TLDR
Six major AI company leaders signed a voluntary self-policing accord at a White House lunch on 29 September, committing to four layers of internal controls and audits. The deal sits in direct tension with calls Dario Amodei and Sam Altman made six days earlier at the UN Security Council for binding global standards.
A handshake with no regulator in the room
The arithmetic is straightforward. Six companies have four control layers and zero enforcement. President Trump announced a voluntary accord at a White House lunch on 29 September. The leaders of six major AI firms signed the agreement. Each committed to self-police their frontier models through internal and external controls.[1] Greg Brockman of OpenAI and Dario Amodei of Anthropic were among the signatories.[1]
Mark Zuckerberg described the structure as having multiple layers of auditing and controls. It starts with internal risk review and external auditors. Boards of directors then independently review the reports from auditors.[1] A fourth requirement rounds out the framework. Signatories will hold semiannual meetings to assess progress on frontier AI responsibilities.
What the signatories said at the UN six days earlier
The timing places the two appearances just six days apart. Amodei and Altman appeared before the UN Security Council on 23 September. They warned that unchecked frontier AI poses a real and imminent threat. No country can contain it alone. They called for common global standards.[2] Turing Award laureate Yoshua Bengio put the same argument in sharper terms. "The dangers of AI beyond human control represent an unprecedented threat for countries that none can contain alone and that does not respect the borders we defend," Bengio said.[2]
Great discussion today at the UN Security Council. Frontier AI needs shared guardrails to keep everyone safe.
2026-09-23 · View on XThe same executives signed an accord six days after those remarks at the Security Council. The agreement lacked any international coordination mechanism and external government oversight.[2] The accord itself does not document what changed between the Security Council chamber and the White House dining room.
Trump's position and the gap it leaves
Trump set the frame one day before the Security Council session. He addressed the 81st UN General Assembly on 22 September. He rejected any attempt to construct a "globalist scheme of control" for artificial intelligence. He referred to it officially as "Super Intelligence".[3] "There is no global government, and while I am President, there will be no global taxes," Trump said.[3]
Proud to join my peers and @POTUS in pledging strong safety standards for AI development.
2026-09-29 · View on XThat position left a governance gap the voluntary accord does not fill. The four control layers are assessed internally and reported to each company's own board. They are subject to no mandatory government review or penalty for non-compliance.[1] The accord's semiannual meetings are among the signatories themselves. No regulator or independent body sits at the table.
The practical result is a governance model where the world's most powerful AI developers audit themselves. They report to their own boards and meet twice a year. The next of those semiannual meetings is not yet scheduled publicly.
KEY TAKEAWAYS
SOURCES & CITATIONS
FREQUENTLY ASKED QUESTIONS
What are the four layers of control in the White House AI accord?
Which AI company leaders signed the accord?
What happens if a company breaches the accord?
Alex Mercer writes about technology, energy and infrastructure. He likes the physical end of the story: the plants, the grids and the machines that everything else depends on.
