The California state legislature has approved a controversial artificial intelligence safety bill that requires companies to ensure their technologies do not cause significant harm.
The bill, named SB1047, has received both criticism and praise from a series of well-known figures in the US. Now, the bill will return to the state Senate for a confirmation vote and may be amended, and then it will be submitted to Governor Gavin Newsom. The bill has already passed in the state House of Representatives with 41 votes in favor and 9 against.
The SB1047 bill stipulates that companies developing powerful AI models must take'reasonable cautionary measures' to ensure their technologies do not cause'serious harm', such as mass casualties or property losses exceeding $500 million. This bill was proposed by California state senator Scott Wiener and passed in the state Senate in May this year. The bill will require companies to take preventive measures, such as introducing a 'kill switch' that can shut down their technology at any time. The bill also requires models to be submitted to third-party testing to ensure that serious risks are minimized.
In addition, the bill will also provide whistleblower protection for AI company employees who wish to share safety concerns. Companies that do not comply with the bill may be sued by the California attorney general.
Innovation and safety can go hand in hand, and California is leading the way, Wiener said in a statement. With this vote, the state House of Representatives has taken a truly historic step and is proactively working to ensure that an exciting new technology protects the public interest as it develops. He also called the bill a'simple and easy common sense measure' that codifies the safety commitments many large companies have already made into law.
Over the past few months, the bill has been strongly opposed by many major tech leaders, startups, and venture capitalists, who believe that the bill represents excessive government interference in a still nascent technology and may stifle California's technological innovation. Last week, OpenAI publicly expressed opposition to the bill, arguing that such policies should be implemented at the federal level rather than the state level.
Politicians including US Congressman Ro Khanna and San Francisco Mayor London Breed have also expressed opposition to the bill, agreeing with the concerns of the tech industry that the bill may hinder California's leadership position in AI innovation.
But in the final stages before this vote, the bill also won the support of some important figures in the field of AI. On August 27, Elon Musk surprisingly expressed support, although he said it was a 'tough decision that will make some people unhappy'. Anthropic, a competitor of OpenAI known for its focus on safety, also cautiously supported the bill and said last week that 'the benefits may outweigh the costs' and that these provisions can be feasibly implemented. After Wiener implemented several amendments suggested by the company, Anthropic became more supportive of the bill.
In the face of critics, Wiener defended the bill, emphasizing that its provisions only apply to companies that spend more than $100 million training large models or $10 million fine-tuning models, which will exempt most smaller startups. The senator also said that while he would support federal legislation, Congress has historically been slow to act in technology regulation, and in the absence of national action, he believes California has a responsibility to take the lead.
Now, as SB1047 is submitted to Governor Newsom, the debate over the bill will continue. OpenAI, the tech incubator Y Combinator, and the venture capital firm Andreessen Horowitz (critics of SB1047) have all registered lobbyists to work on the bill.