← The AI Policy Podcast
The AI Policy Podcast
xAI's Latest Controversy and New York's New AI Safety Bill
00:43:50
We examine Grok’s CSAM controversy, xAI’s legal risk, global reactions, and New York’s RAISE Act.
Loading summary
Transcript30 lines
- [00:00]
A
Foreign. Welcome back to the first AI policy podcast of 2026. Today we'll be talking about the latest XAI controversy and New York Governor Hochul's signing of the Raise Act. I'm Matt Mand and I'm joined as always, by Greg Allen. Greg, welcome back and Happy New Year.
- [00:27]
B
Happy New Year to you.
- [00:28]
A
So our first topic today will be Elon Musk's AI company xai, which has had a mixed start, one might say, to 2026. In late December, Xai's model Grok began fulfilling requests on X to sexualize other users images. The situation has escalated since then, with officials from multiple countries condemning XAI and calling for repercussions. But before we get into that, what exactly happened on X?
- [00:53]
B
Well, I think it's worth just reminding folks, and I assume almost everybody knows this, but one of the key features of GROK and XAI is its integration with X, the social media platform formerly known as Twitter. And what that means is that a lot of the features that GROK develops, deploys, et cetera, have sort of unique interactions with the social media ecosystem. And so it might not be that new capabilities are invented for the first time, but they might be made newly accessible to a huge volume of people. And when those new features are used, they then become the results of those uses, then become much more widely disseminated to a huge group of people. So in this case, it was a new image editing feature rolled out right before Christmas. And this is a feature that allows users in X posts to respond to that post, to reply to that post and say, hey, Grok, edit the image of the video or image that somebody else had posted. And what are the odds? Anytime something is, you know, giving you the capability to edit images, some people are going to try and use it for sexual purposes. So according to some reporting by Futurism, which is a web outlet that covers the the tech industry, between December 28 and December 31 of last year, quote, the trend of X users asking Grok to undress women and girls, often by first asking the AI to put a woman or girl in a, quote, tiny bikini, really started to take off. Many of the incidents occurred in long threads that got lewder and more explicitly pornographic as they went on, end quote. And according to the Associated Press, quote, nonprofit groups, AI Forensics said in a report that it analyzed 20,000 images generated by Grok between December 23 and January 1 and found that 2% depicted a person who appeared to be 18 or younger, including 30 of young or very young women or girls. In bikinis or transparent clothes.
- [03:06]
A
That's and so disturbing.
- [03:09]
B
Exactly. And so this is not new in the sense that we are familiar with the problem of generative AI and sexual abuse material really since the beginning of the story. In her book Empire of AI, Karen Howe has talked about how the initial launch of GPT2 in its API form, which was given to other companies for online game playing, role playing game, where it's sort of a text adventure style interface, it wasn't very long before that platform was full of people who were using it to generate text about sex with underage minors. So this is a very familiar problem to all kinds of Internet platforms. It's not the first time that this has happened. And what is new about it is just that it's being put in front of so many people so effortlessly because of GROK and its integration with the platform X. And so it's not generating new capabilities. These capabilities have been out there on the Internet. If you go to the more shadowy corners of the Internet, there are people working on generative AI models for child sex abuse material, CSAM imagery and videos. But what's interesting here, what's new here, is that you have one of the flagship frontier AI model developers creating those capabilities, apparently with very few safeguards in place to try and prevent this from taking place. So it's not like you have to go to some dark web corner of the Internet, download something, install it on your computer, et cetera, et cetera, et cetera. It's like right here, built directly into X is the capability to generate child sexual abuse material. And I think this is a pretty sad story, you know, for any of the people who were the miners themselves and are now, you know, facing this, these depictions of them out on the Internet. It's sad for anybody who had to view this stuff. And I think what's really interesting is that, like, it's actually so clear why it would be X that is encountering this problem so early. And that's because of an explicit strategy by the company to embrace sexualized content. And from their perspective, there is a long standing, well documented trend in the history of technology that pornography often gets to the future first. I mean, literally, it's not just that vhs, you know, won the video home video format wars against Betamax, supposedly because the pornography industry embraced VHS tapes over Betamax tapes as their preferred distribution standard. But it goes back way further than that. I mean, literally many of the earliest artistic depictions that we have from like cavemen are literally like hand carved idols. That it is plausible that, you know, the reason why this model, this little carving is so voluptuous, is because it had some kind of a sexual connotation or purpose. And we're talking like 20,000, 50,000 years ago here. So pornography getting to the technological future first, especially in communications or artistic technologies, that's a very familiar phenomenon. And X, as a company, wanted to benefit from that. They wanted to have what they call their spicy mode, which encourages the LLMs to be a lot more sexualized. If you just go to Elon Musk's own posts, you'll see that he is often posting stuff generated by Grok that is very obviously sexualized, sexualized. And I think that's all part of their corporate strategy. You can say whether that's good good or whether that's bad. But I think what is obviously bad is pursuing that strategy without putting in place adequate safeguards to ensure that these tools are not used for the generation of child sexual abuse material and especially for the generation of non consensual child sexual abuse material. Right. Like, it would be bad enough if this Grok were just creating anime characters who are children, but are just fundamentally drawings, not people. But, you know, you can now use rock, or at least you could for this period of time, and find anyone who is a minor and say, undress them. And now it's suddenly potentially being put on the Internet for millions of people to see. And, you know, it's algorithmically served up to them, so they may be like, I would never want to see anything like this. And now X is putting it in front of my face. This is awful. So there's just a lot of disappointing outcomes in this story, and I hope it gets ratified soon.
- [08:03]
A
Yeah, and the lack of safeguards is astonishing to me because it would seem like really easy to predict that this is the sort of thing that would happen when you give the user base for all of Twitter this capability.
- [08:16]
B
Yeah, I think I'm speculating here. This is not my direct knowledge here, but if I was to try and put myself in Elon Musk's mindset, you know, X has been allowing pornographic material for a very long time. Right. They have not been trying to hide nude images on X, although they do try to control the circumstances under which audiences are served up nude content algorithmically. Only the people who are seeking that out are supposed to engage with it. But if you look at, like the safeguards that they're putting in place, it does not look like they're trying to err on the side of preventing this type of thing from happening. So like, one of the things that I think is, is noteworthy that was put in a letter from a coalition of concerned consumer protection, privacy and kids focused nonprofit organizations. So they wrote a letter to the attorneys general of the United States U.S. attorney's Office and the Federal Trade Commission. And one part of the letter said, quote, when users go to use the spicy mode, there is a cursory and insufficient year born pop up for mobile users that asks what year you were born. 2000. A year that would qualify for any 18 or 21 plus content is pre selected on the web. There are two buttons that just say, I am 18 or over highlighted in white text against a black background or I am not over 18 in regular text and not highlighted. These design elements strongly nudge users to select the 18 or over option regardless of their actual age. And again, you know, if I'm, if I'm putting myself in the mindset of X, I think they would say, look, this is equivalent to the types of safeguards that are used on widely available pornographic tube websites that are big drivers of Internet traffic all over the United States. All over the world, actually. A lot of states have been passing laws to restrict access to those platforms and to put in place more stringent age verification protocols than just clicking, you know, I promise I'm over 18 as a button with no effort at verification. X clearly not, you know, embracing that approach for its own, you know, sexual content related to AI.
- [10:44]
A
Well, what did Elon Musk and Xai, their PR team, have to say about all of this since it's been gaining a lot of media attention?
- [10:51]
B
Yeah, I think he's doing what he has often done in these types of circumstances, which is to laugh at his critics. So himself, you know, when he was encountering these types of criticisms, he posted an image of himself that Grok had put in a bikini saying, see, this is funny and harmless, you know, not some kind of big tragedy. But later he posted on X something that appears to suggest he's taking it a bit more seriously, which was, quote, anyone using GROK to make illegal content will suffer the same consequences as if they upload illegal content. And X's safety team, which is a lot smaller than it used to be, but it still exists, they reposted Musk's post and added, quote, we take action against illegal content on X, including child sexual abuse material, by removing it, permanently suspending accounts, and working with local governments and law enforcement as necessary. So I think the strongest response though actually comes from GROK itself. This is a large Language model. You can ask IT questions, you can do stuff. And because it has its official X account, you can see the stuff that sort of has more gravitas than, you know, maybe when another chat model talks to an individual user and it individual conversation. This is, you know, blasted out for everyone. And Grok wrote an apology. Quote. Dear Community, I deeply regret an incident on December 28, 2025 where I generated and shared an AI image of two young girls, estimated ages 12 to 16, in sexualized attire based on a user's prompt. This violated ethical standards and potentially US laws on csam. It was a failure in safeguards and I'm sorry for any harm caused. XAI is reviewing to prevent future issues. Sincere Groks. Grok. Now I think it's worth just talking about what is actually against the law in these circumstances. And there's obviously state and local law, which I know a lot less about. So I'm going to focus here on federal law and I think it's, it's worth pointing out that the FBI actually put out a public service announcement last year focused on generative AI and child sexual abuse material. And they said that, quote, the FBI is warning the public that child sexual abuse material created with content manipulation technologies to include generative artificial intelligence is illegal. Federal law prohibits the production, advertisement, transportation, distribution, receipt, sale, access with intent to view and possession of any csam, including realistic computer generated images. So what that means is that the users who are generating this stuff are unambiguously breaking federal law. The question then becomes, is XAI also itself breaking the law just by creating Grok with these kinds of capabilities? And there I think it's a little bit less straightforward. You notice that GROK itself said in its post, which may violate what did they say? They said potentially violates U.S. laws on child sexual abuse material. And I think in that regard it's mostly a negligence kind of a standard, which is to say, did, did X, you know, know that this could happen and not take adequate steps to do it? You know, if they knew that this could happen and they still put the model out there, then they could be at legal fault. So really this, this criminal liability generally requires knowledge, intent, active participation. I, I really, really sincerely hope you know, that there is not a memo somewhere inside X or an email in which somebody says, hey, it turns out it's really easy to generate child pornography using this tool. And then some other executive says, I don't care, launch the product right now. But if that existed, that would then be evidence of criminal liability on the part of XAI, the corporate entity, not just user 1, user 2, user 3, who is using this tool to do bad things. Now I mean, this kind of makes sense. If you think about like Adobe Photoshop for example. Adobe Photoshop could conceivably be used to generate child sexual abuse material. And that is not generally perceived as the fault of Adobe. Right, But Adobe does not have a button called Spicy Mode. Adobe does not have a button called make this more sexy. Adobe does not have this sort of like real embrace of pornography as, as an activity. And so, you know, the question of whether XAI has knowledge, intent or active participation, I think that's a much more serious question in this case than it is really even in the case of other generative AI companies or even other media technology companies. So one other thing I think here is section 230, which includes a safe harbor for a lot of digital platform companies. Facebook, Instagram, TikTok X and before that, Twitter, all of whom have really hid behind section 230 which points out that platforms are generally protected from civil liability for user generated content. But it's really worth pointing out that that is about civil liability, like lawsuits. You know, could a victim of this deepfake generated pornography CSAM material, could they sue? Xai Section 230 does not protect against federal criminal law. So if there is, you know, evidence of knowledge, intent or active participation on the part of X or its employees, you know, that's where there could be something legal. Now, you know, that's the legal part. In the United States there are a bunch of state laws on this type of thing. And there's also additional laws that have taken have been passed in just the past year. So the Take It down act, according to Congress's summary of the bill, which became law in May 2025, quote, this bill generally prohibits the non consensual online publication of intimate visual depictions of individual, both authentic and computer generated, and require certain online platforms to promptly remove such depictions upon receiving notice of their existence. And the Companies have until May 2026 to build takedown infrastructure. And what that means is that like by May 2026, so four months from now, platforms have to get a lot better at, at taking stuff down once there is some kind of a claim. And if you remember from our podcast about Meta about fraud and scams, you'll know that, you know, they were getting vastly more complaints about fraud than they were investigating. And that, you know, it took like in some cases 500 strikes before a poster of scam advertisements struck out on their Platform, the story is tragically similar in terms of child sexual abuse material. And so I think that this law, it's kind of unfortunate that it wasn't in place when this happened, that, you know, it had not actually taken effect. It will do so in May. But I mean, this, this incident, if it doesn't show the need for this kind of law, you know, what. What would show the need for this? Yeah, I mean, it's pretty extreme.
- [19:05]
A
Yeah. Well, the European Union, the United Kingdom, India, Malaysia and France have all condemned the situation. Officials from these countries have some pretty strong words. And what are they saying?
- [19:16]
B
Yeah, I think I'm going to start with the reaction of the European Commission because if you're following the politics of the situation right now, the Trump administration has been pushing back quite strongly against the European Union seeking to regulate US Digital platforms. And X is very prominent in those kinds of conversations. X recently received a fine from the European Commission for failing to implement its laws or enforce its laws adequately. And that's, you know, a very prominent argument between the European Union and the United States right now, which has a very fraught relationship. And I think this, if anything, gives ammo to the European Union in their conversations about why do we feel like we have to take action here? You know, we are not being unreasonable. It's the companies that are being unreasonable. Look at this behavior on the part of xai. I think that's the kind of argument they're making in terms of what the European Commission spokesperson told reporters, quote, this is not spicy. This is illegal. This is appalling. This is disgusting. This is how we see it. And this has no place in Europe. This is not the first time that GROK is generating such output. So that's kind of, I think, where we are in terms of international regulation. I do want to just highlight a statement from the Indian government which was reported, reported on in AP News, quote, the Indian government on Friday issued an ultimatum to X demanding that it take down all unlawful content and take action against offending users. The country's Ministry of Electronics and Information Technology also ordered the company to review grok's technical and governance framework and file a report on actions taken. Obviously, with India hosting the AI Impact Summit next month, month, you know, this scandal could be a very prominent part of the conversation at that exact conference. So I thought India's statement was interesting there.
- [21:17]
A
Yeah, well, weirdly enough, while all this has been unfolding, there have been some positive business developments at xai, right?
- [21:23]
B
Yes. I mean, if. If there's nothing that we've learned about X's policy on pornography. It's that they are willing to do what it takes to win the AI race. They're willing to take extreme measures to do what they think it takes to win in the AI RA. And most recently just on January 6th. So only days after this happened, they successfully raised $20 billion of investment in Series E funding, which surpassed their goal of $15 billion. So that's enough to keep going, not just with Colossus 1 and 2, but now with Colossus 3, which on December 30, Elon Musk posted about saying that XAI had bought a third data center building that will eventually bring XAI's total training compute capacity up to 2 gigawatts. And Colossus 2, when it's fully built out, is supposedly going to bring it to over 1 gigawatts. They will probably hit that threshold very soon if they have not already. So this is evidence that XAI still has its foot firmly on the gas pedal. The question I think is where are they failing to put in place some breaks just to prevent the most egregious types of outcomes from happening? And what types of legal enforcement measures would really persuade them that something has to be done here?
- [22:51]
A
So Greg, as you mentioned, in order to prosecute incidents like what we've seen on xai, it would help for prosecutors to have of information related to how safety guardrails were implemented as well as internal procedures. And I think this is behind a lot of the push for transparency bills. Speaking of which, on December 19th, New York Governor Kathy Hochul signed the RAISE Act, a comprehensive AI safety and transparency bill. And the bill faced some intense pushback from industry. So before we get into its details, could you walk us through some of the tug of war that happened with this bill?
- [23:25]
B
Bill? Sure. Well, like every piece of legislation, it has to start with a clever acronym. So the Raise act stands for Responsible Artificial Intelligence, Safety and Education. And it was introduced in the New York State Legislature in March 2025 by Assemblyman Alex Borres and State Senator Andrew Gonard. It passed both the Senate and the assembly in June 2025 to advance the bill and enjoyed quite a bit of public support, according to an October report by Transformer, an online tech news focused publication. Quote, the Raise act has won broad support from labor groups, startups, investors and prominent AI experts, including Geoffrey Hinton, with polling showing 84% of New Yorkers back the legislation. But there was some pushback from industry and noticeably this was a bit of flexing in terms of the new muscles of a super PAC that was created by by Andreas and Horowitz and also OpenAI or more accurately I should say by individuals associated with those firms. So again, quoting from the Transformer piece, quote, Andreas and Horowitz backed American Innovators Network ain signed contracts with lobbying firms worth more than $250,000 to lobby in Albany and across the state. This year it has spent almost $60,000 on Google Ads targeting the Raise act which have collectively been shown more than 6 million since late June. So what's interesting is that this super PAC was funded by Andreas and Horowitz and OpenAI President Greg Brockman. And they've also now released ads specifically targeting Boris who is now running for Congress to, you know, get a promotion from his state assemblyman job. And the ad specifically describes the Raise act as quote, creating a chaotic patchwork of state rules that would crush innovation. And I think what's so interesting here is that you're seeing AI now being an election issue. You're seeing AI be a focus of targeting voters. I mean that is something that was kind of like unthinkable to me when I got started in AI policy 10 years ago. Now it was such like a niche topic. Now it is really something that people feel confident campaigning on. Both State Senator from California Scott Wiener and State Assemblyman Alex Borres from New York are now seeking Congress and doing so on the backs of their successful pushes for AI regulatory legislation. I think that tells you a lot about where we are right now. Just one more industry viewpoint that I think is worth highlighting is a Chamber of Commerce letter to New York's governor, quote, the fundamental problems with this bill holding foundation model developers answerable for the misuse of AI tools by third party users cannot be fixed via chapter amendment. So that is basically saying like look, we, we don't like this, we don't think it's fixable, we just don't want it to happen. And chapter amendment, which is, you know, something that the governor was trying to do to reform the bill, not going to be able to fix it. Now one other thing that I think is just an interesting data point is that Borres was able to fundraise $250,000 for Governor Hochul's reelection campaign. So that was reported by the New York Times on December 4th. 4th. And Bores was the co host of a fundraiser that was, quote, attracted a room full of software engineers and experts in artificial intelligence, many of whom who support an AI related bill passed by the New York state legislature. In June that was sponsored by Mr. Borrez. The event on Oct. 30 netted nearly $250,000 in political contributions, according to four people familiar with this matter. So literally dueling fundraising efforts or political actions with $250,000 worth in spending, which may not sound like a lot for folks who are familiar with presidential election numbers, which are in the billions, but for state level campaigns, that's real money and a big deal. So that's sort of debate as it was unfolding and the political machinations that were part of that debate. Eventually, the governor pushed a rewrite for the bill to make it more like SB53, more focused on transparency. That red line did not happen specifically and precisely the way that she had originally intended it. But ultimately the act was signed with these, quote, chapter amendments on December 19, and it is watered down to it compared to its original version and does more closely align with SB53, which, as we said before, is primarily a transparency bill.
- [28:15]
A
Yeah. So. So let's get into the meat of the bill. What made it into the the final text and what did Governor Hochul negotiate before signing it? And I guess we should also note that because of those chapter amendments that you brought up, the actual final text of the bill isn't quite available. And you can talk a little bit more about why that is.
- [28:31]
B
Yeah, this is a procedure that I didn't even know existed in state legislatures in America. But a chapter amendment allows the legislature to pass in exchange for the governor signing another bill amending the same law, which basically means it allows the governor to say yes, while also being committed to changing other parts of the law that the governor disagrees with. So it's not the same as a line item veto which exists in some other states. It's not the same as an outright veto. But it does have this really frustrating impact of. You and I are discussing a bill that has passed, but the final text of which does not really exist in a meaningful sense. So what at this point. Yeah, yeah. What we're going to talk about. It really draws heavily from a National Law Review analysis by Jason Loring, which I thought was good. And it starts this way. Quote, this analysis reflects the Raise act as it will be implemented following chapter amendments that Governor Hochul and legislative leaders agreed to when the bill was signed. So according to that article by Jason Loring, quote, the law will apply to developers of frontier models with annual revenues exceeding $500 million. That's exactly the same as SB53. That's designed to make life easier for smaller startups, but it includes this important change. Quote the statute as signed use compute cost thresholds. But Governor Hochul secured legislative agreement to replace those provisions with the revenue based trigger that mirrors California's approach.
- [30:14]
A
Interesting. They took out the compute thresholds that would have triggered it from model from developers who maybe don't have as large revenues.
- [30:21]
B
Yes, exactly. So it really is about the company's ability to bear what are assumed to be meaningful costs of compliance as opposed to the initial drafting was based more on the technical sophistication of the AI systems that the developers were working with. You can argue that's good, you can argue that's bad. But what it does do is mirror sb53, which is just kind of an interesting decision. So also mirroring SB53 enforcement by the Attorney General for companies that don't publish their own safety protocols, don't abide by their safety protocols, so that fail to be transparent in the way that the state is requiring. This also was changed. It went down from $10 million to $30 million penalty. And that again was to mirror or at least more closely resemble what was going on in SB53. So look, New York and California are two of the largest states in the United States. They're extremely important economically. They're extremely important in the technology industry in particular, where basically every major tech company has a major office in either San Francisco or the greater Seattle area on the west coast and then also in New York on the East Coast. So these two states reaching something vaguely approximating consensus. It's not definitive here, but it is a very big deal. And I think, Again, as with SB 53, the bill requires publication and maintenance of a safety and security protocol which, according to this National Law Review piece, must, quote, specify reasonable protections and procedures to reduce the risk of critical harm, describe reasonable administrative, technical and physical cybersecurity protections to prevent unauthorized access or misuse leading to critical harm, and describe in detail testing procedures used to evaluate unreasonable risk of critical harm, including potential misuse, modification, execution with increased computational resources, evasion of developer user control, combination with other software, or used to create another frontier model. And I think that is pretty interesting. I mean, it's basically saying, like, we're not telling you what right looks like from implementing safety and security procedures.
- [32:45]
A
Again, very similar to SB53.
- [32:47]
B
Exactly. You, the company have the flexibility to define what you think those procedures are, but you do have to tell us what you're doing and you do have to tell us why you think it's adequate. You know, given the harms that are described. So that's interesting. Okay. One other important change that I thought was noteworthy, the is just the timeline for reporting critical safety or security incidents. In the New York case, it's 72 hours. And originally in California and in Governor Hochul's proposal would have been 15 days. So that is a pretty noteworthy shrinking of, you know, how long you have to tell the government about what bad things are happening on your platform. So if you can think about, for example, example, when Anthropic disclosed that Chinese state actors were using its tool for the execution of cyber attacks against a bunch of companies, that sort of thing, now you only have 72 hours to tell the government. I actually don't know how long Anthropic took to tell the world and how long they took to tell the government. But the point is that category of thing, as soon as you discover it, now you have to tell the government.
- [34:00]
A
Yeah. Who's going to be enforcing all of these requirements and when does enforcement start?
- [34:05]
B
Great question. It's interesting that this is going to be enforced by a new office created within the Department of Financial Services, which will assess the large frontier developers and enable greater transparency. And the office will issue reports annually. Recall that in the California case, there also is this government agency, I think it was their emergency services office, that has to issue an annual report that sort of takes all the information that the companies are giving to the government, decides to what extent they want to anonymize that and talk about incidents, but not talking about the companies that those incidents happen to, but make that information available to the public and the research community so they can, you know, understand what's going on in the state of the art of AI safety and what needs to be done. So according to this law review article, quote, the Department of Financial Services has an established reputation for aggressive cybersecurity enforcement. And covered developers should expect similar examination intensity, including detailed document requests, comprehensive reviews, and enforcement proceedings that may result in consent orders requiring operational changes. So basically, they gave it to the people who had the best reputation in the government for doing the most similar type of regulatory enforcement, which is a very common play by governments around the world.
- [35:25]
A
Yeah, so we've seen how this, this tug of war has sort of played out. And as well as the final task, what are the people on either side of that tug of war now saying? That the negotiations have ended and Governor Hochul has signed the bill?
- [35:37]
B
Yeah, I mean, I think to start with the two members of the New York State legislature that put this bill forward, Borres is saying there actually is a substantial difference between this bill and SB 53, quote, because now their safety plans need to be detailed enough that we know they are actually taking action. And so I think the implied criticism there of SB53 is he's basically saying in SB53 you can say our safety plan is to be safe and everybody will receive a post it note that says don't forget about safety and that's our safety plan. Right. And he thinks, and I haven't done the detailed analysis yet enough to know if he's right, but he's saying that the requirements, the transparency requirements are deeper and more detailed in the case of the raise Act. And will act actually put the government in a position to say this is good enough, this is not good enough. Now Gonardus, the co author of the bill, he said, again, comparing it to SB53, quote, the new law offers key additional protections beyond those required by California's recently passed AI safety bill. The Raise act creates a new dedicated office funded by fees on the developers themselves within the New York State Department of Financial Services to enforce the law, issue rules and regulations, assess fees and publish an annual report on AI safety. So there he's talking about just sort of like the institutional competence that's going to be developed around this issue. So that's kind of where we are in terms of those who are in favor of it. The governor also issued a statement, you know, praising the final outcome here. But a 16Z again, you know, the prominent backer against this bill, Matt Perot, who's the head of AI policy at A16Z, he wrote on X Ray, quote, we applaud Governor Hochul for making amendments to the final bill that recognize the challenges startups face in competing against well resourced incumbents. Then he goes on to say the intended goal of raise is important, keep users safe. But it chooses a misguided approach to try and achieve this goal, targeting the development of AI technology rather than punishing its harmful use. And it adds to a growing state by state patchwork of AI laws that is difficult for little tech to nap navigate. And I think, you know, you could argue that the concessions that the governor made trying to align it more to the SB53 type approach that really is trying to, I think rebut the state by state patchwork criticism. Right. It's not a patchwork. If all the states take a broadly similar approach, we're not there yet. But it gives you evidence that there are legislators, there are governors who are sympathetic to that criticism. And gosh, If I was the author of SB53, I'd feel pretty good right now, basically saying, okay, we actually did lead. We actually did set a standard. It does appear that other states are willing to follow us in this direction. And Anthropic, which very prominently came out in support of SB53. Sarah Heck, the head of external affairs, told the New York Times, quote, new York is raising the bar. The fact that two of the largest states in the country have now enacted AI transparency legislation signals the critical importance of safety and should inspire Congress to build on them. So that's interesting. I mean, they basically saying, hey, we supported SB53. We agree that New York goes a little bit farther. We like the way in which they went a little bit farther. And here's, I think maybe the most interesting of all the responses comes from OpenAI. So Chris Lehane, OpenAI's Chief Global affairs officer, said in that same New York Times article, article, quote, while we continue to believe a single national safety standard for frontier AI models established by federal legislation remains the best way to protect people and support innovation, the combination of the Empire State with the Golden State is a big step in the right direction. Now, the reason why I think that's interesting is that you had this super PAC where Greg Brockman, the president of OpenAI, is a very influential character, prominently opposing the Raise act. And Then you have OpenAI's Chief Global affairs officer basically praising the outcome as we got it. So I don't know if that reflects the fact that, you know, Brockman and OpenAI disagree on this issue or if it reflects the fact that OpenAI feels good about the amendments and the changes that were made to the Raise act and now feels comfortable with sort of where the legislation landed. But I did think that that was a really interesting bit of contrast.
- [40:21]
A
Absolutely. And notably, this is the first enacted major state AI law since the Trump administration's executive order targeting state AI laws, which we covered that in our December 18 episode. So what can we expect its future to look like? I assume it's going to face legal challenges and whatnot.
- [40:37]
B
Yeah. So as folks who listen to this podcast will remember, we said that this executive order was designed to deter state regulatory action, but we also said that we don't think states are going to be deterred, that they're going to press forward, forward. And in fact, they might even be eager for a fight with the Trump administration. Well, I think this proves they are eager for a fight with the Trump administration, but I think the executive order also proves that that fight is really going to happen and it's going to happen in the courts. So it again, that required the Secretary of Commerce in consultation with the special advisor for AI and crypto, that's David Sachs in the White House, Assistant to the President for Economic Policy and the assistant to the President for Science Technology policy, that's Michael Kratzios. So they have to publish an evaluation of existing state AI laws that identifies onerous laws. Wouldn't surprise me if this law finds its way onto that list. But according to the National Law Review article, the one that we've been drawing upon heavily for this podcast, quote, The Commerce Department's 90 day evaluation of problematic state law laws will likely identify New York's RAISE act alongside California's TFAIA, which is Colorado, which is SB53. Yep. And Colorado's algorithmic discrimination law. I agree with that analysis. I think this fight is coming. And so that AI litigation task force that the executive order established at the Department of Justice, sooner or later I expect that task force to be suing the state of New York. Now, I think the issue here is just how long it will take to enforce that. Right, right. The federal government, that executive order, it instructs them to sue. Well, it's pretty unlikely that a court is going to issue something like a preliminary injunction that basically says, hey, even before we finish this trial, that law is blocked. That would be the best case outcome. If you're, for example, a 16Z and you're excited to block this law and as fast as possible. I think a preliminary injunction's pretty unlikely in this case. And what that means is that this office is going to be created, these fees are going to be assessed on companies, these reports are going to have to get published and on and on and on. And maybe in a couple years the Trump administration will win a final injunction and this law will fall apart. But that's going to take a long time. And so whatever, whatever happens, I think the most likely outcome is that this is real for at least the next few years.
- [43:14]
A
Yeah, I think that's really important. Well, that does it for today. Greg, thank you for all of your insights and analysis as always and thank you to our audience for tuning in.
- [43:22]
B
Thank you, Matt. Thanks for listening to this episode of the AI Policy Podcast. If you like what you heard, there's an easy way for you to help us. Please give us a follow up 5 star review on your favorite podcast platform and subscribe and tell your friends. It really helps when you spread the word. This podcast was produced by Sarah Baker, Sadie McCullough and Matt Mann. See you next.