SB 53 Splits AI Labs: OpenAI vs Anthropic on Regulation

Silicon Valley’s Regulatory Crossroads

Senate Bill 53 represents a significant shift in California’s approach to AI governance, moving from the broad liability clauses of its vetoed predecessor, SB 1047, to a more focused transparency model. Authored by State Senator Scott Wiener and shaped by a governor-convened expert panel, the bill implements a “trust but verify” framework, as described by supporter Anthropic. It avoids prescriptive mandates, instead requiring large AI developers to create and publicly disclose their own safety frameworks for managing “catastrophic risks.” According to reporting from Transformer News, this pivot directly addresses criticisms that led to the previous bill’s failure.

Core mandates include releasing public reports on risk assessments before deploying new models and reporting critical safety incidents to the state within 15 days of occurrence. The legislation also introduces a two-tiered reporting system; companies with less than $500 million in annual revenue submit high-level summaries, while those exceeding that threshold must provide more comprehensive reports, as detailed by hyper.ai. This structure is designed to hold companies publicly accountable to their own safety promises, a direct response to criticisms that the previous bill was overly rigid and applied “stringent standards” regardless of risk level, a point highlighted by TechCrunch. The bill also establishes clear whistleblower protections for employees and contractors who report violations or other substantial dangers.

However, the bill’s verification mechanisms were weakened during negotiations. An initial provision requiring independent, third-party audits of safety claims was removed from the final version. As a result, enforcement now relies more heavily on whistleblowers and state-led investigations, shifting the balance of the “trust but verify” model more toward trust.

Tech Titans’ Battle Lines

The reaction to SB 53 has fractured the AI industry, creating a clear divide in California’s tech ecosystem. On one side, a coalition including OpenAI and Andreessen Horowitz argues against a “patchwork of state regulations,” suggesting it would create duplicative compliance burdens and stifle innovation. Andreessen Horowitz has gone further, raising constitutional challenges by arguing that state-level rules infringe on the federal government’s authority over interstate commerce. In a letter to Governor Newsom, OpenAI argued that companies meeting federal or EU standards should be deemed compliant with state rules to avoid a confusing regulatory landscape.

In stark contrast, San Francisco-based Anthropic became the first major AI lab to endorse the bill. The company stated that SB 53’s requirements would “formalize practices that Anthropic and many other frontier AI companies already follow.” In a statement reported by KQED, CEO Dario Amodei noted that his company has managed to conduct rigorous safety testing while remaining a fast-growing business, arguing that other large labs can and should be held to the same standard. While stating a preference for a federal standard, Anthropic supports SB 53 because, as the company put it, “powerful AI advancements won’t wait for consensus in Washington.”

Newsom’s Pivotal Decision

Despite the high-profile opposition, some policy experts believe the bill has a strong chance of becoming law. Dean Ball, a former White House AI advisor, expressed “roughly 75% confidence, that SB 53 will be signed into law,” largely because its narrower focus aligns with the recommendations from the governor’s own expert panel. Still, a veto remains a distinct possibility. Governor Newsom has a history of vetoing a significant number of bills over cost concerns, a factor that could come into play with SB 53’s provision to create a public compute cluster, as noted by GovTech. His decision will ultimately signal California’s direction in the global effort to balance AI innovation with public safety.