By Khari Johnson, CalMatters

Despite numerous proposals aimed at regulating artificial intelligence in California, one bill has drawn significant ire from major tech companies and investors: the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act.

In letters to lawmakers, Meta expressed that the legislation, Senate Bill 1047, could “deter AI innovation in California at a time where we should be promoting it.” Google echoed this sentiment, suggesting that the bill would render California one of the least favorable places for AI development. A letter from over 130 startup founders and Y Combinator warned that the bill’s “vague language” could “kill California tech.”

Notably, prominent AI researchers are divided on the issue. Yoshua Bengio and Geoffrey Hinton, often referred to as the “godfathers of AI,” support the bill, while Stanford professor Fei-Fei Li, dubbed the “godmother of AI,” opposes it.

Having passed the state Senate with a 32-1 vote in May, the bill must now clear the Assembly Appropriations suspense file and gain final approval by August 31 to reach Governor Gavin Newsom.

Introduced by San Francisco Democrat Scott Wiener, the bill encompasses several key provisions:

• Mandating that developers of high-cost AI tools assess their potential to facilitate attacks on public infrastructure or create weapons of mass destruction.
• Establishing CalCompute, a public cloud resource to support AI development and research.
• Providing protections for whistleblowers working on advanced AI technologies.

These whistleblower protections have garnered support from individuals like Daniel Kokotajlo, a former OpenAI employee, who believes the bill promotes transparency and democratic governance in AI, a field he describes as “completely unregulated.”

Kokotajlo resigned from OpenAI earlier this year, alleging violations of safety protocols. He contends that had SB 1047 been in place, it might have prevented the safety violation he observed during an early deployment of an OpenAI model.

When questioned about Kokotajlo’s claims, OpenAI stated that their policies protect employees’ rights to raise concerns.

While existing laws protect whistleblowers from retaliation in cases of state law violations, SB 1047 would extend protections to those reporting AI models capable of causing significant harm. The bill also prevents employers from obstructing the disclosure of related information.

In June, the Assembly Privacy and Consumer Protection committee recommended expanding whistleblower protections, responding to concerns raised by Google and OpenAI employees.

“Employees must be able to report dangerous practices without fear of retaliation.” – Assemblymember Rebecca Bauer-Kahan

Despite the pushback from tech giants, the bill’s whistleblower provisions have received little attention in their opposition letters. Critics have focused on the bill’s AI testing requirements, arguing that compliance costs could hinder startups and small businesses, potentially harming California’s economy.

However, supporters argue that the bill’s restrictions apply only to systems costing over $100 million, suggesting that most startups will not be affected.

Opponents warn that the bill could discourage the release of open-source AI models, which are vital for startups. Meta has claimed that developers may hesitate to release AI software due to fears of liability.

“If we over-regulate, we can put ourselves in a perilous position.” – Governor Gavin Newsom

Open-source software has played a crucial role in AI development, with companies like Google and Meta releasing influential models. Meta’s Llama model, for instance, has been downloaded over 300 million times.

Ion Stoica, a professor at UC Berkeley, predicts that if SB 1047 passes, open-source models from countries like China may surpass those developed in the U.S.

Sunny Gandhi from Encode Justice believes the backlash against the bill stems from tech firms’ unfamiliarity with accountability for their products.

Wiener has refuted claims that the bill could lead to criminal charges against software developers, emphasizing that it only applies to major AI labs.

As the debate continues, the future of AI regulation in California remains uncertain, with stakeholders on both sides advocating for their interests.

This article was originally published by CalMatters.