Midas Project accuses OpenAI of violating California's AI safety law
Watchdog says OpenAI failed to publish required risk assessments for three model releases, including GPT-6 Astra.
What to know
- The Midas Project alleges OpenAI violated California's SB 53 safety law at least three times by failing to publish required risk tier assessments for GPT-5.6 preview, GPT-5.6, and GPT-6 Astra.
- SB 53 requires AI developers to publish safety frameworks and follow them; violations can incur fines up to $1 million each, scaled by severity.
- OpenAI published its Frontier Governance Framework in May but has not assigned risk tiers in any of its four required categories for any model released since then.
- OpenAI claims confidence in its compliance and says it publishes safety findings through system cards and frameworks, though it uses a separate internal Preparedness Framework for risk assessment.
“Our Preparedness Framework remains the foundation of our approach to managing the most serious risks from advanced AI.”
OpenAI, Company statement · Fortune ↗
OpenAI AI developerTyler Johnston Midas Project founderMidas Project AI industry watchdog nonprofit
Sam Altman OpenAI CEO
How it unfolded 2 developments, newest first · click a bar or a number to jump articles
-
2
OpenAI contests allegations and defends compliance
OpenAI tells Fortune it is 'confident' in its compliance with SB 53, stating it invests heavily in evaluating risks and developing safeguards. CEO Sam Altman calls for federal regulation and international AI development slowdown coordination with China.
“We invest heavily in evaluating emerging risks and developing safeguards, publicly sharing findings through our system cards and safety frameworks.”
— OpenAI spokesperson -
1
Midas Project publishes analysis alleging three SB 53 violations
The Midas Project, an AI industry watchdog nonprofit, releases an analysis accusing OpenAI of violating California's SB 53 at least three times by failing to publish required risk assessments for GPT-5.6 preview, GPT-5.6, and GPT-6 Astra. Potential penalties are up to $1 million per violation.
“California's SB 53 requires AI companies to adopt these safety policies and to follow them. It's totally up to them to choose what the rules are. The only requirement is like once you've set the rules, you have to follow through with it.”
— Tyler Johnston -
background
OpenAI debuts GPT-6 Astra without required risk tiers — OpenAI released GPT-6 Astra last week without assigning risk tiers or publishing assessments for the four required risk categories, though it did designate the model as cyber 'critical' under its separate internal Preparedness Framework.
-
background
OpenAI releases GPT-5.6 model without required risk tiers — OpenAI released the full GPT-5.6 model in July without publishing the required risk tier assessments for any of the four categories specified in its Frontier Governance Framework.
-
background
OpenAI releases GPT-5.6 preview without required risk tiers — OpenAI released GPT-5.6 preview in June without assigning risk tiers in any of the four required categories or including corresponding sections in the model's system card.
-
background
OpenAI publishes Frontier Governance Framework — OpenAI published its Frontier Governance Framework (FBF), the policy document required by SB 53. The framework requires the company to assess each new model on four risk categories—Cyber offense, CBRN, Harmful manipulation, and Loss of Control—and assign them risk tiers from one to three.
-
background
SB 53 takes effect — California's Transparency in Frontier AI Act took effect at the start of 2026, requiring the largest AI developers to publish safety frameworks explaining how they evaluate and mitigate risks.
-
background
California's SB 53 AI safety law signed into law — California's Transparency in Frontier AI Act (SB 53) was signed into law in September 2025, requiring large AI developers to publish safety frameworks and adhere to their own policies.