Anthropic Backs California Bill for Frontier AI Transparency

Anthropic’s endorsement makes SB 53 a political lightning rod

On Monday Anthropic publicly endorsed California’s SB 53, giving the bill a rare high-profile backer as major tech groups press lawmakers to kill or weaken it. Anthropic framed the move as pragmatic: federal rules are preferable, it said, but industry can’t pause development while Washington decides.

SB 53 targets the biggest model developers — companies that meet a revenue threshold — and focuses squarely on "frontier" risks. Its stated aim is to prevent catastrophic harms, defined as events causing at least 50 deaths or over $1 billion in damages.

Key proposed requirements include:

• Develop and maintain documented safety frameworks for frontier models.
• Publish public safety and security reports before deploying powerful models.
• Provide whistleblower protections for employees who report safety concerns.
• Limit extreme misuse—such as expert-level assistance for biological weapons or orchestrating cyberattacks.

A previous draft required third-party audits, but lawmakers removed that provision in September after industry pushback. SB 53 still includes financial penalties for noncompliance and a narrower focus than earlier California bills.

The endorsement is politically significant because powerful trade groups and investors have lobbied against state-level mandates. Critics — including some venture and startup backers — argue states risk overreaching, invoking Commerce Clause concerns and the danger of driving startups out of California.

Anthropic’s blog framed SB 53 as a thoughtful, pragmatic template for AI governance. Co-founder Jack Clark reiterated that while a federal standard is preferred, the industry cannot wait for Washington as capabilities advance rapidly.

OpenAI and other companies have warned that state rules could harm competitiveness, and some former insiders criticized those warnings as misleading. The political debate also includes the Trump administration’s threats to block state AI rules, adding a federalism angle to the discussion.

If enacted, SB 53 would make what many companies already do voluntary into enforceable legal obligations for the biggest labs. That raises immediate operational questions: how to document risk thresholds, prepare pre-deployment disclosures, and protect internal reporters — all without exposing sensitive IP or undermining competitive innovation.

What this means for leaders and regulators

Companies should view SB 53 as a likely blueprint for future rules whether at state or federal level. Practical steps include aligning internal safety frameworks with public reporting needs, establishing secure whistleblower channels, and running tabletop exercises that map capabilities to the bill’s catastrophic-risk threshold.

For policymakers, SB 53 offers a constrained, impact-focused approach: address extreme misuse without attempting to govern every application of AI. That restraint may be why analysts now give the bill a stronger chance of becoming law than earlier, broader proposals.

California’s Senate still needs a final vote, and Governor Newsom has not signaled his position. But with a major lab publicly supportive and a narrower, technically grounded text, SB 53 has moved from theory to a realistic regulation that developers, investors, and regulators must plan for now.