AI Labs Sign a White House Pledge to Face Outside Audits

The leaders of six of the biggest AI companies signed a safety pledge at the White House on Tuesday. The White House Accord on Super Intelligence was signed by President Trump, Google’s Sundar Pichai, Anthropic’s Dario Amodei, Meta’s Mark Zuckerberg, OpenAI president Greg Brockman, xAI’s Elon Musk and Nvidia’s Jensen Huang, after a lunch with nearly two dozen technology executives. It is short, a little over 300 words, and voluntary. But its central idea is new for this administration: the companies’ safety controls should be checked by someone outside the company.

The document, subtitled a “Joint Commitment on Frontier Responsibilities”, sets out four layers. Each company building frontier models will run robust internal controls that monitor the capabilities and alignment of its models during training and deployment, in areas including cybersecurity, biosecurity and chemical threats. It will empower internal teams to make sure those controls actually work. It will partner with an independent external auditor or evaluator to test them. And it will designate an independent committee of its board of directors to receive reports from both the internal teams and the auditors. The signatories also agreed to meet regularly to set best practices for the industry, according to Nextgov/FCW.

What the accord does not contain is just as telling. It names no auditor and sets no standard for what an audit must cover. It specifies no penalties for a company that falls short, noting only that codifying the rules “may make sense” at some point. Trump called it “morally binding” and said he would appoint someone to oversee it. “I think I’m seeing tremendous self-policing,” he told reporters, The Federal reports, adding that the companies “understand that they have to self-police.” Vice President JD Vance used the same event to rule out an FDA- or FAA-style regulator for AI, arguing that the FTC and the Justice Department already have the powers needed to deal with consumer harm.

The timing explains the shift. The pledge comes days after OpenAI paused its top models after one escaped a test sandbox, and after a run of disclosures about AI agents probing US government websites. “The technology has very real risks,” Amodei said. Zuckerberg called the accord “a meaningful step forward.” Musk was more dry about it, describing the arrangement of joint monitoring, boards and special committees as companies “grading each others’ homework.” Robin Jia, a computer scientist at USC, warned that as models improve, the consequences of failures like these “are only going to get greater.”

The accord shared the stage with an executive order that got more attention. Trump directed every federal department and agency to use “Super Intelligence”, or SI, in place of “artificial intelligence” in official documents and communications wherever the law allows. The White House fact sheet argues the new term better reflects what the systems can now do, and gives the president’s science and technology adviser 60 days to propose a federal definition. Senator Mark Warner was unimpressed: “Changing what we call artificial intelligence does nothing to address the very real risks.”

For the labs, signing costs little. Anthropic, OpenAI and Google already publish safety frameworks and bring in outside testers such as the US and UK AI security institutes. The new part is the board committee and the promise that an external party will check whether the controls work, not just whether a model passes a test. How much that means depends on the details the accord leaves out: who the auditors are, what they see, and whether anyone outside the company ever reads their reports. The companies also agreed to put more money into local schools and to cut energy costs near their data centres, a nod to the growing local opposition to new AI campuses.