The White House Accord on Super Intelligence: who signed it and what it says

Who signed

Signatory	Organisation	Our page
Donald J. Trump	President of the United States	The rename timeline
Sundar Pichai	Google, chief executive	Google DeepMind and AGI
Dario Amodei	Anthropic, chief executive	Anthropic and advanced AI
Mark Zuckerberg	Meta, chief executive	Meta Superintelligence Labs
Greg Brockman	OpenAI, president	OpenAI and AGI
Elon Musk	xAI	xAI
Jensen Huang	Nvidia, president and chief executive	None yet

OpenAI signed through its president rather than its chief executive. Safe Superintelligence Inc., the company with “superintelligence” in its name, did not sign. All six signatory companies are American; the accord has no signatories from Europe, the UK or China.

What it asks companies to do

The accord opens with a statement of principle: “we believe every company is responsible for developing its own technology safely and in a way that builds trust with customers and the public.” It then sets out four layers of controls that each company should apply to its most capable, or “frontier”, models.

1. Internal controls and monitoring. Controls to monitor the capabilities of its models, including in sensitive areas such as cybersecurity, biology and chemistry, and the alignment of its models during training and deployment, together with controls “to ensure that its models do not hack or access technical systems in unintended ways”.
2. An internal team to check them. A team inside the company whose job is to check that those controls and that monitoring work, and to get problems fixed.
3. An independent external auditor or evaluator. An outside body that assesses whether the controls are effective.
4. An independent board committee. A committee of the company’s board, independent of management, that oversees the reports of the internal team and the external auditor and makes sure problems are put right.

The signatories also say they will meet regularly to develop standards and best practices. The accord says that over time it may make sense to codify such steps in law or regulation, but that the signatories regard them as important whether or not that happens.

What it does not do

• It is not legally binding. There are no penalties for not meeting it. The President described it as “morally binding”, with what he called a “tremendous self-policing aspect”.
• It gives government no role. No agency checks compliance, receives the audits or can act on them.
• It requires nothing to be published. Companies are not asked to publish their audit results, their board committee’s reports, or even who their external auditor is.
• It does not say how. It does not specify which tests to run, what results would be unacceptable, or who counts as independent. Each company decides.
• It does not stop anything. Nothing in it requires a company to delay or cancel a release. As our page on frontier AI safety frameworks explains, no outside body currently has that power over these companies’ models.
• It is not about superintelligence in the research sense. The title uses the US administration’s term for AI in general (what the difference is). The controls apply to today’s most capable models.

When we checked on 2 October 2026, the text had not been published on whitehouse.gov. The President posted it on Truth Social; the Washington Examiner published it in full, and the American Presidency Project has archived it.

What the accord does and doesn’t specify

The accord asks each signatory to work with “an independent external auditor or evaluator” and to designate “an independent committee of the board of directors” to oversee the work. It doesn’t name any auditors or committees, and it sets no deadlines for publishing results or standards. It says the companies “will meet regularly to establish standards and best practices”.

Not to be confused with the Super Intelligence Force

The Super Intelligence Force, announced on 4 October 2026, is a separate government initiative. It is not an oversight body set up by the accord.

What people have said

This page records what has been reported; it does not take a view on whether the accord is adequate.

• At the signing, Elon Musk said: “We agreed to a number of things that include joint monitoring, board special committees, just generally grading each other’s homework.”
• Vice President JD Vance told the industry leaders that “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.” Defense One reported that he was cool on proposals for a new regulator modelled on the Food and Drug Administration or the Federal Aviation Administration.
• Al Jazeera noted that several signatories have also called for mandatory rules. OpenAI’s Chris Lehane has said industry standards should “complement – not replace – mandatory federal safeguards”, and Dario Amodei has called for “more serious and binding regulation of AI”.
• SiliconANGLE reported that the President also said he would form a ten-person AI safety committee and name an “AI czar”.

How it compares

Measure	Where	Status	What it requires
White House voluntary commitments	United States	Voluntary, July 2023	Seven companies, including Google, Meta, OpenAI and Anthropic, committed to security testing before release, information sharing and other safeguards
White House Accord on Super Intelligence	United States	Voluntary, September 2026	Four layers of controls: internal monitoring, an internal team to check it, an independent external auditor or evaluator, and an independent board committee
Transparency in Frontier AI Act (SB 53)	California	Law since 1 January 2026	Large frontier developers must publish a safety framework, comply with it and report critical safety incidents to the state, with civil penalties
EU AI Act, general-purpose AI rules	European Union	Law since August 2025	Providers of the most capable general-purpose models must assess and reduce systemic risks and report serious incidents
No statutory requirement	United Kingdom	Voluntary testing	The AI Security Institute tests models only with companies’ cooperation

Like the 2023 commitments, the accord is voluntary; what it adds is oversight by an independent committee of each company’s board. Unlike the California and EU rules, it carries no legal duty and nothing has to be published. Our frontier safety frameworks page compares these in more detail.

What it means for the UK

Directly, nothing. The accord is between the US President and US companies and creates no obligations in Britain. Indirectly, it matters because these six companies build or power most of the frontier models used here. The accord’s “independent external auditor or evaluator” could in principle be a body like the UK’s AI Security Institute, but it names none, and the institute’s access to models remains voluntary. Our page on how AI is regulated in the UK sets out the rules that do apply.

What we are watching

• Whether any signatory says how it is meeting the accord: who its external auditor is, and whether it has set up the board committee.
• Whether the signatories’ regular meetings produce published standards.
• Whether other developers sign, including Safe Superintelligence Inc., Microsoft and Amazon.
• Whether the ten-person AI safety committee and the AI czar are appointed, and with what remit.
• Whether an official text appears on whitehouse.gov.
• Whether Congress moves to put any of these steps into law.

This page is reviewed monthly, or sooner when something changes. For the separate executive order signed the same day, see Executive Order 14434 explained.