Anthropic has thrown its support behind SB 53, a California proposal designed to set transparency and safety baselines for the most powerful AI systems. The endorsement is striking in a debate where many industry groups, including the Consumer Technology Association and Chamber of Progress, have argued the bill would slow innovation. Coming from a frontier model developer, the move signals a pragmatic willingness to accept guardrails for high‑risk AI, especially as federal action remains uncertain.
The company framed its support as a recognition that advanced models are accelerating faster than consensus policymaking. While it favors federal standards over a patchwork of state rules, Anthropic’s message is clear: waiting for Washington could leave critical gaps. SB 53, in its view, is a workable way to formalize practices responsible labs already use.
What SB 53 would require
SB 53 targets “frontier” developers—think OpenAI, Anthropic, Google, and xAI—by requiring documented safety frameworks and public safety and security reports before deploying high‑capability models. The aim is to make pre‑deployment risk assessments routine rather than discretionary.
The bill centers on preventing catastrophic risks, defined as events that could cause at least 50 deaths or more than $1 billion in damages. That framing directs attention to concrete misuse scenarios, such as expert‑level biological threat assistance or high‑impact cyberattacks, rather than everyday harms like deepfakes or model sycophancy.
SB 53 also adds whistleblower protections so employees can flag safety concerns without retaliation. And to avoid sweeping in small companies, it focuses on the largest players—those with more than $500 million in gross revenue—reflecting the reality that extreme capability and deployment scale are concentrated in a handful of firms.
Why Anthropic’s endorsement matters
Industry alignment has been elusive on AI safety regulation. By backing SB 53, Anthropic is effectively saying the bill’s compliance burden is manageable and worthwhile. The company already publishes model cards and red‑team results; codifying such disclosures would turn voluntary norms into enforceable obligations, with penalties for noncompliance.
The endorsement could also alter the political math. Lawmakers frequently hear that state rules will chill investment. A prominent developer supporting state‑level accountability undercuts the narrative that any regulation is fatal to competitiveness, and it may embolden a coalition of researchers, civil society groups, and responsible‑AI teams pushing for concrete safeguards.
Opposition and constitutional questions
Trade associations and venture investors have warned that state mandates will fracture the regulatory environment and expose firms to inconsistent obligations. Policy leads at Andreessen Horowitz, Matt Perault and Jai Ramaswamy, recently argued that several state AI bills risk violating the Constitution’s Commerce Clause by burdening interstate commerce.
OpenAI’s global affairs chief, Chris Lehane, urged California not to adopt measures that could push startups out of the state, though his letter did not name SB 53. That position drew sharp pushback from former OpenAI policy researcher Miles Brundage, who said the concerns mischaracterized the bill’s scope. The text is explicit: it zeroes in on the largest companies, not early‑stage startups.
How SB 53 evolved from earlier efforts
California’s previous frontier AI bill, SB 1047, was vetoed after sustained criticism from parts of the tech ecosystem. SB 53 is narrower. Lawmakers recently removed a requirement for mandatory third‑party audits, addressing one of industry’s biggest objections about operational burden and confidentiality.
That recalibration has earned cautious praise from some policy experts. Dean Ball at the Foundation for American Innovation—an earlier critic of SB 1047—called SB 53 more technically grounded and restrained, increasing its chances of becoming law. The bill’s drafters also drew on an expert panel convened by the governor, co‑led by Stanford’s Fei‑Fei Li, to align obligations with what labs can realistically implement.
The stakes for California and beyond
California is home to most of the world’s leading AI labs and the largest concentration of AI talent. Standards set in Sacramento often ripple outward; privacy law is the obvious precedent. If SB 53 passes, it could become a template for other jurisdictions or a reference point as federal agencies refine guidance like the NIST AI Risk Management Framework.
Anthropic co‑founder Jack Clark has argued that the sector cannot wait for perfect federal consensus while model capabilities advance. In that light, SB 53 is less a ceiling than a floor—establishing baseline obligations for risk analysis, transparency, and internal escalation before the next generation of frontier systems arrives.
The legislative path is not finished. A final vote remains, and the governor has not indicated a position after previously vetoing SB 1047. But with a major lab now publicly endorsing SB 53—and with the bill scoped to the highest‑risk actors—the center of gravity in California’s AI debate may be shifting toward codified, enforceable safety norms.