency mechanism is not about detection, but about the recording, the marking of the record. And so we've actually changed the paradigm in which to think about battling this problem. I think we've, you know, we're developing some of the tools. As of today, we have set down some of the rules, and with C2PA and content provenance, we have the record, and the record connects the tools and the rules.

Senator Josh Becker 17:35

Any other questions? With that, I'll join the speakers here in thanking Ian Higgins for his tremendous work, uh, being in the details of all this. A lot of questions were raised by companies, some legitimate, some not, and we've had a lot to work through. For example, you know, for those who are disabled, how do they access content and how does that fit in? Had a lot of, um, just little important pieces to work through. I'm just really excited. We're at this moment, I'm really excited about the collaboration with the European Union. I thank you all for being here. Alright.