OpenAI wants tougher AI rules than Trump just gave it

OpenAI is asking for more regulation than the White House is willing to give it.

Last week, Trump signed a scaled-back AI executive order that leans heavily on voluntary cooperation between the government and the big labs. This week, OpenAI published a policy paper arguing the federal government should go further and require the most powerful AI models to be evaluated by a federal agency before they're released to the public.

That agency is CAISI, the Center for AI Standards and Innovation. It sits inside NIST and was set up to test frontier models for safety and security risks. OpenAI already shares data with CAISI voluntarily, but now the company wants that arrangement made mandatory for everyone building at the frontier.

Chris Lehane, OpenAI's head of global policy, told Politico the Trump order was a "validation" of the safety work OpenAI has been doing, and said CAISI has "developed into something" capable of running serious evaluations of advanced models. In its policy paper, the company spelled out the ask more directly: policymakers should require frontier models to go through a CAISI evaluation before public release, with the agency recommending mitigations rather than approving or denying launches.

This is the part most readers wouldn't expect. OpenAI is the company everyone assumes wants Washington to stay out of its way. Instead it's openly asking for a tougher regime than the one the White House just delivered.

The skeptical read is that OpenAI is asking for the kind of regulation it already complies with, which would mostly inconvenience everyone else. Sanchit Vir Gogia, chief analyst at Greyhound Research, put it bluntly: "Governance of this shape can become a moat dressed as maturity." He also noted the CAISI model is really an "evidence-producing regime rather than an accountability-producing one," meaning it makes labs more visible to the government without necessarily making them more answerable for what their models do.

There's also a practical problem. The Institute for Progress estimates CAISI would need at least $84 million a year to actually carry out this kind of mandate. Its current operational budget is closer to $20 million. The agency OpenAI wants to be its overseer can't really do the job at its current size.

OpenAI isn't alone in pushing the safety conversation, either. Last month Sam Altman and Anthropic's Dario Amodei signed a joint letter calling for stronger guardrails on AI-assisted biosecurity risks, one of the rare moments the two CEOs have agreed on anything in public. The frontier labs are starting to sound similar when it comes to what they want from Washington.

INTO THE VALLEY

OpenAI asking for tougher rules sounds like a company breaking ranks with its own interests, until you remember that the company already passes the test it's proposing. A rule that says "every frontier lab must be evaluated by CAISI" is a rule OpenAI complies with on Tuesday and a rule a smaller competitor spends a year preparing for. The cleaner version of this story isn't that OpenAI suddenly wants to be regulated. It's that OpenAI figured out which regulation it can live with, and decided it would rather help write that one than wait to see what California, New York, and twenty other states write instead.