Dossier · Global Coordination

Demand an AI Safety Treaty.

Negotiate and ratify a binding international treaty on frontier artificial intelligence before any system approaching general intelligence is deployed at civilization scale.

No private company should hold unilateral power over a technology that can reshape every nation on Earth.

Summary

No nation should be permitted to unilaterally release a system that could reshape every economy, every election, and every labor market on Earth. A treaty is how civilizations historically handle weapons of last resort, nuclear, chemical, biological. Frontier AI belongs in that category until proven otherwise. Demand that your government open formal negotiations on enforceable thresholds, shared inspection regimes, and red lines that cannot be crossed for short-term commercial advantage.

What is being asked

A binding international treaty on frontier artificial intelligence. Not a press release. Not a voluntary pledge. A treaty, with verification, inspection rights, and meaningful penalties for violation.

The model is not new. The world built such instruments for nuclear weapons, for chemical weapons, for biological weapons, for the atmosphere itself. Each time, the same objection was raised: it is too early, too complex, too political. Each time, the absence of agreement cost more than the agreement would have.

Frontier AI is now in the same category. Systems being deployed today can already influence elections, generate functional malware, impersonate human beings at scale, and accelerate the design of biological and chemical agents. The companies building them say so themselves, in their own published safety reports.

Why a treaty, and not self-regulation

Self-regulation has been tried. It has produced safety teams that were disbanded, charters that were quietly amended, and commitments that expired the moment a competitor moved faster. The pattern is consistent across every major laboratory and every major jurisdiction.

This is not a moral failing of any one company. It is the predictable behavior of any actor inside a race they did not start and cannot unilaterally exit. A treaty is what allows the race to slow without any single participant being punished for slowing first.

What history tells us

Civilization has, before, succeeded in placing limits on weapons and technologies whose unchecked development would have ended it. It has done so imperfectly, slowly, and over the objections of those whose power depended on the absence of limits. It has, nonetheless, done so.

The instruments exist. The legal craft exists. The diplomatic machinery exists. What is missing is the public demand.

The demand, specifically
  1. 01.Negotiation of a binding multilateral treaty on frontier AI systems within the next legislative session.
  2. 02.Mandatory pre-deployment registration of any model trained above a defined compute threshold.
  3. 03.An independent international body with authority to inspect, audit, and require remediation.
  4. 04.Clear, enforceable penalties for non-compliance, including restrictions on cross-border deployment.
  5. 05.Public reporting of all critical safety incidents within a fixed disclosure window.
If nothing is done
  • A small number of private firms will continue to set the global rules of an irreversible technology.
  • States that prefer no rules will set the floor for everyone else.
  • Each new generation of systems will be deployed before the previous one has been understood.
  • By the time consensus becomes politically possible, the technology in question will already be ambient and uncontainable.
Objections, answered
Objection

"A treaty would let other countries get ahead."

Answer

The reverse is true. The absence of a treaty guarantees that the country willing to take the most risk sets the pace. Treaties are not about slowing one side. They are about agreeing on a floor neither side will undercut.

Objection

"AI is too fast-moving to regulate."

Answer

Every transformative technology has been called this. The argument is always made by those who benefit from delay. Speed of development is an argument for governance, not against it.

Objection

"Companies will simply move offshore."

Answer

Frontier AI requires extraordinary capital, specialized hardware, and predictable jurisdictions. There is no realistic offshore alternative for an actor seeking commercial scale. This is a talking point, not a fact.

The species has done this before. It is being asked to do it again, on a tighter clock, with stakes that compound daily.

Pillar

Human Future Framework · Global Coordination

Ways to act