OpenAI Backs State AI Safety Baseline As Federal Cyber Tests Near

California, New York and Illinois have advanced frontier AI safety legislation that OpenAI views as the start of a common state-level baseline, while federal cyber-testing rules for advanced models remain unfinished.

Published by OpenAI Chief Global Affairs Officer Chris Lehane, the position argues for a national frontier AI framework that combines state disclosure laws, federal testing capacity and international standards rather than leaving labs to set safety rules alone.

State AI Safety Bills Define The Baseline

Across the three states, the proposals share core elements: documented safety frameworks with risk assessments, public disclosure of those assessments and serious-incident reporting.

Independent, objective audits are also part of the accountability structure.

Lehane characterizes that alignment as "reverse federalism," with states moving in similar directions before a federal statute is complete.

In that outline, California is establishing the core disclosure framework, New York is showing that the approach can move across jurisdictions, and Illinois is adding independent verification for key disclosures.

Support from OpenAI remains conditional.

The post warns against state rules expanding into national-security decisions or highly technical model reviews that require federal expertise, classified-system access and direct work with frontier labs.

It adds that fragmented state rules could be difficult for regulators, consumers and smaller developers.

Even so, the state section leaves room for action outside frontier-safety testing.

States can continue serving as laboratories of democracy on youth protection, electricity and environmental policy, and AI education and literacy.

Federal Cyber Testing Framework Is Due In Early August

At the federal level, the post describes the Trump Administration as working with technical and national-security experts on US government testing for the most capable AI models on cyber.

The framework is expected to establish testing standards, timelines and processes.

Separately, OpenAI is engaged in discussions with the administration, peer companies, business groups and other stakeholders.

It also expresses appreciation for the administration's goal of having the cyber-evaluation framework in place by early August.

That federal section names the Center for AI Standards and Innovation, or CAISI, as the durable federal capacity that could evaluate advanced models.

Any federal legislation should consider how CAISI works with the rest of government and what role it plays in testing.

For developers of the most capable systems, the blueprint includes independent audits, incident reporting, strong security standards and whistleblower protections.

Those items are presented as part of a federal framework, not as evidence that a final bill has passed.

The blueprint gives the federal government the lead role in testing and evaluation of the most advanced systems.

Frontier AI raises national-security and public-safety questions that need national standards, national capabilities and national institutions.

In the same argument, federal and state work are presented as mutually reinforcing.

State laws will not all be identical, but OpenAI wants to work with policymakers so those laws strengthen safety while maximizing economic benefits.

Global Standards Depend On US Legislation

OpenAI links national legislation to a US-led international framework for AI standards.

The idea was discussed several weeks ago at the G7 with Brazil, Egypt, India, Kenya and Korea, where frontier-lab CEOs discussed the need for an international framework.

The section also points to Sam Altman's Financial Times proposal for a US-led international forum covering accepted standards, independent analysis of capabilities and risks, and access for participating nations and companies that follow the rules.

Google DeepMind CEO Demis Hassabis advanced related ideas in a new paper this week.

Bipartisan federal legislation is framed as the foundation for that international effort.

Congress is moving on the issue, with lawmakers in both chambers and both parties putting forward proposals for a federal framework.

OpenAI did not identify a final federal bill text.