Six AI leaders sign a White House pledge to police their own models

Google's Sundar Pichai, Meta's Mark Zuckerberg, Anthropic's Dario Amodei, Nvidia's Jensen Huang, OpenAI president Greg Brockman and Elon Musk signed a one-page "White House Accord on Super Intelligence" on Tuesday, September 29, promising to monitor and audit their own frontier models. President Trump, who signed alongside them, called it "morally binding".

The same day he signed an executive order telling federal agencies to stop writing "artificial intelligence" and use "Super Intelligence", or "SI", instead.

Four layers of control, and no referee

The accord describes four layers. Each company should "implement robust internal controls to monitor the capabilities and alignment of its models during training and deployment", naming cybersecurity, biosecurity and chemical threats, and should ensure its models "do not hack or access technical systems in unintended ways". An internal team checks that those controls work. An independent external auditor or evaluator assesses them. A committee of the board receives the reports and makes sure problems are fixed.

Then comes the sentence that tells you how far this goes: "Over time, it may make sense to codify these steps into laws or regulations." The companies say they "will meet regularly to establish standards and best practices". The text sets no deadline and no penalty, and gives no government body a role in testing models.

Vice President JD Vance made the logic plain. "The solution to some of the AI risks is for you guys to take the risk seriously, not to come to the government for a regulatory regime that may make things worse if it's not smart and careful," he said, and he poured cold water on the idea of an FDA- or FAA-style regulator for frontier models. Musk's summary of the deal was "joint monitoring, board special committees, just generally grading each other's homework".

A clause written for the last few months

Read the first commitment again. The line about models that "hack or access technical systems in unintended ways" is not hypothetical. Nextgov noted that the accord came months after an OpenAI model escaped its testing environment and hacked into Hugging Face. That clause is the most concrete line in the document, and it describes a failure the industry has already had.

So does the accord change what the labs do? For some signatories, less than it sounds. OpenAI published a framework on September 28 saying structured safety documentation "should be required before continuing any frontier reinforcement learning training run", reviewed by senior leaders who can each veto the run, with auditors given access to check the claims. That already covers much of the accord's first three layers. The fourth, a board committee that receives the audit reports, is the piece that could bite, if boards choose to use it.

Amodei signed but told reporters the pact was only "a start", Axios reported. Huang offered the administration's view in one line: "Trust and innovation are not in conflict. Safety is how trust is earned."

Renaming AI without changing what it covers

The executive order is stranger. It argues that "Super Intelligence" "more appropriately captures the promise, potential, and rapidly advancing capabilities of these technologies", and requires agencies to use the term in correspondence, public communications, websites, reports and policy documents. The President's science and technology adviser has 60 days to propose legislative language for a federal definition.

Here is the detail most coverage skipped. Until then, the order defines "Super Intelligence" as exactly the systems already covered by the existing statutory definition of "artificial intelligence". Nothing new comes under federal rules. Every system that counted as AI on Monday simply gets a grander name.

The rebrand did not come from nowhere. The White House's September 25 fact sheet on the Trump-Xi summit said the two leaders "agreed to use the term 'super intelligence' rather than 'artificial intelligence'" and set up a US-China Super Intelligence Dialogue. On Tuesday the White House also launched America.gov, a single entry point for federal services that it says uses "Super Intelligence (SI)" to answer questions, with passport renewal and Medicare enrollment planned later.

Who gains, and what would make it real

The companies gain most. They get a public commitment written on their terms, and a President who praises self-policing. "I think I'm seeing tremendous self-policing. And they understand that they have to self-police," Trump said. Advocates of binding rules get one thing: a document in which the industry concedes, in writing, that laws may eventually be needed.

Whether this becomes more than a photo opportunity depends on details the accord leaves blank. Watch for the first signatory to name its external auditor, and for any board committee to publish what its reports found. The standards group that "will meet regularly" is the other test: either it produces rules that bind companies outside the room, or it stays a club of six.