California Governor Vetoes AI Safety Bill

California Governor Gavin Newsom has officially vetoed the highly anticipated Frontier AI Model Safety Innovation Act (SB 1047). In his veto statement, Governor Newsom elaborated on several key factors influencing his decision, including the potential burdens the bill could impose on AI companies, California's leading position in the global AI sector, and concerns over the bill's possible overreach.

SB 1047 is designed to establish safety standards for the development of large AI models, particularly targeting developers with training costs exceeding $100 million or high computational demands. The bill mandates that these developers conduct rigorous testing before deploying their models, including simulated cyberattacks, implementing cybersecurity defenses, and providing protections for whistleblowers, to ensure the models' security.

SB 1047 was initially submitted to the California Legislature in February this year, promptly sparking widespread debate. After multiple revisions, the latest version of the bill was released by the California legislative body on August 19. While the bill received support from renowned AI experts such as Geoffrey Hinton and Yoshua Bengio, it also faced opposition from several scientists, including Yann LeCun and Fei-Fei Li, as well as leading AI companies, who argued that the legislation could "hinder innovation."

Governor Newsom remarked, "There is no doubt that SB 1047 has a positive intention, but it fails to fully consider the specific application scenarios and risk levels of AI systems. Regardless of whether an AI system operates in high-risk environments, makes critical decisions, or handles sensitive data, as long as it is a large system, the bill imposes stringent safety standards on it. I believe this approach is not the optimal way to protect the public from the potential threats posed by AI technologies."

Additionally, Governor Newsom highlighted that SB 1047 could instill a "false sense of security" among the public, while in reality, some smaller and more specialized AI models might present even greater risks. He stressed that overly stringent restrictions on large AI models could weaken the innovative momentum that drives the advancement of public interests.

Governor Newsom expressed his support for developing AI safety protocols and safeguards, ensuring that malicious actors face "clear and enforceable" consequences. However, he believes that the state should pursue solutions grounded in empirical trajectory analysis of AI systems and their capabilities, rather than relying solely on a broad and potentially unrealistic piece of legislation.

In response to Governor Newsom's decision, Meta's Public Affairs Manager Jamie Radice stated, "We are pleased to see Governor Newsom veto SB 1047. The bill would significantly hinder AI innovation, negatively impact business growth and employment opportunities, and disrupt California's long-standing tradition of fostering open-source development. We support responsible AI regulation and are committed to continuing our collaboration with legislators to advocate for improved solutions."