New York’s governor is trying to turn the RAISE Act into an SB 53 copycat
EXCLUSIVE: Gov. Kathy Hochul is proposing to strike the entire text of the RAISE Act, replacing it with verbatim language from SB 53, sources tell Transformer.
by Issie Lapowsky
New York Governor Kathy Hochul is proposing a dramatic rewrite of the RAISE Act, the AI transparency and safety bill that recently passed the state legislature, according to two sources who reviewed the governor’s redlines on the bill.
The governor’s proposal would strike the RAISE Act in its entirety and replace it with verbatim language from California’s recently enacted law, SB 53, with minimal changes. SB 53 is generally viewed as a lighter touch approach. One source who spoke with Transformer on the condition of anonymity said the proposal would effectively make SB 53, a law that “was always meant to be a floor” for AI regulation, “suddenly become the ceiling.”
Kristin Devoe, a spokesperson for Gov. Hochul, declined to comment on the details of the bill, but said “the governor contributes to the review of the legislation.” Gov. Hochul has until December 20 to decide whether to sign or veto the bill.
The proposed changes come amid an ongoing fight about federal preemption of state AI laws, with President Trump saying this week that he would sign an executive order to establish “One Rulebook” for AI. Some have floated SB 53 as a potential framework for a national standard.
The RAISE Act, which was introduced by Assembly Member Alex Bores and co-sponsored by State Senator Andrew Gounardes, has become a political lightning rod, inspiring vehement pushback from AI companies and organizations representing major players in AI. Transformer has previously reported on the $300,000+ spent by the Andreessen Horowitz-backed American Innovators Network to oppose AI regulation in New York this year.
Just this week, Leading the Future, a super PAC funded by Andreessen Horowitz and OpenAI president Greg Brockman, launched an attack ad targeting Bores, who is running for Congress, and describing the RAISE Act as “creating a chaotic patchwork of state rules that would crush innovation.”
While the ad is ostensibly targeting Bores’ congressional campaign next year, the timing of its release — many months out from the New York primary — suggests the PAC may have also been angling to influence Gov. Hochul’s decision regarding the RAISE Act. “My gut reaction is that the super PAC scared her,” said one source who reviewed the governor’s redlines. That is despite efforts from pro-AI regulation advocates to sway her in the other direction; Bores recently held a fundraiser for Hochul which raised $250,000 for her campaign.
Devoe told Transformer: “Campaign donations have no impact on government decisions. Governor Hochul evaluates all legislation based on the merits and makes determinations solely on what is best for New Yorkers.”
Leading the Future did not immediately respond to a request for comment, but its leaders have described the RAISE Act as “a clear example of the patchwork, uninformed, and bureaucratic state laws that would slow American progress and open the door for China to win the global race for AI leadership.”
Under the new proposal, the bill would require large frontier model developers to publish AI safety frameworks on their websites and conduct reviews of those frameworks once a year, not unlike the RAISE Act. It would also, like the RAISE Act, create new enforcement mechanisms and fines for violations.
But the two documents differ in significant ways. The language in SB 53 and the governor’s proposal, for instance, define a “large frontier developer” as any developer of a frontier model with annual gross revenues over $500m. RAISE defines it as any developer (with some academic exceptions) that has trained at least one frontier model and spent more than $100m in training compute costs. That means major provisions in the RAISE Act would apply to developers that are not yet bringing in substantial revenue, but have still built powerful models. Under Gov. Hochul’s proposal, they wouldn’t.
The RAISE Act also prohibits companies from deploying models that “would create an unreasonable risk of critical harm.” The new language includes no such prohibition. While it does require that these large frontier developers publish and review safety standards and the results of their safety tests, it doesn’t prevent them releasing a model if, in fact, it fails those tests.
Likewise, where the RAISE Act requires large frontier developers to report on safety incidents once they have “demonstrable evidence” of an increased risk of harm, the new language would require such disclosures only after the harm has occurred.
Gov. Hochul’s rewrite would also cap fines for violations at $1m, compared to RAISE, which would impose a maximum of $10m fines for first violations, and $30m fines for subsequent ones.
Finally, the new language omits provisions in the RAISE Act aimed at companies that use large AI models to train smaller ones, producing similar capabilities — what’s known as “knowledge distillation.” This provision was included to ensure that a company like China’s DeepSeek, which reportedly trained its models in this way, would still be subject to New York’s rules. “They just let China off the hook,” one of the sources said.
The sources who spoke with Transformer said the new language waters down a bill that was meant to build upon SB 53. “The RAISE Act would, in a couple of key areas, really enhance the safety and security of these models, above and beyond what California’s law currently does,” one of the sources said."
‘The message has been very clear: Take it or leave it.’
In some ways, the effort to align different states around the same light touch legislation is evocative of the tech industry’s push to pass copycat privacy laws in states across the country. The difference in those cases is that if states didn’t pass their own privacy laws, their citizens would have no rights under another state’s laws.
But SB 53 works differently by imposing transparency and safety requirements on the companies themselves. Because large frontier developers are already subject to these requirements in California, replicating the law in New York is largely redundant. The primary difference is that it would give New York’s attorney general the power to punish violators, too. Still, one of the sources suggested that there was little difference between “signing this bill and a veto.”
The changes Gov. Hochul is proposing also go beyond what New York’s own tech lobbying group, Tech:NYC, was pushing for, according to the sources. During negotiations with lawmakers over the RAISE Act, sources said Tech:NYC indicated that it preferred some of the language in RAISE to SB 53, but those preferences are not reflected in Gov. Hochul’s proposal. Tech:NYC declined Transformer’s request for comment.
Gov. Hochul and the legislature now have a chance to negotiate over her proposal, but the sources who spoke with Transformer said the governor’s opening salvo — striking all of the text of RAISE in one fell swoop with just days to go before her deadline — does not bode well for those negotiations.
“This feels like they’re doubling down,” one source said. “The message has been very clear: Take it or leave it.”
Issie Lapowsky is a freelance journalist focused on tech, politics, and national affairs.
Fantastic reporting on the RAISE/SB53 swap. The knowledge distillation carveout is wild because it basically creates a regulatory arbitrage where companies can just use model distilation to sidestep the revenue threshold entirely. Seems like the $500m revenue bar versus $100m training compute criteria means startups burning cash on frontier models get a pass until they actually monetize, which feels backwards when the risk profile is about capability not profitibility.