OpenAI Proposes Civilian-Led AI Safety Rules, Diverging from White House

OpenAI has released a regulatory framework for advanced artificial intelligence that puts the company at odds with the Trump administration's recent executive order on AI safety, setting up a policy clash as CEO Sam Altman meets with White House officials and lawmakers this week.

The core disagreement centers on who should evaluate advanced AI systems for potential risks and whether such reviews should be mandatory.

Civilian versus intelligence oversight

OpenAI's proposal calls for the federal government to require mandatory evaluations of advanced AI models, with oversight responsibilities falling to the Center for AI Standards and Innovation, an office within the Commerce Department's National Institute of Standards and Technology.

This stands in direct contrast to the White House executive order issued Tuesday, which established a voluntary framework for evaluating advanced AI systems for cybersecurity risks under the leadership of the National Security Agency.

Chris Lehane, a top OpenAI executive, told reporters Wednesday that while the company views Trump's order as validation of regulatory efforts, OpenAI will push for CAISI to take the lead role. He noted that OpenAI, Anthropic, and other major AI companies already share information about new models with CAISI, while no such relationship exists with the NSA.

Concerns about classified benchmarks

OpenAI also expressed reservations about the White House plan to develop AI model benchmarking criteria on a classified basis. Lehane suggested this approach could create uncertainty for companies trying to determine when their models will trigger government scrutiny.

"I think one of the items here is when do you hit the capability threshold?" Lehane said, indicating the company wants clear, transparent criteria.

According to details first reported by Politico, Altman is meeting with White House officials and lawmakers from both parties Wednesday to discuss OpenAI's preferred approach. The executive order gives agencies 60 days to finalize implementation details, leaving a window for the company to influence the outcome.

Why it matters

The dispute highlights a fundamental tension in AI governance: whether advanced systems should be evaluated primarily through a national security lens or a civilian regulatory framework. OpenAI's push for mandatory civilian oversight could shape how the U.S. government approaches AI safety for years, particularly if Congress moves to codify evaluation requirements into law. The outcome will determine not just which agencies have authority, but whether AI safety reviews remain voluntary or become enforceable requirements.

While OpenAI advocates for mandatory evaluations, Lehane stopped short of supporting pre-release approval requirements for new models, arguing that a robust evaluation framework must be established first before considering more aggressive regulations.

These details were first reported by Politico.