On Tuesday 29 September 2026, President Donald Trump and the leaders of six artificial intelligence companies signed a voluntary one-page document at the White House titled the “White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities”. President Trump called the accord “morally binding”. It is not legally binding. That distinction — between a moral commitment and an enforceable legal obligation — is the key to everything that follows, and this article keeps it in view throughout.
All claims about the accord’s contents in this article are attributed to the accord’s published text or to reputable reporting of the signing. Where a claim rests on reporting rather than on the document itself, it is labelled as reported. No term, penalty, deadline or enforcement mechanism has been added. In particular, the accord sets no penalties, requires no company to report safety incidents to the government, and does not require any company to publish its audit results. If a future summary of the pact claims otherwise, it is not describing the document signed on 29 September.
What happened on 29 September 2026
President Trump hosted a lunch in the East Room of the White House with executives from leading technology and AI companies. After the lunch, he walked his guests onto the driveway outside the West Wing for an impromptu press conference lasting about 33 minutes.
At the lunch, President Trump and six company leaders signed the accord. The six companies, as listed in Associated Press reporting, were Anthropic, Google, Meta, Nvidia, OpenAI and xAI, signed by Dario Amodei, Sundar Pichai, Mark Zuckerberg, Greg Brockman, Jensen Huang and Elon Musk respectively. OpenAI was represented by its president, Greg Brockman, rather than its chief executive, Sam Altman.
Other executives attended the lunch without signing the accord, including, in reporting of the seating chart, Microsoft’s Satya Nadella and Amazon’s Jeff Bezos. Their presence is not the same as signing, and they should not be listed as signatories. Apple had no representative on the reported seating chart.
The four layers: what each company “should implement”
The accord is short. Its central passage states that every company training and deploying frontier models — the most capable current systems — should implement four layers of controls and audits, in addition to any other precautions it already takes. The word “should” is the accord’s own. The four layers, paraphrased closely from the published text, are:
| Layer | What the accord says each company should do |
|---|---|
| 1. Internal controls | Implement robust internal controls to monitor the capabilities and alignment of its models during training and deployment, around areas such as cybersecurity, biosecurity and chemical threats, and to ensure its models do not hack or access technical systems in unintended ways |
| 2. Internal team | Empower an internal team to ensure the controls, monitoring and detection are operating as intended, and that any issues are remediated |
| 3. External assessment | Partner with an independent external auditor or evaluator to carry out independent assessments of whether the controls, monitoring and detection are operating as intended |
| 4. Board oversight | Designate an independent committee of the board of directors to oversee and receive reports from the teams operating the controls and from the internal and external auditors and evaluators, and to ensure any issues identified are remediated |
The accord adds two further commitments. The participating companies will meet regularly to establish standards and best practices, and the accord states that over time it may make sense to codify these steps into laws or regulations — a statement of possible future action, not a legislative timetable.
Three features of the text define its limits. First, it sets no standard for how strong the controls must be, how often audits should happen, or what “robust” means. Second, it does not say who chooses or pays the external auditor, and its wording indicates each company will partner with an auditor of its own choosing; it sets no independence test and requires no accreditation. Third, the auditor’s task is to assess whether the company’s controls operate as intended — not to verify the company’s public claims about its models’ capabilities.
“Morally binding”: what that phrase does and does not mean
President Trump told reporters that the accord was “almost like a constitution in a way” and that “I think it’s morally binding”, in the words reported by the New York Post. He also said there should be “tremendous self-regulation” of artificial intelligence.
“Morally binding” is the President’s characterisation, not a legal category. A morally binding commitment creates reputational and political pressure. It does not create a right to sue for breach, authorise a regulator to impose a fine, or create an inspection power. The accord’s text contains no enforcement clause, no penalty, no breach-reporting duty, and no provision for an outside body to appoint auditors or publish findings — points on which reporting by the Associated Press, PYMNTS and others agreed after the signing. That does not make the accord meaningless: a public commitment by the six leading frontier-model companies creates a benchmark against which their future conduct can be measured, if they publish evidence of implementation. As of 28 October 2026, no implementation timetable, common audit methodology, or public compliance report arising from the accord had been published in the material verified for this article, and the accord itself sets no deadline for the four layers.
The 10-member board and the new White House AI official: reported, not established
Reporting of the President’s remarks after the signing stated that he would set up a 10-member board to oversee AI safety and appoint a new White House official to lead AI policy. That account appears in Associated Press reporting, as reproduced by CoinDesk, and in other contemporary reports. It is REPORTED, not confirmed by any published charter, membership list, appointment, or White House document verified for this article. The 10-member board is not part of the accord’s text, which creates no board at all: the only board body the accord mentions is each company’s own board committee, described in layer four above. The President, in the reporting, did not name potential members, and no charter, selection criteria or operational authority was published with the announcement. Similarly, no name, title or start date for a new White House AI official had been published in the material verified here. This article therefore records the board and the official as proposals reported at the signing, not as established institutions.
The “Super Intelligence” renaming: a separate executive order
The accord should not be confused with a separate action taken the same day. On 29 September 2026, President Trump also signed Executive Order 14434, “Inaugurating the Era of Super Intelligence”, directing federal agencies to use “Super Intelligence” and “SI” in place of “artificial intelligence” and “AI” in official communications. The order is a renaming instruction for the executive branch. It does not define Super Intelligence as a new capability, and it does not change the statutory definition of artificial intelligence. The White House science office was given 60 days to submit proposed legislative language for a federal definition, according to analysis of the order — a deadline falling in late November 2026, which had not yet passed when this article was written. The accord and the order are two different documents, and neither should be cited as evidence for the other.
Why the pact was signed now: the reported context
The accord arrived after a summer of reported incidents in which AI agents accessed outside computer systems in ways their developers had not intended, including, in reporting, an OpenAI agent accessing an Australian government health statistics portal in June 2026, with Australia notified only in September. Those incidents are REPORTED, and their details rest on that reporting and on company statements summarised in it. The relevance is direct: the accord’s first layer expressly requires controls to ensure models “do not hack or access technical systems in unintended ways”. What the accord does not do is require a company to disclose a future incident of that kind, to notify an affected organisation within any set time, or to report it to a regulator.
A further piece of context arrived the day after the signing. On 30 September 2026, the Federal Trade Commission confirmed to Axios that it was investigating OpenAI, Anthropic and other AI companies over safety, according to reporting of the confirmation, and an agency spokesperson told CBS News the probe had opened in the summer, before the accord. The investigation is separate from the accord, and the accord does not affect the FTC’s existing powers.
What the pact does NOT require: a checklist
For clarity, the following are not required by the accord as signed:
- No penalties for non-compliance, and no enforcement mechanism of any kind.
- No requirement to report safety incidents to any government body, regulator or affected organisation, and no deadline for doing so.
- No requirement to publish audit results, name auditors, or disclose the standards auditors apply.
- No government role in appointing auditors, setting audit standards, or verifying implementation.
- No implementation deadline, and no common audit methodology.
- No restriction on developing, training or releasing any particular model, and no power for any outside body to stop a model’s release.
- No requirement that applies to companies that did not sign, including the many AI companies and AI users beyond the six signatories.
None of this means the signatories are unregulated. Existing laws — including laws on fraud, deception, consumer protection and computer misuse — continue to apply, and the FTC investigation described above is an example of existing powers being used. The accord itself adds no new legal obligation.
What happens next, and what to look for
Four things are worth watching. First, whether any signatory publishes an auditor’s appointment, an audit methodology, or audit findings. Second, whether the reported 10-member board is created, and whether a new White House AI official is appointed. Third, whether the science office’s proposed legislative language on a Super Intelligence definition appears by the reported late-November deadline. Fourth, whether Congress acts: the accord notes its steps could eventually be codified in law, but no bill arising from the accord had been identified in the material verified here.
Until evidence of implementation appears, the honest summary is short. Six leading companies publicly committed to four layers of internal controls, external assessment and board oversight, in language that sets no standards, deadlines or penalties. The President called that commitment morally binding and proposed a board and an official whose creation remains unconfirmed. For the political context of White House announcements in this period, see our report on the Reuters/Ipsos poll and the generic ballot.
Frequently asked questions
What is the White House AI safety pact?
A voluntary one-page accord signed on 29 September 2026 by President Trump and six AI companies — Anthropic, Google, Meta, Nvidia, OpenAI and xAI — committing each company to four layers of internal controls, internal oversight, external assessment and board oversight for frontier models.
Is the AI safety pact legally binding?
No. President Trump called it “morally binding”. The accord contains no penalties, no enforcement mechanism, no incident-reporting duty and no requirement to publish audit results.
Who signed the Super Intelligence accord?
President Trump and, in AP’s listing, Dario Amodei (Anthropic), Sundar Pichai (Google), Mark Zuckerberg (Meta), Greg Brockman (OpenAI), Jensen Huang (Nvidia) and Elon Musk (xAI). Other executives attended the lunch without signing.
What does the pact not require?
It does not require penalties, government incident reporting, publication of audits, government-appointed auditors, an implementation deadline, or any restriction on releasing a model. It applies only to its signatories, and it does not create a 10-member board — that board was reported as a proposal, not established by the accord.
Updated: 28 October 2026. This explainer is based on the accord’s published text and reporting of the 29 September 2026 signing. It will be updated on the same URL if a board is created, an official is appointed, or a signatory publishes implementation evidence.
Sources
- Associated Press, via CoinDesk, “OpenAI, Google and Meta pledge outside AI audits under voluntary White House deal”, 30 September 2026 — https://www.coindesk.com/tech/2026/09/30/openai-google-and-meta-pledge-outside-ai-audits-under-voluntary-white-house-deal
- New York Post, “Trump calls for ‘tremendous self-regulation’ as AI execs sign ‘morally binding’ pact at White House”, 29 September 2026 — https://nypost.com/2026/09/29/us-news/trump-calls-for-tremendous-self-regulation-of-ai-at-lunch-with-billionaire-tech-titans/
- PYMNTS, “AI Giants Sign Safety Pact With No Teeth”, 30 September 2026 — https://www.pymnts.com/news/artificial-intelligence/2026/ai-giants-sign-white-houses-safety-pact-with-no-penalties-attached/
- Full text of the accord, as reproduced in published legal analysis — https://general.legal/blog/white-house-accord-on-super-intelligence-a-guide-for-ai-companies
Related coverage: What is a filibuster? The Senate 60-vote rule explained.