California’s State Assembly has advanced a significant piece of legislation aimed at regulating large-scale artificial intelligence models, potentially setting a precedent for AI oversight in the United States.
The bill, known as the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act (SB 1047), passed a key vote on Wednesday and is now poised for a final review by the State Senate before reaching Governor Gavin Newsom’s desk.
The proposed legislation aims to address the risks associated with advanced AI systems that can process massive amounts of data. It mandates that companies developing AI models requiring over $100 million in data to train must implement stringent safety measures and publicly disclose their safety protocols. These measures are designed to mitigate risks such as the potential for AI systems to be used for malicious purposes, such as disrupting critical infrastructure or facilitating dangerous technologies.
The bill’s supporters argue that it is a necessary step toward ensuring AI safety and preventing catastrophic misuse. “It’s time that Big Tech plays by some kind of rule,” said Republican Assemblymember Devon Mathis, highlighting the need for regulatory safeguards.
Major tech companies and venture capital firms have spoken out against the legislation. Critics, including tech titans like OpenAI, Google, and Meta, argue that such regulations should be handled by the federal government rather than individual states. They argue that the bill’s emphasis on developers rather than users of AI systems may not adequately address potential threats.
Senator Scott Wiener, who introduced the bill, defended it as a balanced approach to regulating powerful AI models. “Innovation and safety can go hand in hand,” Wiener said, emphasizing that the bill aims to establish fundamental safety measures without stifling technological advancement. Wiener has also criticized those who dismiss the potential risks of AI, claiming that such concerns are not unfounded.
The bill has been revised several times based on industry feedback. It no longer includes criminal penalties for noncompliance, and the state attorney general’s authority to enforce the rules has been reduced. Additionally, the bill now specifies that developers who refine existing open-source models with less than $10 million in investment are exempt from the same regulatory requirements.
The bill’s fate now depends on a final vote in the State Senate and a decision from Governor Newsom by the end of September. If passed, it could serve as a model for other states, potentially influencing federal AI policy as well.
The Associated Press, Axios and the Verge contributed to this report.
