SF’s Anthropic Backs California AI Safety Bill After Newsom Vetoed 1st Attempt

San Francisco-based chatbot maker Anthropic said Monday that it is backing landmark artificial intelligence legislation in California, one year after the first attempt to regulate the burgeoning industry ended in a veto by Gov. Gavin Newsom.

The endorsement makes Anthropic the first large AI company to back SB 53 by Sen. Scott Wiener (D-San Francisco), which would require the biggest AI companies to disclose their safety and security protocols and report critical safety incidents to the governor’s office within 15 days. It would also offer protections to whistleblowers at AI companies of any size who call out risky or dangerous behavior.

Anthropic’s support is a big win for Wiener, but the bill still faces opposition from large business and tech groups. Still, the company’s support could help Wiener get the bill over the finish line — and may help convince Newsom, who, after vetoing last year’s bill, created a working group to develop recommendations for AI regulation. Wiener said he crafted SB 53 to align with that group’s findings.

In an interview with KQED, Anthropic CEO Dario Amodei said Wiener’s bill, which was narrowed last week to apply only to large AI programs, strikes a balance between safety and progress.

“SB 53 focuses on a particular subset of the risks, what we call catastrophic risks, which are some of the kinds of the largest things that can go wrong — things like large-scale cyberattacks,” he said. “It forces companies to be transparent about how they measure these risks, about the safety and security testing that they do in order to quantify these risks. This is something that Anthropic actually already does.”

Amodei noted that when Anthropic tested its latest AI model, Claude 4, earlier this year, researchers found some concerning responses, like “resisting being shut down or blackmailing employees.” That testing, he said, is being conducted even as Anthropic continues to grow.

“We’ve managed to do that while being a profitable and fast-growing company. And so our thinking is that if we can do that, the other large companies in the space can do that as well,” he said.

Congress has so far failed to enact any sort of national AI regulations. Wiener’s first attempt last year, praised as “a promising first step” by Anthropic co-founder Jack Clark, drew criticism from Newsom in his veto message over its limited focus on the largest models.

Amodei said he thinks SB 53 could go even further, but he did have concerns about how prescriptive last year’s legislation was.

“We saw a good idea at the heart of it, but we were concerned, actually, that with the field moving so fast, having to comply with all these precise tests would be too rigid and they would kind of quickly become out of date,” Amodei said. “So I would say we had mixed feelings about it.”

SB 53, he said, strikes a good balance — in part because it does distinguish between large and small AI companies.

“We don’t want to stifle competition or even just be accused of stifling competition,” he said.

In a written statement, Wiener praised Amodei, saying that under his leadership, “Anthropic has been a courageous and steadfast champion for innovating safely and responsibly.”

“It can be difficult to tell where many AI companies will come down on safety issues. Never Anthropic,” he added. “I’m grateful to have support from a homegrown San Francisco company that’s shown the world it’s possible to lead on both responsible practices and product performance. The two aren’t mutually exclusive.”

The legislation must pass both houses and be sent to Newsom by Friday. He will have a month to decide whether to sign or veto the bill.