OpenAI Supports Federal Audits of Major AI Models, Not the Full FRONTIER Act

OpenAI has endorsed outside safety audits for the biggest AI developers, but not the full federal bill that would require them. Its position gives the FRONTIER Act support on independent review while preserving a key uncertainty: which other obligations in the proposed framework the company will accept.

Chris Lehane, OpenAI’s chief global affairs officer, expressed support for the audit provision during a Washington, D.C., roundtable with lawmakers, according to CBS News. CBS described the move as the first time OpenAI has supported a federal mandate for a third party to assess AI developers’ work.

An endorsement with limits

That distinction is more than wording. An OpenAI spokesperson said the company has not endorsed the FRONTIER Act in its entirety. The bill combines its audit requirement with reporting, risk-management, and ongoing-assessment duties, so backing one provision does not amount to backing the complete package.

What the proposed review would cover

Under the proposal, an independent organization would assess whether a developer’s safety protocols are rigorous enough to mitigate what the bill calls acceptable levels of catastrophic risk. The audit duty would apply to companies that exceed specified revenue and computing-power thresholds, rather than every AI developer.

Other duties in the bill

• Publish model cards and maintain risk-management frameworks.
• Report safety incidents and complete ongoing assessments.

A general policy becomes a specific stance

The move follows a July OpenAI policy statement by Lehane that identified independent, objective audits as part of AI governance and accountability. That statement also argued for a national standard over a patchwork of state rules, and said the most capable AI developers should face requirements including audits and incident reporting.

OpenAI has now taken that principle into a live congressional debate. It supports a rule under which the largest developers would face outside review of their safety protocols, while withholding approval from the full set of rules surrounding that review.