The scoreboard
29 September 2026
The text, titled “White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities”, runs just over 300 words. It starts from a principle, that “every company is responsible for developing its own technology safely and in a way that builds trust with customers and the public,” and asks each company training frontier models to stack four layers of checks. Semafor reported that Zuckerberg raised the idea with Speaker Johnson at the state dinner for Xi Jinping, worked out the principles with Jensen Huang, and circulated a draft before the lunch; on All-In David Sacks described himself as one of its architects.
Each one watches the one before it.
Read against the earlier pledges and the 2023 executive order.
The seating chart as All-In showed it. The six executives who signed are in blue. The other company leaders at the table, Nadella, Bezos, Lisa Su and Alex Karp among them, did not sign; none has given a reason. The bottom of the table was out of the frame, so a few seats are missing.
signed · company, did not sign · government · investor or adviser
“I think it’s morally binding … It’s almost like a constitution, in a way.”
“The technology has very real risks. And the mechanism, how we address those risks is still under discussion.”
“The president’s response? To rename it and tell the companies developing it to regulate themselves.”
The “Bretton Woods of super intelligence”: the first meeting to bring the key players of the new AI order to one table.
Third time
The Accord is the third time the frontier labs have made voluntary safety commitments in public. Each round asked for something new, and none came with a way to enforce it. Microsoft and Amazon signed the first two rounds but not this one.
| Round | Who signed | The main promise | Outside check | Made public | Enforced by |
|---|---|---|---|---|---|
| White House commitments21 Jul 2023 · Biden | 7 companies (Amazon, Anthropic, Google, Inflection, Meta, Microsoft, OpenAI); 15 by September | Test models for safety and security before release; share risk information; watermark AI content | Yes: testing “carried out in part by independent experts” | Yes: report capabilities and limitations | No one |
| Frontier AI Safety Commitments21 May 2024 · Seoul summit | 16 companies, including Zhipu AI from China and xAI | Publish a safety framework with thresholds of risk at which a model will not be deployed | Not required | Yes: the frameworks are published | No one |
| White House Accord on Super Intelligence29 Sep 2026 · Trump | 6 companies (Google, Anthropic, Meta, OpenAI, xAI, NVIDIA); Microsoft and Amazon present but not signing | Four layers of controls and audits, up to the board | Yes: an auditor the company chooses | Not required | No one: “morally binding” |
The word
The bar used to be set by the best humans in every field. The executive order brought it down to any AI system. The marker on each card is my reading of how high that definition sets the bar.
ASI, transformative AI, powerful AI: the same idea has gone by many names, some for a mind past the best humans, some for the human-level rung below it, and some defined by what it would do to the world rather than how smart it is.
| Name | Who, when | Where it sets the bar | What it stresses | Source |
|---|
Each event’s sources are listed under it. For 2026, many come from news reports and law-firm summaries rather than the primary text. The Accord and EO 14434 are quoted from their published texts (American Presidency Project; Federal Register), and EO 14409 follows the Congressional Research Service’s summary. Calling CAISI the testing body under EO 14409 rests only on secondary blogs, so this page doesn’t. Hassabis’s essay was published on X; it is cited here through summaries. The seating chart is transcribed from the All-In episode; the off-frame guests are from Axios’s list.