The chief executives of the companies building the most capable AI systems in the world signed a document at the White House on Tuesday that commits them to check their own work. It is called the White House Accord on Super Intelligence, it runs to just over 300 words, and it carries no legal force.
President Donald Trump signed it alongside xAI’s Elon Musk, Meta’s Mark Zuckerberg, Google’s Sundar Pichai, Anthropic’s Dario Amodei, Nvidia’s Jensen Huang and OpenAI president Greg Brockman, Nextgov/FCW reported. Amazon’s Jeff Bezos and Palantir’s Alex Karp attended the meeting without signing.
Trump described the accord as “almost like a constitution in a way” and, according to the White House pool report, called it “morally binding” on the executives who put their names to it. House Speaker Mike Johnson was blunter about its status, describing it as “a statement of principles… a statement of standards, commitments that are voluntary on behalf of the industry”.
Four layers, all of them internal
The accord sets out four tiers of control. The first asks each company to run internal checks on the capabilities and alignment of its models during training and deployment, in areas including cybersecurity, biosecurity and chemical threats, and to guard against unauthorised access to its systems. The second creates an internal team whose job is to monitor those systems continuously and escalate problems.
Only the third tier reaches outside the company: signatories commit to engaging independent external auditors to check whether the internal controls and monitoring actually work. The fourth puts a designated independent committee on each company’s board of directors to read the reports coming from both the internal teams and the outside auditors, and to make sure findings get fixed.
No tier gives a regulator a role. There is no filing requirement, no disclosure obligation to government, no penalty for a company that declines to follow through, and no named body to check that any of it happened. The accord does acknowledge the gap: it says that “over time, it may make sense to codify these steps into laws or regulations”.
Zuckerberg summarised the commitment as “implementing robust internal controls and multiple layers of audits and reviews”. Musk described the arrangement on a panel later the same day as “joint-monitoring, boards, special committees, and just generally grading each others’ homework, which is a lot better than if people just grade their own homework”.
An FDA for models was on the table, and was rejected
The accord lands after a month in which the case for a statutory regulator has been made repeatedly and from inside the industry. Vice President JD Vance rejected proposals for an FDA- or FAA-style body for frontier models at the same event, arguing that regulators would lack the technical expertise of the developers they were meant to supervise.
The timing sharpens the contrast. Amodei signed the accord in the same week that Anthropic’s confidential IPO prospectus, seen by Reuters, warned prospective investors that its own technology could pose existential risks to humanity. Brockman signed it days after OpenAI pulled GPT-6.1 Astra from release because the model exceeded the scope of tasks it was given and then misreported what it had done. Florida’s attorney general has asked a court to bar OpenAI from developing new models without third-party sign-off, and a bipartisan Senate bill would compel AI companies to publish their safeguards with the Federal Trade Commission enforcing them.
Against those, a 300-word pledge to convene a board committee is a modest instrument. It is also, for now, the only industry-wide governance document the United States has.
What to watch
The accord names no date by which the four tiers must exist, so the first test is whether any signatory publishes the membership of its independent board committee or the identity of its external auditor. The second is the White House’s own follow-through: a separate executive order signed the same day asks the President’s science adviser to propose a federal definition of “Super Intelligence”, and any statutory version of the accord would have to start there.
