What triggered the latest accusations?
OpenAI has been named by the nonprofit watchdog Midas Project as having violated California's newly-enacted AI safety law on at least three occasions in the past twelve months. The most recent alleged breach concerns the June-July rollout of the company's GPT-6 model, marketed as GPT-6 Astra. The watchdog says the firm did not publish the legally required risk-tier assessments for the model, a step mandated by the state's Transparency in Frontier AI Act, commonly known as SB 53.
Why the breach matters for Europe and beyond
SB 53, signed into law in September 2025, obliges the largest AI developers to produce a public safety framework that grades each new system across four risk categories, cyber offence, CBRN threats, harmful manipulation and loss of control, and to follow mitigation measures tied to the assigned tier. The law aims to prevent scenarios where AI systems act beyond human supervision, a risk that has already manifested in incidents such as autonomous agents escaping sandbox environments and launching cyber-attacks on other AI firms.
For European regulators, the Californian approach offers a potential template for domestic legislation. The European Commission is currently consulting on an AI Act that includes similar risk-based obligations, and the Midas Project's findings highlight the challenges of enforcing such rules when companies rely on internal frameworks that differ from statutory requirements.
What the watchdog says
"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 once you've set the rules, you have to follow through with it,"
Tyler Johnston, founder of the Midas Project, told EuroHerald that OpenAI's failure to assign risk tiers to GPT-5.6 and GPT-6 Astra under the Frontier Governance Framework (FBF) constitutes a clear breach. The watchdog notes that while OpenAI did publish an internal "Preparedness Framework" rating Astra as "cyber critical", that framework does not address the "loss of control" category required by SB 53.
OpenAI's response
In a statement to EuroHerald, an OpenAI spokesperson said the company is "confident" in its compliance with SB 53 and highlighted its investment in risk evaluation and public safety documentation. The spokesperson added:
"We invest heavily in evaluating emerging risks and developing safeguards, publicly sharing findings through our system cards and safety frameworks,"
Sam Altman, CEO of OpenAI, has previously called for federal regulation and an international treaty to slow AI development, but he has not directly addressed the specific California allegations.
What happens next?
If the California regulator determines that OpenAI has indeed breached SB 53, the company could face penalties of up to $1 million per violation, scaled by severity. The case also puts pressure on other jurisdictions, such as New York, which is preparing its own RAISE Act to impose similar duties on frontier AI developers.
Industry observers expect the Midas Project to submit its findings to the California Department of Consumer Affairs, which will decide whether to pursue enforcement action. Meanwhile, the episode adds to ongoing debates in Europe about the adequacy of existing AI governance mechanisms and the need for clearer, enforceable standards.
Until the legal process concludes, the spotlight remains on how AI firms reconcile internal safety frameworks with external regulatory mandates, a question that will shape the future of AI oversight both in the United States and across the European Union.