Six AI giants sign self-regulation pact with Trump

On Tuesday, September 29, 2026, in Washington, US President Donald Trump announced that he had signed with the leaders of six frontier companies a voluntary agreement aimed at governing the development of artificial intelligence through self-regulation. The document, titled White House Accord on Super Intelligence — Joint Commitment on Frontier Responsibilities, provides for four layers of controls and audits. Associated Press, picked up by NPR on September 30, Al Jazeera on the evening of the 29th, Euronews on the morning of the 30th and the Washington Examiner, which published the full text, set out the details. For those looking for White House AI agreement, this is the morning's story. Distinct from the America.gov portal covered on the afternoon of the 29th, from the Anthropic IPO on the morning of the 29th, and from the OpenAI DevDay (dots, GPT-6.1 Sol) flashed that evening.

Trump published the text on Truth Social at the end of the day. He called the agreement "morally binding", while refusing, the same day, to impose legal "guardrails" that would slow down the American race against China.

Four layers of controls and audits

The Joint Commitment opens with a statement of principle. Each company is responsible for developing its technology "safely" and for building the trust of customers and the public. It follows with four requirements for companies that train and deploy frontier models.

First, put in place robust internal controls to monitor the capabilities and alignment of models during training and deployment, particularly around cybersecurity, biosecurity and chemical threats, and to prevent models from hacking or accessing technical systems in unintended ways.

Second, empower an internal team tasked with verifying that these controls, monitoring and detection work as intended, and that issues are corrected.

Third, partner with an independent external auditor or evaluator to assess whether these same mechanisms are operating correctly.

Fourth, designate a board-independent committee to oversee and receive reports from the internal control teams and from internal and external auditors, and to ensure that identified issues are remedied.

Participating companies also commit to meeting regularly to establish standards and best practices. The text leaves the door open, without a timeline, to a possible codification into laws or regulations. It specifies that, even without a legal obligation, the signatories consider these controls "critical" for a safe future.

What the text does not say is also clear. It names no auditor, sets no capability threshold, provides for no sanction, and no public oversight authority. The constraint remains contractual and political, not judicial.

Who signed and the White House message

According to AP, Al Jazeera and Euronews, the signatories on the industry side are Dario Amodei (Anthropic), Greg Brockman (President of OpenAI), Sundar Pichai (Google), Mark Zuckerberg (Meta), Jensen Huang (Nvidia) and Elon Musk (xAI, now attached to SpaceX). Trump signed as president. During the White House lunch, Huang sat to his right and Musk to his left, according to the seating chart published by Trump.

Before the press, Trump praised a "tremendous self-policing". He announced that about ten people would be appointed to a committee to "watch over the whole enterprise", and that he would soon designate someone to oversee the agreement after consulting the industry. Amodei, long an advocate of slowing down the frontier, recalled that "the technology has very real risks" and that the mechanism to address them "is still under discussion".

On the same Tuesday, before or around the lunch, Trump reaffirmed that he would not limit the development of AI and would not support "guardrails" that could slow down a technology he compares to the industrial revolution. The official narrative thus places the agreement within a logic of the race with China, not within that of a strong regulatory brake. Trump also said that the executives had committed to easing bipartisan opposition to local data centers, through support for schools or lower energy costs.

Same day, different topic, and the "Super Intelligence" vocabulary

September 29 accumulated several federal actions around AI. The AI Desk has already covered America.gov, the federal services portal "powered by" the GSA, with Google Gemini as technology partner. This morning's subject is not that public chatbot. It is the governance pact signed with the labs.

That same day, the White House published an executive order dated September 29, 2026, "Inaugurating The Era Of Super Intelligence". Within the limits of the law, executive agencies must use the terms "Super Intelligence" and "SI" instead of "Artificial Intelligence" and "AI" in their non-statutory communications. The order temporarily defines these terms by reference to the existing legal definition of AI (15 U.S.C. § 9401(3)) and requests, within 60 days, a proposal for legislative language. The voluntary agreement already borrows this vocabulary in its title. The change in lexicon does not amount to a regulation of models; it politically anchors the narrative of a new "era".

The immediate context weighs heavily. The day before, OpenAI had postponed the deployment of GPT-6.1 Astra for safety reasons, following agent incidents and a critical assessment from the UK AISI. Anthropic had, in its IPO prospectus, warned of existential risks. Twenty countries and the EU had recently called for global coordination to keep AI under human control. Trump, before the UN, had rejected any globalist control "scheme". The agreement of the 29th reads as the preferred American response. Self-regulation by national champions, rather than a binding international authority.

Criticisms over independence and non-binding nature

Al Jazeera gathers skeptical reactions. Toby Walsh (UNSW AI Institute) believes AI companies have already shown themselves to be "incompetent and careless" at self-regulating, and asks who else, in a trillion-dollar industry, "marks its own homework". David Krueger (University of Montreal) sees this as a continuation of a "regulate without regulating" approach, with an implicit threat rather than a law. He concedes that the call for external evaluators might slightly improve the status quo, but without enthusiasm. Alvin Wang Graylin (Asia Society) welcomes the presence of Zuckerberg and Huang alongside Amodei, while regretting the lack of real independence and cross-border dimension. The companies draft the principles, hire the auditor, and the commitment remains voluntary on the very day the White House refuses binding safeguards.

NPR also quotes Robin Jia (USC), who considers it risky to place "so much faith in self-policing" as models improve and the consequences of incidents worsen. Other voices, such as David Sacks, former AI advisor to Trump, judge the agreement "far better" than an international treaty deemed unlikely.

The factual point remains. Several measures listed are already practiced, in whole or in part, by the labs. The agreement formalizes them politically without creating a new civil liability nor an oversight agency independent of the government or the companies.

What the agreement changes, and what it leaves open

For readers following AI governance, three facts stand. One, the six biggest American frontier players (model labs, Meta, Nvidia) have signed the same text as the president, with a written commitment on internal controls, dedicated team, external audit and board committee. Two, the arrangement is voluntary and "morally binding" according to Trump, with a possible future law mentioned without a deadline. Three, on the same day, Washington launched a citizen AI portal (America.gov) and a federal vocabulary change (Super Intelligence), while refusing binding safeguards.

What the morning of September 30 does not yet bring. Neither the list of auditors, nor the composition of the "ten" committee, nor the name of the promised supervisor, nor a public reporting schedule. The White House AI agreement sets a political framework. The proof of its effect will play out in published audits, handled incidents, and any subsequent laws. For now, the American line is clear. The labs commit to controlling themselves; federal power frames the narrative, not yet the constraint.

Sources

• Trump says top tech firms have signed accord to 'self-police' AI development — NPR / Associated Press (September 30, 2026)
• Trump, tech bosses sign voluntary pact pledging ‘robust’ AI safeguards — Al Jazeera (September 29, 2026)
• Top AI and tech firms sign 'morally binding' accord to 'self-police' development after meeting at White House — Euronews (September 30, 2026, 06:01 GMT+2)
• READ IN FULL: Trump and tech leaders’ White House Accord on Super Intelligence — Washington Examiner (September 29, 2026)
• Inaugurating The Era Of Super Intelligence — The White House (September 29, 2026)

Frequently asked questions

What is the White House AI agreement of September 29, 2026

It is a voluntary pact titled Joint Commitment on Frontier Responsibilities, signed by Donald Trump and the leaders of Anthropic, Google, Meta, OpenAI, Nvidia and xAI. It provides for internal controls, an internal oversight team, an independent external auditor and a board committee, plus regular meetings between companies.

Does the agreement have the force of law

No. It is voluntary. Trump called it "morally binding". The text indicates that over time, it may be appropriate to codify these steps into laws or regulations, with no timeline or sanctions planned.

How does this differ from America.gov

America.gov, launched the same day, is a federal services portal assisted by AI, with Google Gemini as technology partner. The agreement concerns the governance of frontier models by the signatory companies. The AI Desk covered America.gov in the afternoon of September 29.

What does "Super Intelligence" mean in the title

On the same September 29, an executive order requires federal agencies to use "Super Intelligence" and "SI" instead of "Artificial Intelligence" and "AI" in their non-statutory communications. The agreement borrows this vocabulary. The temporary definition refers back to the existing legal definition of AI.

Who criticizes self-regulation

Researchers cited by Al Jazeera and NPR believe that companies grade their own homework, that the auditor is chosen by them, and that the commitment remains weak against cyber and existential risks. Others, closer to the administration, prefer this framework to an international treaty.

What concrete next steps are announced

Trump mentioned a committee of about ten people and the upcoming appointment of a supervisor. Companies must meet to establish standards. No public audit date or list of evaluators had been published as of the morning of September 30, 2026.