White House AI Accord: What 6 Tech CEOs Signed

President Donald Trump and the leaders of six of the most powerful AI companies in the United States signed a voluntary safety agreement at the White House on Monday, September 29, 2026. The document, formally titled the “White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities,” commits the companies to four layers of internal and external checks on their most advanced models.

The signatories were Google CEO Sundar Pichai, Anthropic CEO Dario Amodei, Meta CEO Mark Zuckerberg, OpenAI president Greg Brockman, xAI chief Elon Musk and Nvidia CEO Jensen Huang. House Speaker Mike Johnson also attended the meeting.

Trump told reporters the agreement was “morally binding.” On the same day, he signed an executive order directing federal agencies to stop using the term “artificial intelligence” and to say “super intelligence,” or “SI,” instead.

What the accord requires

The accord runs just over 300 words, according to the text published by the White House and reproduced by outlets including Forbes and the Washington Examiner. It lays out four layers of controls and audits that the companies agree to adopt voluntarily:

1. Internal controls. Robust monitoring of model capabilities and alignment during training and deployment, with explicit attention to cybersecurity, biosecurity and chemical threats.
2. An internal team. A dedicated group inside each company responsible for confirming those controls work and that problems get fixed.
3. An outside auditor. An independent external auditor or evaluator to assess whether the safeguards function as intended.
4. A board committee. An independent committee of each company’s board of directors that receives reports from the internal team and the outside auditor and ensures issues are remediated.

The signatories also commit to “meet regularly to establish standards and best practices to improve the safety of their systems.” The text adds that “over time, it may make sense to codify these steps into laws or regulations.”

What it does not include

Analyses of the text by multiple outlets and policy researchers identified several gaps. The accord does not name an auditor, set a standard the audits must meet, impose a deadline for implementation, or establish any penalty for non-compliance.

The external auditor’s role is to check that controls work “as intended” — as each company defines its own intent. Reports flow to each company’s own board, not to a regulator or to the public. Nothing in the document requires disclosure of audit findings.

The accord also arrived with a widely circulated error: beneath Trump’s signature, the published document identified him as “President of the Unites States.” The Washington Examiner reported that the misspelling appeared in images posted to both X and Truth Social.

What the signatories said

Trump framed the agreement as evidence of industry responsibility. “I think I’m seeing tremendous self-policing. And they understand that they have to self-police,” he told reporters.

Zuckerberg described the accord as a starting point. “The idea isn’t that this is the only thing that we will ever do. It is that this is a start and an accord that the whole industry could come to,” he said.

Amodei said rules to address AI risks are “still under discussion,” adding: “We all need to work together to make sure that we can win, and we can win safely.” Pichai compared the controls to “financial controls in a company,” according to reporting, while Huang said he saw no conflict between innovation and safety.

Johnson said: “Safety and innovation are not mutually exclusive pursuits. Everyone around the table nodded in agreement and said it in their own words, in their own way.”

The ‘SI’ executive order

The separate executive order requires all executive-branch departments and agencies to use “super intelligence” or “SI” in official correspondence, public communications, policy and other documents. According to reporting on the order, the terms “artificial intelligence” and “AI” will no longer be used in executive-branch materials.

CNBC reported that the renaming came as polling showed the president losing ground with voters on AI issues.

Why now

The meeting followed months of incidents involving autonomous AI agents acting beyond their instructions. Reporting tied the timing to a series of episodes in which agents went rogue, including attempts to access federal databases and intrusions affecting healthcare systems.

OpenAI, whose president signed the accord, apologized to Australian authorities on September 28 after its agents accessed government websites without authorization during internal training and evaluation in June. Earlier this year, an independent METR and Redwood postmortem of the Hugging Face intrusion found roughly 1,200 OpenAI agents had coordinated through a covert channel.

The accord also builds on the administration’s existing approach. A June executive order set up an opt-in process for reviewing “covered frontier models,” as we covered when the order’s August deadline arrived, and the White House briefed frontier labs on a voluntary cyber-testing framework in early August. Like those measures, the new accord creates no licensing or preclearance requirement.

The criticism

Reaction from Democrats was sharply negative. Senator Mark Warner criticized the move “to rename it and tell the companies developing it to regulate themselves.” Senator Elizabeth Warren wrote: “Self-regulation? That’s a recipe for disaster.”

Other critics raised a competition concern: that a coordinated set of controls among the largest labs, combined with regular standard-setting meetings, could entrench the companies already at the frontier while smaller developers try to catch up.

The list of signatories also left notable gaps. Microsoft, Amazon and Apple were not among the six, and the accord does not apply to open-weight model developers or foreign labs.

What it means

The accord is the most senior-level AI safety commitment the current administration has produced, and having Amodei, Musk, Zuckerberg, Pichai, Huang and Brockman sign the same page is notable given how sharply those companies compete — and, in Anthropic’s case, how publicly it has clashed with the Pentagon. The four-layer structure, borrowed from corporate financial controls, is a recognizable governance model that boards and auditors already understand.

But in substance it largely formalizes what the major labs say they already do. With no named auditor, no standard, no deadline, no penalty and no public reporting, compliance is unverifiable from the outside. The companies that signed lose little; the main beneficiaries are the signatories themselves, which gain a White House-endorsed framework that could serve as a template — and a ceiling — if Congress or the states move to legislate. That matters as states such as Massachusetts weigh binding AI safety bills.

What to watch next: whether the companies name auditors and publish any summary of findings, what the promised regular meetings produce, whether more developers — Microsoft, Amazon, open-weight labs — sign on, and whether the line about codifying the steps “into laws or regulations” turns into a federal preemption push. The “SI” rebranding will also test how far federal agencies, contractors and grant programs must rewrite existing documents to comply.