Consultations on a Bold and

Proper US-China-led AI Treaty

Closed-door sessions — one-to-one, small-group and remote — where delegates of key humaninst US AI power brokers help draft a strategic memo for President’s negotiations of a bold and proper an AI treaty with Xi.

Washington, DC and remote · Now to Doral, and beyond · Chatham House Rule

A standing working group of the delegates of key US AI policy powerbrokers, looking for common ground on an AI treaty-making process they could accept, and China plausibly could too.

The missing lane

Domestic campaigns are winning the argument for keeping humans in charge. None answers the question scientists, lab leaders and President Xi now ask aloud: what stops the race to superintelligence across borders? Washington has plenty of convenings on guardrails, standards and China. Few ask the question now on the table.

If Trump and Xi were to move or be moved beyond protocols, towards a bold AI treaty-making process, what process should the US be aiming for? And can a critical mass of US AI powerbrokers converge around a successful startegic plan to reliably steer the President to success — for the US and for Humanity?

Officials, lab leaders, members of Congress, national-security voices and humans-first figures hold very different predictions and red lines. They rarely compare them in private. That is the lane — Amodei's step two, "democratic coordination," widened from the labs to everyone who shapes the President's AI policy.

Shared goals, different predictions

As in the 1946 Acheson–Lilienthal consultations for the feasibility of a bold nuclear treaty, we assume that contributors largely share goals and diverge mainly on predictions. Where each power broker stands on a treaty follows largely from the odds they assign to seven deeply uncertain ones:

  • whether China wins the race to superintelligence;

  • whether superintelligence leads to human extinction, near-extinction or dystopia;

  • whether it would be unconscious — or conscious but unhappy;

  • whether it would discard the values its creators built in, after rewriting itself many times;

  • whether an aligned superintelligence would have to fight unaligned ones, at a disadvantage;

  • whether a proper treaty can prevent both runaway AI and AI-enabled authoritarianism;

  • whether lab leaders shape the future more by fostering a treaty than by building the first superintelligence.

The evidence on each is moving fast. History suggests that one key figure revising one prediction can reshape a whole consensus.

What happens

  • Who. Delegates: senior staff, advisors and envoys of executive-branch officials, frontier labs, Congress, and the political, media and religious leaders who shape US AI policy — for markets, Bessent; for security, Rubio, NSA Director Rudd and Cyber Director Cairncross; for industry, Amodei, Altman, Musk and Hassabis; for the President's base, voices such as Bannon, Carlson and Rogan — plus a few experts in verification, arms control and China.

  • How. Bilateral, trilateral and small multilateral sessions, in Washington and remote, under the Chatham House Rule. Each works on the latest draft: what's wrong, what's missing, what their principal could live with. Structured methods — Delphi pre-surveys, double crux, adversarial collaboration — so convergence is measured, not asserted.

  • Growth. Each contribution lowers the bar for the next. Once two or three offices have shaped the text, the fourth is joining something already respectable.

  • What. The ladder: first steps (an incident rulebook, misuse, pre-deployment testing), triggers to the next rung, the communication and verification infrastructure to build now, the safeguards that make scaling safe — and how safety commitments fit their economic impact and a fair share for citizens.

What it produces

No public statement, declaration or signatory list. Instead, three private outputs:

  • a better memo: every contribution is absorbed into Strategic Memo v3.0, and attributed by name only if the contributor agrees;

  • a coalition map: where US power brokers converge and diverge, shared privately with contributors — showing each how close the others already are;

  • a coalition ready to carry it. If enough of them converge, they can ask the President for a meeting together. If not, the memo stays on file with the administration, ready when he reaches for it.

A one-sided Track 1.5

In early 1946, the Acheson–Lilienthal consultations let Washington align internally, months before it brought a proposal to the UN for the international control of nuclear weapons. The consultations have similar aims but are driven by NGOs, involving key national stakeholders, and the aim is a proposal the other side can accept.

When

Now through the G20 at Doral (December 14–15), and beyond. By invitation.

Take part

Where to go next