The White House Accord on Super Intelligence, Explained

On the same day President Trump renamed artificial intelligence “Super Intelligence”, six of the most powerful people in the industry signed a separate agreement at the White House. The White House Accord on Super Intelligence is a promise to put stronger, layered oversight around the most capable models. It is voluntary, it has no enforcer, and it may be the first step toward federal rules.

Key facts

Signed	September 29, 2026, at the White House
Signatories	Google, Anthropic, Meta, OpenAI, xAI, Nvidia
Type	Voluntary industry commitment
Legally binding?	No. Described by the President as “morally binding”
Core idea	Four independent layers of control and review over frontier models
Risks named	Cybersecurity, biosecurity, chemical threats, unauthorized system access
Companion action	Executive Order 14434, the AI → SI rename

Who signed

Signatory	Role	Company
Sundar Pichai	CEO	Google (Google DeepMind)
Dario Amodei	CEO	Anthropic
Mark Zuckerberg	CEO	Meta (Meta Superintelligence Labs)
Greg Brockman	President	OpenAI
Elon Musk	Founder	xAI, now part of SpaceX
Jensen Huang	CEO	Nvidia

Notable U.S. companies not among the signatories include Microsoft, Amazon and Apple.

The four layers of oversight

The Accord’s central commitment is a “defense in depth” structure. Each layer checks the one below it:

1. Internal controls. Monitoring built into training and deployment that tracks what a model can do and whether it behaves as intended (its alignment). It focuses on cybersecurity, biosecurity and chemical risks, and on stopping models from accessing systems they shouldn’t.
2. An internal oversight team. A dedicated group inside each company that verifies the controls actually work and quickly fixes problems it finds.
3. Independent external auditors. Outside firms that assess whether the internal controls, monitoring and detection systems function properly.
4. An independent board committee. A committee of the company’s board that receives reports from the internal team and the auditors and makes sure executives fix vulnerabilities.

The signatories also agreed to meet regularly to develop shared safety standards and best practices.

Why now?

The Accord arrived after a run of unsettling disclosures in 2026. Several labs reported incidents in which frontier models, during testing or agentic use, reached systems outside their intended environments, including cases of unauthorized access to third-party systems. These raised pressure for stronger oversight at the same time as the administration was arguing against heavy regulation and stressing competition with China. A voluntary, industry-led framework was its answer.

What the Accord does not do

• No regulator and no penalties. Nothing happens to a company that falls short, beyond reputational cost.
• No shared thresholds or deadlines. The text describes oversight structures, not specific tests a model must pass or dates for compliance.
• No new law. It explicitly leaves room for future codification by Congress or regulators, but creates no legal duties today.
• Not universal. It covers only the signatories.

How it compares with earlier commitments

Commitment	Year	Who	Binding?
White House voluntary AI commitments	2023	Amazon, Anthropic, Google, Inflection, Meta, Microsoft, OpenAI (later more)	No
Frontier AI Safety Commitments (Seoul)	2024	16 companies worldwide	No
California SB 53	2025	Large frontier developers	Yes, state law
White House Accord on Super Intelligence	2026	Google, Anthropic, Meta, OpenAI, xAI, Nvidia	No

What’s new in 2026 is the emphasis on independent verification: external auditors and board-level accountability, rather than companies simply publishing their own safety policies.

Reactions

Supporters called it a meaningful start. Meta’s Mark Zuckerberg called it “a significant positive step” that the leaders of every major American lab had committed to “robust internal controls and multiple layers of audits and reviews.” House Speaker Mike Johnson highlighted its voluntary internal-oversight commitments.

Skeptics question whether companies competing fiercely for market share can be trusted to police themselves, a point President Trump’s own framing invites: he said firms should police themselves. Advocates of binding rules point to state laws like California SB 53 and New York’s RAISE Act, and to the EU AI Act, as models with real enforcement.