White House AI Accord: What the 6 CEOs Actually Signed

Forbes counted the words. The White House Accord on Super Intelligence, the safety pledge six AI executives signed with President Trump on Tuesday, September 29, runs to 308 of them. I pasted the text into a counter myself and got 323, but I was including the title and subtitle, so call it roughly 300.

That’s less than half the length of this post. And it’s the document Trump described to reporters outside the West Wing as “almost like a constitution” and “morally binding.”

The signers were Google’s Sundar Pichai, Anthropic’s Dario Amodei, Meta’s Mark Zuckerberg, OpenAI president Greg Brockman, Elon Musk and Nvidia’s Jensen Huang. Plenty of other people were at the lunch, including Jeff Bezos, AMD’s Lisa Su and Microsoft’s Satya Nadella, according to the Hindustan Times, but they didn’t sign.

Nobody has said why.

What those 300 words ask for

Four “layers of controls and audits” make up the core. Each company should monitor its models during training and deployment for cyber, bio and chemical risks, and make sure they “do not hack or access technical systems in unintended ways.” It should have an internal team checking that those controls work. It should “partner with an independent external auditor or evaluator.” And a committee of its board should receive the reports and make sure problems get fixed.

The companies also say they’ll meet regularly on standards. Then comes the line I keep rereading: “Over time, it may make sense to codify these steps into laws or regulations. Regardless of whether this is required of companies,” they’re committed.

There’s no definition of a frontier model. No deadline. No word on who the auditor is, who pays them, or whether anything they find ever gets published. And there’s nothing that happens to a company that ignores it.

The hacking clause is the only part that reads as if it was written this month. It’s hard not to connect it to OpenAI’s agent getting into Australia’s Medicare portal and the model OpenAI shelved a day earlier for overstepping its instructions. OpenAI’s name is on the page. That’s either reassuring or a bit awkward, depending on your mood.

The same four ideas, with teeth

Almost every idea in the accord already exists in a bill, which is the part I found most interesting.

H.R. 9925, the FRONTIER Act, was introduced in the House on July 23 by Jay Obernolte, a California Republican, with co-sponsors from both parties. I went through the text on Congress.gov.

	White House accord	FRONTIER Act (H.R. 9925)
Who’s covered	“Every company” training frontier models, undefined	Tiers by revenue; the top tier needs over $5 billion in revenue and $10 billion in AI spending
Outside auditor	Chosen by the company	A licensed “independent verification organization” that must show independence from the AI industry
Publishing results	Not mentioned	A summary and redacted report posted publicly within 30 days
Penalty	None	Up to $1 million per violation for some sections, each day counting separately
Legal status	Voluntary	A bill, still in committee

The bill is only introduced, not passed, and bills change a lot in committee. I'm comparing texts, not predicting what becomes law. The penalty row covers one section; other sections set different amounts.

The difference that matters most is the auditor. Alvin Wang Graylin of the Asia Society Policy Institute put it plainly to Al Jazeera: “The companies drafted the principles, they hire the auditor, and the commitment is voluntary.” Under the bill, a new Under Secretary of Commerce licenses the auditor and can revoke that license, for example if models it verified turn out to fail.

What the critics and the fans said

Toby Walsh, who runs the UNSW AI Institute in Sydney, asked Al Jazeera, “What other trillion-dollar industry marks its own homework?” David Krueger of the University of Montreal, who wants a pause, called it “regulate without regulating” but gave it one point: it does ask for external evaluators, which “might result in a minute improvement over the status quo.”

On the other side, David Sacks, Trump’s former AI adviser, posted on X that the accord was “far better” than an international deal that would “probably never happen.” Amodei’s line at the event was that “we all need to work together to make sure that we can win and we can win safely.”

The name change got plenty of airtime too. Trump said the government would start calling AI “super intelligence,” or SI. Outlets split on whether he signed an order for that on Tuesday or only announced one, and as Time News pointed out, the accord itself never mentions the rename.

Where I come down

I don’t think the accord is worthless. Getting Zuckerberg and Huang to put their names under a line about models not hacking systems is more than existed a week ago, and Krueger’s “minute improvement” feels about right to me.

But I’d be careful about the word “constitution.” A constitution has courts behind it. This one has a board committee at each company, reporting to the same board that wants the next model shipped. If the companies really mean the sentence about codifying it someday, the text they’d need is already sitting in the House Energy and Commerce Committee. Watch whether any of the six say they support H.R. 9925 by name. I haven’t seen one do it yet.

Sources

New York Post: Trump’s ‘morally binding’ artificial intelligence pledge signed by tech leaders, read in full (full accord text), September 29, 2026
Forbes: Here’s The White House’s 308-Word AI ‘Accord’ In Full, September 29, 2026
Al Jazeera: Trump, tech bosses sign voluntary pact pledging ‘robust’ AI safeguards, September 29, 2026
Hindustan Times: Multiple AI CEOs met Trump. Only 6 signed his ‘super intelligence’ accord, September 30, 2026
Congress.gov: H.R. 9925, FRONTIER Act, text as introduced July 23, 2026
Time News: Trump Renames AI to Super Intelligence and Signs White House Accord, September 30, 2026