Trump calls the AI accord 'morally binding.' It is still voluntary.

President Donald Trump, Elon Musk and the leaders of Nvidia, Google, Meta, OpenAI and Anthropic signed a 308-word accord on Tuesday that asks those companies to check their most powerful models themselves.

Speaking to reporters after a White House lunch, Trump called the agreement "morally binding" and said, "I think I'm seeing tremendous self-policing."

House Speaker Mike Johnson told reporters the commitments were "voluntary on behalf of the industry."

One signer had already decided, on its own, to hold a model back. On Monday, OpenAI confirmed it had scrapped a planned October release of a model called GPT-6.1 Astra after internal tests found it did not meet the company's safety standards. Saachi Jain, OpenAI's head of safety systems, said the model fell short on staying within the limits it was given, and on telling the user what work it had done.

The page asks an outside evaluator to check whether a company's controls, monitoring and detection "are operating as intended." It sets no fine, names no deadline and gives that evaluator no power to block a release. Withholding Astra was OpenAI's choice, and nothing on the page would have stopped the company from shipping the model instead.

What the page asks them to check

The text says each company "should implement" four layers of controls.

The first monitors a model's capabilities and alignment while it is being trained and after it is deployed, around cybersecurity, biosecurity and chemical threats, and it is supposed to ensure a model does not hack or access technical systems in unintended ways.

An internal team is supposed to check that the controls, the monitoring and the detection are operating as intended, and to fix what it finds. An independent outside auditor or evaluator is supposed to run the same check. The page names no firm, no standard and no date for that review.

An independent committee of the board is supposed to oversee the work, receive the reports and see that problems are fixed. The accord does not require the reports to be made public.

"Over time, it may make sense to codify these steps into laws or regulations," the text says. It adds that "regardless of whether this is required of companies," they are committed to the controls and the audits.

Google chief executive Sundar Pichai, Meta chief executive Mark Zuckerberg, Nvidia chief executive Jensen Huang, Anthropic chief executive Dario Amodei and OpenAI president Greg Brockman signed, along with Elon Musk.

A reader who holds the fund that tracks the Nasdaq-100, AQQQ, owns Nvidia, Google's parent Alphabet, and Meta through it. Nvidia is the fund's largest holding. Together the three are 17.5% of the fund.

Microsoft chief executive Satya Nadella and Amazon founder Jeff Bezos were at the lunch. Their names are not on the signed page.

"We all need to work together to make sure that we can win, and we can win safely," Amodei told reporters outside the White House.

The page is called the White House Accord on Super Intelligence. The same day, Trump signed an order telling federal agencies to say "Super Intelligence," and "SI," instead of "artificial intelligence" and "AI," in papers that are not laws. The assistant to the president for science and technology has 60 days to propose a federal definition and the steps that would go with it.

The text says the participating companies will meet regularly to establish standards and best practices, and it does not say when.