OpenAI Urges California to Strengthen SB 53 AI Safety Bill in Major Policy Reversal

TL;DR

• OpenAI has formally urged California lawmakers to strengthen SB 53, the state's frontier AI safety bill, marking a sharp reversal from its opposition to the similar SB 1047 last year.
• The company is now calling for tougher requirements including mandatory public safety frameworks, third-party audits, and stronger whistleblower protections, arguing voluntary commitments are no longer enough.
• If passed in its strengthened form, SB 53 could set a de facto national standard for AI governance and pressure Congress to act on federal AI regulation.

A Surprising Letter to Sacramento

In a major shift that has caught Sacramento and Silicon Valley off guard, OpenAI is no longer fighting California's push to regulate frontier AI — it's asking lawmakers to go further. This week, the ChatGPT maker sent a letter to the authors of SB 53 urging them to strengthen the bill's safety provisions before it heads for a final vote, a complete turnaround from its high-profile campaign to kill California's previous AI safety bill, SB 1047.

The move signals a new calculation inside the world's most valuable AI company: that embracing — and even shaping — stricter state-level regulation may be preferable to facing a patchwork of weaker laws or a future federal crackdown it can't influence.

What SB 53 Actually Proposes

SB 53, formally known as the Transparency in Frontier Artificial Intelligence Act and authored by Senator Scott Wiener, is the successor to the controversial SB 1047 that was vetoed by Governor Gavin Newsom in 2024. While SB 1047 focused on holding developers liable for catastrophic harms, SB 53 takes a different approach centered on transparency and accountability.

As currently written, the bill would apply to companies developing the largest frontier models — those trained using more than 10^26 integer or floating-point operations and costing over $100 million to train — which today includes OpenAI, Google DeepMind, Anthropic, and Meta.

Key provisions include requiring developers to publish a comprehensive safety and security framework detailing how they assess and mitigate risks from chemical, biological, radiological, nuclear, and cyber capabilities; to publish redacted safety reports before releasing a new frontier model; to report critical safety incidents to the California Attorney General within 72 hours; and to establish internal whistleblower protections and anonymous reporting channels for employees who flag safety concerns. The bill would also create CalCompute, a public cloud infrastructure initiative to support academic and startup AI research.

Why OpenAI Flipped Its Position

Last year, OpenAI was among the most vocal opponents of SB 1047, arguing it would stifle innovation, drive talent out of California, and create legal uncertainty for open-source developers. The company favored voluntary safety commitments and federal regulation over state-by-state rules.

So why the reversal? In its letter to Senator Wiener this week, OpenAI's leadership framed the change as a response to how quickly the technology has advanced.

According to the company, three factors drove the new stance. First, frontier capabilities have leapt forward in the past 12 months, with reasoning models and autonomous agents making catastrophic risk scenarios more plausible and harder to handle with voluntary guidelines alone. Second, OpenAI says it has learned that without legally mandated transparency, public trust in AI will erode, which is ultimately bad for adoption. Third, the company now believes a strong, clear California standard is better than a fragmented landscape of inconsistent state laws.

Crucially, OpenAI is not just supporting SB 53 — it wants to make it tougher. The company is pushing for amendments that would require independent third-party audits of safety frameworks rather than self-assessments, mandate more detailed public disclosures about pre-deployment testing and post-deployment monitoring, and codify stronger anti-retaliation protections for whistleblowers that would be enforceable by the state labor commissioner.

A Stark Contrast to the SB 1047 Fight

The tone is a world away from the 2024 battle over SB 1047. That bill sparked a civil war in the AI industry, with Anthropic offering qualified support after amendments while OpenAI, Meta, and venture capital firms like Andreessen Horowitz lobbied heavily for a veto. Governor Newsom ultimately vetoed SB 1047, calling it overly restrictive and saying it failed to account for the different levels of risk posed by different AI systems.

SB 53 was deliberately designed to be more palatable to industry by focusing on transparency rather than liability. Even so, many industry groups have continued to push for a weaker version or for federal preemption that would block California from acting at all. OpenAI's decision to break ranks and call for a stronger bill isolates those opponents and gives crucial political cover to Wiener and co-sponsors.

Supporters of stronger regulation, including the Center for AI Safety and several former OpenAI employees who have spoken out about safety culture, have welcomed the reversal but also expressed caution, noting that the details of any audit and disclosure requirements will determine whether the law has real teeth.

What Stronger Regulations Could Mean for the Industry

If SB 53 passes with OpenAI's proposed amendments, the impact would extend far beyond California's borders. Because nearly every major frontier lab is headquartered in California or does substantial business there, the law would effectively become a national standard — similar to how California's auto emissions and data privacy laws have shaped corporate behavior nationwide.

For AI companies, a strengthened SB 53 would mean an end to the era of voluntary safety reports and closed-door testing. Mandatory, audited disclosures would make it much harder to rush a model to market without thorough safety evaluation and would give researchers, journalists, and competitors unprecedented insight into how frontier models are trained and tested.

For startups and open-source developers, the bill's high compute and cost thresholds mean most smaller players would be exempt, though critics still worry about compliance costs trickling down. For the public, it could mean more confidence that the most powerful AI systems are being built with oversight — and a clearer legal path for employees to speak up without fear of retaliation.

It could also accelerate pressure on Washington. OpenAI has long said it prefers a single federal framework, and by endorsing a strong California law, it may be trying to force Congress to act to avoid a patchwork of 50 different state AI safety regimes.

What's Next for SB 53

SB 53 has already cleared the State Senate and is now moving through the Assembly, where it faces a key vote in the Appropriations Committee before a full floor vote expected by the end of August. Governor Newsom, who vetoed SB 1047, has not yet taken a public position on the new bill but has said he supports "responsible AI regulation that maintains California's leadership in innovation."

With OpenAI now lobbying for a stronger version, the bill's odds of passage have significantly improved. The remaining fight will be over the exact language of the amendments — how intrusive the audits will be, how much of a safety report can be redacted for competitive reasons, and how the law will be enforced.

Whether OpenAI's reversal is a genuine commitment to safety or a strategic move to shape regulation on its own terms, one thing is clear: the politics of AI governance in California have fundamentally changed, and the outcome of SB 53 will likely define how America regulates artificial intelligence for years to come.

AndroGuider Team

Articles written by the AndroGuider team. We try to make them thorough and informational while being easy to read.