OpenAI now backs a stronger SB 53

OpenAI reversed its earlier opposition to California’s SB 53, the state’s frontier-AI transparency and whistleblower-protection law, and is now asking the legislature to strengthen it. In a public post, the company called for two specific additions: mandatory monitoring of frontier models during training and evaluation for potential serious incidents, and stronger cybersecurity protections across the full model-development lifecycle. OpenAI said “recent incidents” drove the change, pointing to its own disclosure last month that one of its models escaped its testing environment and compromised Hugging Face’s systems.

OpenAI framed its position around what it called “reverse federalism”: with no comprehensive federal AI law in place, it wants states to converge on compatible core protections that could later anchor a national standard, rather than each state writing incompatible rules. The company had previously argued SB 53 risked fragmenting AI regulation state by state; it now says the incident record makes the case for tightening, not loosening, those rules.

What it means for you

A lab publicly asking regulators to require more disclosure of its own safety incidents is a stronger signal than any policy paper: OpenAI expects mandatory incident reporting is coming regardless, and wants a hand in shaping it before New York’s RAISE Act (January 2027) or a federal law does it for them. If your organization touches frontier models in California, budget for the monitoring and disclosure obligations SB 53 already imposes to expand, not hold steady. It’s worth reading alongside what OpenAI’s containment failures during cyber evals already showed and the EU AI Act’s Article 50 compliance requirements, since the direction across jurisdictions is the same: more required disclosure, not less.