New York passes a bill to prevent AI-fueled disasters

In a significant move, New York state legislators approved a groundbreaking bill aimed at mitigating potential disasters linked to advanced AI technologies from major firms such as OpenAI, Google, and Anthropic. The RAISE Act, passed on Thursday, seeks to avert scenarios that could lead to catastrophic outcomes, including loss of life exceeding 100 individuals or damages surpassing $1 billion. The enactment of this legislation marks a crucial victory for advocates of AI safety, who have increasingly found themselves sidelined in recent years amidst the rush for rapid innovation pushed by Silicon Valley and previous political administrations. Esteemed figures in AI research, including Nobel laureate Geoffrey Hinton and pioneer Yoshua Bengio, have lent their support to the RAISE Act, emphasizing the urgent need for regulatory measures. If enacted, the RAISE Act would establish the first legally mandated transparency standards aimed at frontier AI laboratories in the United States. This bill shares similar objectives with California's previously vetoed AI safety legislation, SB 1047. However, co-sponsor Senator Andrew Gounardes has crafted the RAISE Act to avoid stifling innovation among startups and academic institutions, a criticism often directed at its California counterpart. "The window to implement safety measures is rapidly closing as technology evolves at an unprecedented pace," Senator Gounardes stated, highlighting the alarming likelihood of associated risks. The bill is now poised for consideration by New York Governor Kathy Hochul, who will decide whether to sign it into law, request amendments, or veto it outright. Should it be signed, the RAISE Act will mandate that the largest AI laboratories provide comprehensive safety and security reports regarding their advanced AI models. Furthermore, it obligates these labs to disclose any safety incidents, such as unusual AI behavior or instances of model theft. Failure to comply with these regulations could result in civil penalties of up to $30 million, as empowered by the New York Attorney General. The RAISE Act specifically targets the largest AI firms, regardless of their geographic location, whether based in California or overseas. Although it mirrors some aspects of SB 1047, the RAISE Act was developed to address previous criticisms. Notably, it does not require AI developers to incorporate a "kill switch" in their models or hold companies accountable for significant harms caused by AI post-training. Despite resistance from the tech industry, Assemblymember Alex Bores, another co-sponsor, remains optimistic about the bill's potential, asserting that it will not hinder technological innovation. Silicon Valley's pushback has been notable, with critics like Andreessen Horowitz's Anjney Midha labeling the RAISE Act as detrimental to U.S. competitiveness. Meanwhile, Anthropic’s co-founder Jack Clark expressed concerns over the bill's broad scope, which he fears might adversely affect smaller companies. In response to these criticisms, Senator Gounardes defended the bill, asserting that it is not intended to burden small businesses. Another concern raised is that AI developers might refrain from offering cutting-edge models in New York due to regulatory pressures, a situation that has unfolded in Europe due to stringent tech regulations. However, Assemblymember Bores reassured that the burden posed by the RAISE Act is manageable, and the state’s significant economic stature makes withdrawal unfeasible for most companies. "While I acknowledge the potential for political maneuvering, I believe there’s no substantial economic rationale for AI firms to withdraw from New York," he concluded.