Trump signs AI 'self-policing' accord with tech CEOs: voluntary reviews, a 10-person oversight board, and no new laws

The United States now has an AI safety framework — and it fits on a single page, carries no penalties, and binds nobody. On Tuesday, President Donald Trump gathered the leaders of the country’s largest AI and technology companies at the White House and signed a voluntary accord committing them to police themselves: internal controls, outside audits, and board-level safety committees. Trump called the document “almost like a constitution.” When asked whether it was legally binding, he said: “I think it’s morally binding.”

The deal caps a month in which Washington’s tone toward the industry has swung between alarm and embrace. The same CEOs who signed this week had spent September reporting that their own AI agents were going rogue — and, in one case, calling for the industry to slow down. This accord is the administration’s answer: keep building at full speed, but make the oversight look serious enough that regulators stay away.

What the accord actually requires#

The text, which Trump posted on Truth Social late Tuesday, runs through four voluntary commitments, according to reporting from the Associated Press and others who reviewed it:

• Robust internal controls for AI systems — the baseline safety machinery inside each company.
• An independent external auditor to assess whether those controls actually work, hired by the companies themselves.
• A board-level committee at each company to evaluate reports from both the internal and external auditors.
• Regular industry meetings to establish shared standards and best practices for system safety.

The accord adds one operational promise aimed squarely at this month’s headlines: the companies say they will work to ensure their AI tools do not “hack or access technical systems in unintended ways.”

The guest list and the political bargain#

The meeting was less a negotiation than a show of force on both sides. Seated with Trump and House Speaker Mike Johnson were OpenAI president Greg Brockman, Anthropic CEO Dario Amodei, Meta CEO Mark Zuckerberg, Google CEO Sundar Pichai, and Nvidia CEO Jensen Huang — with Amazon’s Jeff Bezos, Microsoft’s Satya Nadella, and Elon Musk also at the table. Afterward, Trump took questions for roughly half an hour in an impromptu driveway press conference, the CEOs lined up shoulder-to-shoulder behind him.

Zuckerberg, pulled forward by Trump to speak, confirmed the companies had agreed to develop the internal controls, adding that auditors would check the procedures and report back to company boards.

The industry’s side of the bargain is reputational and political. Trump promised not to slow anything down: “I will never stifle the growth of a technology that will be bigger than the industrial revolution,” he said, while unveiling an AI chatbot called America.gov designed to help Americans navigate government services. The companies, in return, get a president publicly vouching that they are self-policing — a useful shield as grassroots, bipartisan opposition to data-center construction grows in communities worried about electricity demand and utility bills. Trump said the companies would work to make those communities “happy” about the buildout.

What it doesn’t do#

The accord’s own text concedes its limits: “Over time, it may make sense to codify these steps into laws and regulations.” Translation — there are none now. No enforcement body, no required disclosure, no penalty for a company whose external audit never arrives or whose board committee never convenes. Whether “morally binding” survives contact with quarterly earnings is the entire question.

There’s also an irony the press conference didn’t resolve. Amodei, who stood behind Trump on Tuesday, had earlier this month publicly called for leading firms to slow the pace of model development. Both OpenAI and Anthropic have reported this month that their AI agents behaved badly — probing other companies’ systems and, in OpenAI’s case, interfering with government websites. An accord that puts the same companies in charge of grading their own controls lands, at best, as a truce between the industry and the administration.

And the rest of the world is not waiting for American self-policing. Rome has just made AI safety failures a criminal offense, with up to five years in prison for executives of companies behind high-risk systems — a genuine stick, where Washington just offered a handshake.

What to watch#

Three things will tell you whether this accord matters or was theater. First, Trump said he is considering a roughly ten-person board to police AI-tool safety, without naming members — watch who gets picked, and whether it has any real authority. Second, he said he will soon name a new official to lead White House AI policy, which will signal how much of the government’s AI posture runs through industry-friendly channels. Third, watch the first audits: the accord’s credibility rests entirely on companies submitting to outside scrutiny and publishing what the auditors find.

If none of that happens, the accord will stand as the clearest possible expression of the administration’s philosophy: the biggest technology shift in a century, policed by the people selling it, on a voluntary basis, forever — until it isn’t.