The controversial California AI bill that has divided the tech world

Catch up quick: California's SB 1047 would require AI developers to comply with certain rules before developing their models.

• It passed the state Senate 32-1, with the Assembly facing an Aug. 31 deadline to approve the measure in time for it to head to Gov. Gavin Newsom.

Driving the news: Anthropic has offered cautious support for the bill, following certain changes, while OpenAI opposes it, saying it could "stifle innovation."

What they're saying: Elon Musk said Monday that California "should probably pass the SB 1047 AI safety bill," arguing that AI's risk to the public justifies regulation. "This is a tough call and will make some people upset," he said.

• Musk's motivations could be more complicated than a simple desire for regulation. He previously called for a six-month pause on AI development while his own AI efforts have proceeded, often with fewer safeguards than rivals.
• "I reject the false claim that in order to innovate, we must leave safety solely in the hands of technology companies and venture capitalists," bill sponsor Sen. Scott Wiener said in a statement earlier this month. He also said he'd support a strong federal law that preempts California's.
• Landon Klein, director of US policy at the Future of Life Institute, told Axios the bill's authors landed at an impressive balance, and naysayers simply don't want to be regulated.
• "I think the concern is that if [AI companies] need to make sure that their systems are safe before they put them out, that they then hold some responsibility and some accountability for the potential harms that occur when they put those models out, and frankly, they just don't want to face that accountability," he said.

The other side: Nancy Pelosi — and several other Californian members of Congress — have urged their home state to reject the bill, with a number of industry and business groups also voicing opposition, including the U.S. Chamber of Commerce and the Software and Information Industry Association.

• In a letter first shared with Axios, tech advocacy coalition Chamber of Progress and other groups including the Bay Area Council, Context Fund, Engine, NetChoice, R Street Institute, Silicon Valley Leadership Group and TechFreedom wrote to Newsom urging a veto, calling the bill "fundamentally flawed and mistargeted" even with recent changes.
• San Francisco Mayor London Breed announced Tuesday that she is also opposed to the bill's passage.

Zoom in: The bill has seen several big changes since it was originally proposed thanks to aggressive industry feedback, including from AI firms like Anthropic.

• There are no longer any criminal penalties in the bill; civil penalties remain.
• The attorney general can no longer seek civil penalties until after a harm has occurred. There's also no longer a Frontier Model Division, which would have been a new regulatory body just for AI.
• The legal standard for ensuring developers are complying with the bill has shifted from providing a "reasonable assurance" of safety to exercising "reasonable care," a lower standard.
• If an open-source developer spends less than $10 million fine-tuning an existing model, they are not considered a "developer" under the bill and liability would rest with the original developer. This is meant to protect open source developers from burden.

Critics like OpenAI say that the federal government, not individual states, is best suited to regulate AI, while backers of the proposal note that Congress has thus far failed to act. And, should it do so later, it could pass legislation that preempts state laws.

The bottom line: As with many technology policy areas, California is determined to be the first mover while Congress drags its feet.

Editor's note: This story was corrected to reflect that, under the the bill, if less than $10 million is spent on fine-turning an existing model any liability would rest with the model's original developer.