Seventeen minutes into his final State of the State address on Jan. 8, Gov. Gavin Newsom declared California’s newly enacted artificial intelligence safety law “a template for the nation.” He boasted that New York “adopted our approach” in regulating the Silicon Valley companies that are releasing increasingly powerful chatbots. But his jab at New York was more than braggadocio: It was an accurate representation of legislative machinations unfolding in real time.
That day, two New York lawmakers who had spearheaded the state’s AI law weeks earlier undid many of its key provisions. Though Gov. Kathy Hochul had signed it Dec. 19, she did so with the proviso that it be quickly revised with an industry-friendly amendment, which she had negotiated with the law’s authors. In her formal signing memo, she requested “certain clarifications” about what dangers AI companies would be required to report to regulators.
The Albany politicians agreed to delete key provisions that eliminated the law’s preventive power. Instead of alerting regulators to possible risks posed by their products before any disaster strikes, under the 2026 revision companies would have to file a report only if an incident leads later to people dying, suffering injuries or incurring billion-dollar losses.
This was the same change California lawmakers made in the weeks before passing their bill in September, transforming a first-in-the-nation AI safety law from an early warning system into an exercise in post-disaster documentation. New York’s alteration required copying just 12 lines of text from California’s weaker law.
The move to gut and rewrite New York’s Responsible Artificial Intelligence Safety and Education or RAISE Act came amid a multimillion-dollar bicoastal lobbying campaign that sprang up in mid-2025, funded by California-based technology interests that had lobbied heavily against federal legislation to oversee AI development. The companies and their allies exploited the political vulnerabilities of ambitious politicians — and arcane administrative processes — on both coasts nearly simultaneously.
The result was not just the watering down of strict regulatory language. The relaxations, loopholes and exceptions now common between California’s law and the amendment pending in New York are a demonstration of the power of wealthy industries to press lawmakers and governors to enact their preferred legal language.
New York’s 455-line amendment is not a tweak. It is a wholesale rewrite more than twice the length of the law it replaces. It incorporates large sections of California’s Senate Bill 53, which Newsom signed in late September, nearly verbatim. The amended RAISE Act is so strongly aligned with California’s industry-preferred law as to make the two nearly indistinguishable in their regulatory effect.
There is broad public support for artificial intelligence safety regulation across the country. But lobbyists for large tech firms, having failed in Washington to achieve a single, nationwide standard, got something almost as good: strongly aligned state laws shaped by industry intervention.
Elected Democrats in the two biggest blue states, who often criticized the deregulatory agenda of President Trump and his Republican allies in Congress, were now able to promote their new regulations as public safety wins while also pleasing Bay Area-based frontier model developers OpenAI, Anthropic, Google, Meta, xAI and others.
The RAISE Act, which Hochul signed in December, was to go into effect this month. But the amendment as submitted by New York lawmakers — and which is slated to go up for a formal vote within days — pushes implementation to next January, meaning California’s will be the only such regulation for at least nine more months.
In a Dec. 19 press release issued after Hochul signed the RAISE Act, its authors in New York’s legislature made no mention of their last-minute deal, as they touted a regulation that “creates nation-leading protections from AI’s risks.”
Bicoastal lobbying blitz
Technology lobbyists exploited two unusual features of New York’s legislative process. The governor may delay signing or vetoing a bill until the end of the calendar year. She may then negotiate “chapter amendments,” details of which are not publicly disclosed, along with a more general signing letter.
This backdoor process provided the AI industry ample time to mount a multipronged campaign to weaken the regulations both before and after they were signed.
The California process also was opaque, but in a different way. The main sponsor of SB 53, state Sen. Scott Wiener, did not respond to a public-records request for documents addressing contact with industry representatives as the law was being written late last summer, and officials in the Legislature declined to release them. If the documents contain details of who requested or drafted these changes, they remain hidden by an exemption in the state’s Legislative Open Records Act that shields lawmakers’ correspondence from public view.
In California, tech industry lobbyists communicated their disapproval of early versions of SB 53 to Wiener. Late pre-approval amendments to his bill slashed fines, narrowed whistleblower protections, relaxed reporting requirements and eliminated third-party audits. Rolling Stone reported that in the run-up to passage, tech industry representatives descended on the state Capitol in the final few weeks.
The people lobbying to weaken California’s law then turned their sights on New York. They established a super PAC, Leading the Future, armed with more than $100 million from venture capital firm Andreessen Horowitz, OpenAI President Greg Brockman and other business leaders. The political action committee mounted a three-month blitz to steer the RAISE Act toward its preferences. Besides donating to Hochul’s bid for reelection this fall, the group publicly attacked one of the law’s co-authors and mounted an astroturf campaign to “champion responsible AI” through advertising and messaging voters on social media.
New York State Sen. Andrew Gounardes said many provisions of the law he sponsored last year held AI companies to more stringent rules than did California’s. But he acknowledged in a recent interview that corporate lobbying and political activism undercut many of his accomplishments through the revisions pending this year.
“As we were negotiating the chapter amendment, we felt a lot of pressure from the VC community,” Gounardes said, referring to venture capitalists, and adding that Andreessen Horowitz and industry groups “were trying a tag-team approach of putting out content” disparaging the regulation.
He said he saw well-financed organizations “parachute into New York and pay for all these online ads, and all these text messages saying the RAISE Act will destroy rainbows, and cookies and chocolate, and the kids will have no more ice cream.”
The business groups also targeted Gounardes’ Assembly co-sponsor, Alex Bores, by “activating their political arm.” Bores announced in October that he was running to replace retiring U.S. Rep. Jerrold Nadler. In late November, Wired reported that Leading the Future was working to torpedo Bores’ congressional bid. Gounardes said the opposition was “focused solely and wholly” on Bores’ sponsorship of the AI legislation.
From prevention to forensics
The January amendments to the RAISE Act incorporated large sections copied from California’s SB 53. Taken together, industry-friendly provisions transplanted into New York’s law match California’s text by about 95%, with some sections copied verbatim.
The RAISE Act’s authors claimed victory in their deal with Hochul by preserving a clause requiring safety alerts to be shared with the state within 72 hours of a triggering event (compared with California’s 15 days). But a subtle wording change, mirroring an edit made in California’s law, altered its meaning dramatically. In the amended RAISE Act, the reporting timeframe refers to a “critical safety incident” as opposed to a “safety incident.” The new term changes the definition to apply only after harm has occurred.
Now, the only circumstance in which companies must report safety incidents to regulators beforehand is when they pose a “catastrophic risk,” defined as an event caused by rogue AI systems that could kill or injure 50 or more people or cause at least $1 billion in damage. The catch: This distinction would require a company to accurately predict the enhanced scale of damage for the event to be subject to the advance reporting mandate.
The original RAISE Act specifically cited scenarios in which early warning could prevent catastrophic harm. For example, a hacker might steal an AI chatbot’s model weights — the mathematical structure based on its training data. The unaltered weights ought to prevent the model from revealing how to build a bioweapon. This scenario warrants immediate reporting to prevent catastrophic harm, an AI safety expert told the Public Press in September. But a last-minute amendment to California’s SB 53 requires reporting the theft only if it causes death or injury, and the same change made its way to New York’s amendment.
Gounardes saw California’s conditional reporting requirement as a dangerous flaw and wrote the original RAISE Act to require unconditional advance reporting. “We wrote the theft of model weights provision that way because we intended to close a loophole,” said Gounardes’ communications director, Billy Richling. “We saw it as enough of a red flag to warrant making it a critical safety incident.”
While Gounardes’ distinction, avoiding California’s mistake, survived the original law’s passage, December negotiations with Hochul undid his accomplishment and he accepted the reversal.
New York pressure campaign
In summer and fall 2025, tech industry lobbyists worked to shape California’s SB 53 into a law that satisfied industry while allowing politicians to claim victory for tough action. The same dynamics unfolded in New York. Gounardes recognizes that they are up against a massive lobbying network working to block AI regulation nationwide.
The New York Times reported that on Oct. 3 Bores co-hosted a fundraiser at a Manhattan restaurant, drawing high-profile AI experts and elite software engineers. The RAISE Act, having passed in June, awaited Hochul’s signature or veto — or demands for amendments. The event netted nearly $250,000 in contributions toward Hochul’s November 2026 reelection bid.
A later Times story said Hochul traveled to San Francisco to raise donations with the help of Silicon Valley angel investor Ron Conway, a principal backer of Leading the Future, and Julie Samuels, president of industry group Tech:NYC. Both groups had lobbied against the RAISE Act.
The American Innovators Network, another group supported by Andreessen Horowitz, spent more than $100,000 on digital ads and bombarded New Yorkers’ mobile phones with anti-AI-regulation text messages in the weeks before Hochul’s Dec. 19 decision deadline, warning that the RAISE Act would destroy jobs and stifle innovation, according to reporting by Transformer News and The American Prospect.
Strongest provisions dismantled
The RAISE Act draft amendment is a brute-force rewrite. More than twice the length of the original law, it replaces nearly every provision that made it stronger than California’s SB 53. Among the significant changes, the amendment:
- targets only companies with gross revenues of more than $500 million, rather than considering the cost of training models or venture capital funding, which might cover more AI firms.
- slashes penalties by 90%. First violations fell to $1 million from $10 million. Repeat violations fell to $3 million from $30 million. The targeted companies have yearly revenues of billions of dollars.
- strips a prohibition on releasing potentially unsafe models.
- exempts catastrophes caused by “lawful activity of the federal government” from the usual reporting and liability provisions.
In the absence of federal AI regulation, amid opposition by the Trump administration, California’s SB 53 has become the default AI oversight standard for the nation. Wiener, said his “light touch” regulation balanced public safety with innovation. His previous attempt to wrangle the burgeoning industry, SB 1047, was stronger. But it met fierce industry opposition and Newsom vetoed it in September 2024.
Similarly in New York, the RAISE Act’s sponsors claimed the law “came despite ferocious opposition from tech lobbyists and venture capitalists.” Bores framed it as a righteous fight with vested interests, writing that his bill was “about ensuring that innovation and new technology is safe and serves people, not corporate power and profits.” And he had strong public backing: 84% of New Yorkers supported the RAISE Act, polls showed.
The original RAISE Act drew overwhelming support as it advanced through the Legislature. Seventy-three of 79 letters submitted to legislators backed the bill. Supporters included AI safety groups, startup companies, venture capital firms and independent AI experts.
Hochul’s pre-signing renegotiation to weaken the law came despite all this support. Why would the law’s authors agree to such a deal? Gounardes described it as a pragmatic compromise. “For me, the battle over AI safety is only just beginning,” he said in a January interview. “I am happy where we ended, but I’m not satisfied that this is the end of the conversation.”
He pointed to previous victories — on social media regulation and children’s online safety — to demonstrate his commitment to reining in big tech.
Gounardes confirmed that Hochul’s office insisted on importing California’s approach. “We could not get them to see our point of view on that,” he said, “but that doesn’t mean we’re not going to come back trying to fix that in the future.”
The tech industry is also rapidly arming itself against further attempts to regulate it, with Leading the Future raising $50 million late last year. On the agenda for 2026: to influence the midterm congressional elections, including targeting Bores’ race, according to Federal Election Commission filings.
Anthropic, the only major AI company to publicly support SB 53, spent $200,000 on lobbying to influence the law it ended up endorsing, California lobbying records show. Meta, the parent company of Facebook, poured $65 million into two super PACs mostly focused on supporting pro-tech California candidates and countering what the company cast as burdensome AI legislation, Politico reported.
In New York, identical amendments introduced by Gounardes and Bores must pass both chambers and get Hochul’s signature before March 19. If they fail, the original RAISE Act would take effect instead. Richling expects the legislature to move quickly.
“If we haven’t done this by March 19,” he wrote in an email, “then we — the entire New York State legislative apparatus — are not doing our jobs correctly.”
Listen to an interview with Jason Winshell on KALW Public Radio discussing his recent coverage of Senate Bill 53, and AI companies’ lobbying against whistleblower protections that got watered down in the final legislation.
This reporting was supported by a grant from the Tarbell Center for AI Journalism.