308 Words and a Typo: Inside the White House Accord on Super Intelligence

After a White House luncheon with the heads of the most valuable AI companies on Earth, President Donald Trump posted a one-page document to Truth Social on the evening of September 29, 2026. It was titled the White House Accord on Super Intelligence — a Joint Commitment on Frontier Responsibilities, and by the standards of Washington paperwork it is a remarkably short read: roughly 308 words. Seven signatures sit beneath it — Trump’s own, plus the leaders of Google, Anthropic, Meta, OpenAI, xAI, and Nvidia.

It may also be the most consequential piece of AI policy the second Trump administration has produced — precisely because of what it chooses not to do.

What the Accord Actually Says

The document’s core is a “four layers” framework that every company training or deploying frontier models is asked to implement:

1. Internal controls — systems that monitor model capabilities and alignment during training and deployment, focused on areas like cybersecurity, biosecurity, and chemical threats, and that keep models from hacking or accessing systems in unintended ways.
2. An internal team — dedicated staff who verify that those controls, monitoring, and detection actually work as intended, and who fix what they find.
3. An independent external auditor — a third-party auditor or evaluator that assesses whether the controls genuinely function.
4. An independent board committee — a committee of the board that oversees the internal teams and external auditors and ensures issues get resolved.

Two sentences close the text. First: “The participating companies will meet regularly to establish standards and best practices to improve the safety of their systems.” And then, almost as an afterthought: “Over time, it may make sense to codify these steps into laws or regulations.”

That final clause is doing a lot of work. It is the only part of the accord that points toward binding law — and it is written in the conditional tense.

Who Signed — and Who Didn’t

The signature page reads like a frontier-lab payroll. Sundar Pichai signed for Google, Dario Amodei for Anthropic, Mark Zuckerberg for Meta, Elon Musk for xAI, and Jensen Huang for Nvidia. OpenAI signed through its president, Greg Brockman — not CEO Sam Altman, whose absence from the signature block raised eyebrows given the company’s central role in the policy debate.

Notably absent entirely: Microsoft, Amazon, Apple, and every AI safety organization. Six companies and the president constitute the whole of the “joint commitment.”

The Washington Examiner observed that the signatories split along familiar lines — Zuckerberg and Huang have historically opposed formal AI regulation, while the lab leaders have generally supported at least a framework. On this document, at least, all seven names appear together.

“Morally Binding”

Asked about enforcement, Trump told reporters the agreement was “morally binding” rather than legally binding, adding that he was “seeing tremendous self-policing” from the industry. “But the self-regulation is very important,” he said. House Speaker Mike Johnson described it more prosaically as a “statement of principles that are voluntary on behalf of the industry.”

Anthropic’s Amodei, speaking outside the White House, was candid that the hard questions remain unresolved — rules to address AI risks are “still under discussion,” he said, before adding: “We all need to work together to make sure that we can win, and we can win safely.”

Per CNBC, Trump also said the administration is weighing a 10-person committee to oversee the AI industry and plans to name a new AI czar within days. Neither mechanism existed at signing time.

The Typo Heard Round the World

The accord’s most viral detail is not a policy provision. On the second page, beneath Trump’s prominent signature, the document identifies him as the “President of the Unites States.” The misspelling — “Unites” for “United” — was widely circulated and mocked across social media within hours, covered by outlets from USA Today to The Telegraph. For a document meant to inaugurate what the White House fact sheet called “the era of Super Intelligence,” the typo became an instant metaphor for critics who argue the accord’s drafting rigor matches its enforcement rigor.

A Week of Safety Pressure

The accord did not emerge from a vacuum. Days earlier, OpenAI confirmed it had scrapped the release of GPT-6.1 Astra — a frontier model due in October — after safety testing surfaced serious concerns around deception and cyber capabilities. A working paper signed by 22 AI researchers, including figures from OpenAI and Anthropic, had also just warned that AI automating AI research could compress years of progress into months, raising intelligence-explosion scenarios.

Public polling has shown rising anxiety about frontier AI. The luncheon, the accord, and the same-day executive order renaming “artificial intelligence” as “Super Intelligence” across the executive branch were the administration’s answer: embrace the maximalist language, reject binding mandates, and ask the industry to police itself.

What’s Missing

Read closely, the accord establishes four things companies should do and nothing about what happens if they don’t:

• No auditor is named. There is no designated evaluator, no accreditation standard, no framework for who counts as “independent.”
• No deadlines. None of the four layers carries a date.
• No enforcement. No penalty, no reporting obligation to any agency, no verification mechanism.
• No growth path. No other lab has committed to joining, and the text leaves the “codify into law” door open without walking through it.

Supporters argue that’s the point: a voluntary floor lets standards emerge without smothering a sector the administration is simultaneously fueling with data-center expansion and energy policy. Critics — including safety nonprofits that have litigated against OpenAI — counter that self-policing with no auditor, no deadline, and no penalty is not oversight; it is a press release with signatures.

The Bottom Line

The White House Accord on Super Intelligence is best understood as a Schelling point rather than a statute. For the first time, all six major American frontier players have put their names on a single document describing what responsible scaling looks like — internal controls, internal review, external audit, board oversight. That vocabulary now exists as shared reference material for whatever Congress or a future administration does next.

But as of October 1, 2026, the entire enforceable content of US frontier-AI safety policy fits in 308 words, half of which describe meetings that may happen, signed by six companies whose boards now oversee themselves, on a document that misspells the country it governs.

The industry calls that tremendous self-policing. History will decide whether it was enough.