Anthropic Endorses California’s Ai Safety Bill, Sb 53

On Monday, Anthropic announced an charismatic endorsement of SB 53, a California measure from authorities Senator Scott Wiener that would enforce first-in-the-nation transparency requirements connected nan world’s largest AI exemplary developers. Anthropic’s endorsement marks a uncommon and awesome triumph for SB 53, astatine a clip erstwhile awesome tech groups for illustration CTA and Chamber for Progress are lobbying against nan bill.

“While we judge that frontier AI information is champion addressed astatine nan national level alternatively of a patchwork of authorities regulations, powerful AI advancements won’t hold for statement successful Washington.” said Anthropic successful a blog post. “The mobility isn’t whether we request AI governance—it’s whether we’ll create it thoughtfully coming aliases reactively tomorrow. SB 53 offers a coagulated way toward nan former.”

If passed, SB 53 would require frontier AI exemplary developers for illustration OpenAI, Anthropic, Google, and xAI to create information frameworks, merchandise nationalist information and information reports earlier deploying powerful AI models. The measure would besides found whistleblower protections to labor that travel guardant pinch information concerns.

The measure specifically focuses connected limiting AI models from contributing to “catastrophic risks,” which nan measure defines arsenic nan decease of astatine slightest 50 group aliases much than a cardinal dollars successful damages. SB 53 focuses connected nan utmost broadside of AI consequence — limiting AI models from being utilized to supply expert-level assistance successful nan creation of biologic weapons, aliases being utilized successful cyberattacks — alternatively than much near-term concerns for illustration AI deepfakes aliases sycophancy.

California’s Senate approved a anterior type of SB 53, but still needs to clasp a last ballot connected nan measure earlier it tin beforehand to nan governor’s desk. Governor Gavin Newsom has stayed silent connected nan measure truthful far, though he vetoed Senator Weiner’s past AI bill, SB 1047, which included galore of nan aforesaid measures.

Bills regulating frontier AI exemplary developers person faced significant pushback from some Silicon Valley and nan Trump administration, which some reason that specified efforts could limit America’s invention successful nan title against China. Investors for illustration Andreessen Horowitz and Y Combinator led immoderate of nan pushback against SB 1047, and successful caller months, nan Trump management has repeatedly threatened to artifact states from passing AI regularisation altogether.

One of nan astir communal arguments against AI information bills are that states should time off nan matter up to national governments. Andreessen Horowitz’s Head of AI Policy, Matt Perault, and Chief Legal Officer, Jai Ramaswamy, published a blog post past week arguing that galore of today’s authorities AI bills consequence violating nan Constitution’s Commerce Clause — which limits authorities governments from passing laws that spell beyond their borders and impair interstate commerce.

However, Anthropic co-founder Jack Clark argues successful a post connected X that nan tech manufacture will build powerful AI systems successful nan coming years, and can’t hold for nan national authorities to act.

“We person agelong said we would for illustration a national standard,” said Clark. “But successful nan absence of that this creates a coagulated blueprint for AI governance that cannot beryllium ignored.”

OpenAI’s Chief Global Affairs Officer, Chris Lehane, sent a letter to Governor Newsom successful August arguing that he should not walk immoderate AI regularisation that would push startups retired of California — though nan missive did not mention SB 53 by name.

OpenAI’s erstwhile Head of Policy Research, Miles Brundage, said successful a post connected X that Lehane’s missive was “filled pinch misleading garbage astir SB 53 and AI argumentation generally.” Notably, SB 53 intends to solely modulate nan world’s largest AI companies — peculiarly ones that generated a gross gross of much than $500 million.

Despite nan criticism, argumentation experts opportunity SB 53 is simply a much humble attack than erstwhile AI information bills. Dean Ball, a Senior Fellow astatine nan Foundation for American Innovation and erstwhile White House AI argumentation advisor, said successful an August blog post that he believes SB 53 has a bully chance now of becoming law. Ball, who criticized SB 1047, said SB 53’s drafters person “shown respect for method reality,” arsenic good arsenic a “measure of legislative restraint.”

Senator Wiener antecedently said that SB 53 was heavy influenced by an expert argumentation sheet Governor Newsom convened — co-led by leading Stanford interrogator and co-founder of World Labs, Fei-Fei Li — to counsel California connected really to modulate AI.

Most AI labs already person immoderate type of nan soul information argumentation that SB 53 requires. OpenAI, Google DeepMind, and Anthropic regularly people information reports for their models. However, these companies are not bound by immoderate laws to do so, and sometimes they fall behind their self-imposed information commitments. SB 53 intends to group these requirements arsenic law.

Earlier successful September, California lawmakers amended SB 53 to region a conception of nan measure which would person required AI exemplary developers to beryllium audited by 3rd parties. Tech companies person fought these types of 3rd statement audits successful different AI argumentation battles previously, arguing that they’re overly burdensome.