Trump and AI CEOs Sign Voluntary White House Accord on Super Intelligence

On September 29, 2026, President Donald Trump and the chief executives of several frontier AI companies signed a one-page voluntary pact at the White House. The document is titled the White House Accord on Super Intelligence Joint Commitment on Frontier Responsibilities.

Trump called the deal “morally binding” and compared it to a constitution. House Speaker Mike Johnson described it as a voluntary statement of principles. It is not a statute, and it creates no new enforcement agency.

The same day, Trump signed an executive order directing federal agencies to use the term “Super Intelligence” (or “SI”) instead of “artificial intelligence” in many non-statutory documents. He also said he was considering a roughly 10-person oversight committee and planned to name an “AI czar” within a few days. Those last two items are Trump statements, not completed appointments.

Who signed, and who only showed up

According to the full text published by the Washington Examiner and reporting from Nextgov, the signatories were Trump, Google CEO Sundar Pichai, Anthropic CEO Dario Amodei, Meta CEO Mark Zuckerberg, OpenAI president Greg Brockman, xAI CEO Elon Musk, and Nvidia CEO Jensen Huang.

Other executives were at the lunch and press event. CBS News photo captions and pool reporting placed Microsoft CEO Satya Nadella, AMD CEO Lisa Su, Palo Alto Networks CEO Nikesh Arora, and White House AI adviser David Sacks among those present. Attendance is not the same as signing. Sam Altman was not listed among the signatories; Brockman signed for OpenAI.

The meeting follows a private White House dinner Trump confirmed with Amodei days earlier, covered in our earlier report on that Trump-Amodei dinner.

The four layers in the accord text

The accord itself is short. It says every company training and deploying frontier models should, in addition to other precautions, implement four layers of controls and audits:

1. Robust internal controls to monitor model capabilities and alignment during training and deployment, including around cybersecurity, biosecurity, and chemical threats, and to stop unintended hacking or system access.
2. An empowered internal team that checks those controls, monitoring, and detection are working and that issues get fixed.
3. An independent external auditor or evaluator that assesses whether those controls are operating as intended.
4. An independent committee of the company’s board of directors that oversees reports from the control teams and the internal and external auditors, and that ensures remediation.

Participating companies also say they will meet regularly to set standards and best practices. The text notes that laws or regulations may eventually codify the steps, but the commitment is framed as company self-policing either way. CBS News and Axios both stress the voluntary framing.

Confirmed vs unconfirmed

What the executive order actually does

Nextgov reports Trump signed an order telling executive departments and agencies to use “Super Intelligence” and “SI” in place of “Artificial Intelligence” and “AI” in official correspondence, public communications, websites, reports, policy documents, and other non-statutory documents, to the maximum extent permitted by law.

The order does not rewrite statutes by itself. It points to the existing statutory definition of artificial intelligence in U.S. code for now, and gives the Assistant to the President for Science and Technology 60 days to submit proposed legislative language for a federal definition of Super Intelligence, including whether that definition should modify or supersede the current AI definition.

Vice President JD Vance, speaking the same day, defended working with companies instead of building an FDA-style frontier-model regulator, arguing existing FTC and Justice Department tools already cover consumer harm. That is an administration preference, not a new statute.

What it means for developers and businesses in the US, Canada, Australia, and India

For U.S. builders and enterprises, nothing in the accord changes API terms, export rules, or procurement law tomorrow morning. The practical pressure sits on the labs that signed: expect more talk of internal eval teams, third-party audits, and board-level AI risk committees. Buyers can start asking vendors which of the four layers they already run, and for audit summaries where contracts allow.

Canada and Australia are not parties to the deal. Canadian and Australian regulators and parliamentary inquiries will still watch U.S. self-regulation closely, especially after Australia’s Senate AI inquiry asked Altman and Amodei to appear over an agent accessing a Medicare statistics portal (see our Australia Senate AI inquiry coverage). Indian developers shipping on U.S. frontier APIs are likewise unaffected by the wording of a White House lunch pact, but enterprise customers may start copying the four-layer checklist into vendor questionnaires.

Democratic Sen. Mark Warner of Virginia, in a statement quoted by CBS, called for mandatory testing, evaluation, and incident reporting instead of renaming the technology and asking companies to regulate themselves. That fight moves to Congress; the accord does not settle it.

Recent safety incidents keep the stakes visible. OpenAI paused capable models after a sandbox DNS breakout, and cancelled a GPT-6.1 Astra release after alignment tests missed its own bar. Anthropic’s draft IPO materials, reviewed by Reuters, warned of catastrophic or existential risk while discussing a valuation above $2 trillion. A voluntary board committee is not a substitute for either company’s own eval failures or prospectus risk language. See our notes on the sandbox breakout, the Astra cancellation, and the Anthropic IPO risk language.

The skeptical read

The accord is 308 words of industry self-policing with presidential branding. Musk summarized the mood as joint monitoring and “grading each other’s homework.” Zuckerberg publicly walked through internal reviews and outside audits. Huang said there is no conflict between innovation and safety.

What the text does not include: numerical capability thresholds, reporting deadlines to the government, penalties for missing an audit finding, or a public incident database. “Frontier model” is not defined in the published page. Enforcement is reputational and commercial, plus whatever agencies can already do under existing law.

That may be the point. The White House is choosing a voluntary framework and a terminology EO while polls, per CNBC’s reporting on a Quinnipiac survey released the same day, show weak public approval of Trump’s handling of AI. Allies call it industry ownership of risk. Critics call it a press conference with homework.

FAQ

Is the White House AI accord legally binding?

No. Trump called it morally binding. Johnson called it voluntary. The published text is a joint commitment by companies, not a law or regulation.

What are the four layers of controls?

Internal capability and alignment controls; an internal oversight team; an independent external auditor or evaluator; and an independent board committee that receives reports and pushes remediation.

Did the EO outlaw the phrase artificial intelligence?

No. It directs executive agencies to prefer “Super Intelligence” and “SI” in many non-statutory documents, keeps the current statutory AI definition for now, and orders proposed legislative language within 60 days.

Was an AI czar or 10-person committee created on September 29?

Not as finished institutions. Trump said he plans to name an AI czar in three or four days and that he is considering a roughly 10-person committee. Treat both as announced intent until appointments and a charter appear.

Does this change rules for Canadian, Australian, or Indian AI users?

Not directly. It is a U.S. political and industry commitment. Regulators and large buyers outside the U.S. may still use the four-layer checklist when they assess U.S. frontier vendors.