A Treaty
Like the Bomb.
Whether an international AI ceasefire is possible, what it would take, and what happens if we get it wrong — the regulation question nobody wants to answer honestly.
On 16 July 1945, the first nuclear device was detonated at Trinity in New Mexico. Within weeks, atomic bombs had destroyed Hiroshima and Nagasaki. The Soviet Union tested its first device in 1949. The Nuclear Non-Proliferation Treaty was opened for signature in 1968 — more than two decades after Trinity, and only after the technology had already rearranged global power.
The NPT was not a success because it stopped all proliferation — it didn't. India, Pakistan, Israel and North Korea either never signed or subsequently developed weapons. It was a partial success because it slowed and constrained something that, without it, might have spread much further and faster. It established a norm. It created inspection infrastructure. It made the possession of nuclear weapons by non-signatory states politically costly in a way it would not otherwise have been.
Can something similar be built for AI? The honest answer — the answer that the experts closest to the problem will give you when they are not optimising for reassurance — is: possibly. Partially. With enormous difficulty. And probably not before something goes seriously wrong.
What Exists Today
The regulatory landscape for AI in 2026 is not nothing. It is, however, fragmented along lines that reflect geopolitical competition more than shared technical understanding.
The gap between what exists and what a treaty would require is significant. These frameworks regulate rights, deployment, providers and national markets. None creates a shared system for verifying frontier training runs, assessing dangerous capabilities, or making restraint rational for actors with strong incentives to defect.
What Would Actually Work
The most serious proposals for meaningful AI governance converge on a common mechanism: compute governance. The argument is that compute — the hardware required to train frontier AI models — is the most controllable chokepoint in the entire development chain.
Unlike software, which can be copied instantly and distributed globally, the hardware used for the largest training runs is physical, expensive and supplied through a concentrated global chain. That makes compute more visible than algorithms or data. It does not make it simple to control: hardware can be repurposed, systems can become more efficient and policy can concentrate power as easily as it constrains risk.
An agency would not magically ban development. States would still hold the power to act. Its value would be verification: a shared record of what is being trained, what has been tested, and where risks have been found. Without a credible way to observe compliance, a treaty is a promise made in the dark.
The strongest governance proposals are not simple demands to “pause AI.” They ask for the technical and institutional capacity to slow, inspect or stop a development path if evidence justifies it. The distinction matters. A brake is not a destination. It is an acknowledgement that a system without a brake leaves its passengers with only hope.
Why the Nuclear Analogy Breaks Down
The nuclear analogy is instructive but imperfect. Understanding where it breaks down is as important as understanding where it holds.
Verification has physical anchors. Nuclear weapons require detectable infrastructure. The largest AI training runs require data centres, power, networks and advanced hardware.
Existential stakes justify international cooperation. When both sides genuinely fear the worst outcome, treaties become rational for all parties.
Actor count. Nuclear weapons required states. Frontier AI can increasingly be developed by well-funded private companies, and eventually by smaller actors. The NPT model controls states. It cannot easily control non-state actors.
Verification depth. You can count warheads. You cannot easily verify that a training run has been paused, or that a model has not been secretly fine-tuned to acquire new capabilities.
The Three Scenarios
Governance frameworks exist, but no shared verification regime does. From here, three broad futures are plausible.
The Compute Lever — Most Promising, Most Contested
The most technically tractable governance proposal is compute caps — limiting, by regulation, the amount of AI compute that can be produced or used for frontier training runs annually. Compute is physical. It leaves traces. It can be exported only through known channels. It is the closest thing to a verifiable chokepoint that exists in the AI development chain.
Compute caps alone are insufficient. Algorithmic progress means a capability once requiring a large budget may later be reached with less. Compute governance therefore needs to work with model evaluations, reporting obligations, cybersecurity controls and safety cases above defined thresholds. The lever is promising precisely because it is not the whole machine.
The US–China dimension is the hardest part. A durable regime needs participation from the states and companies capable of supporting frontier-scale development. Export controls can shape incentives and buy time, but they do not automatically create cooperation. They can also accelerate substitution, fragmentation and domestic supply chains.
What a Real Brake Requires
A serious international framework would need more than a shared statement of concern. It would need three capabilities that do not yet exist at the required scale:
2. Governance tools for coordination. International frameworks that would allow a deliberate, coordinated pacing of frontier development to be enacted if the monitoring indicated it was necessary — which requires diplomatic infrastructure, verification mechanisms, and enforcement tools that also do not currently exist.
3. Build them before the emergency. Once a coordinated slowdown is urgently needed, the diplomatic and technical infrastructure will take too long to invent. The brake must exist before the vehicle reaches the speed at which it is needed.
This is more modest than a treaty, but also more honest. The preparatory work — common measurements, credible evaluations, reporting channels and institutions — is what makes a treaty conceivable when political conditions allow.
The Question Nobody Answers
There is a question at the centre of all these proposals that the public debate consistently avoids: what happens if the governance fails?
The nuclear case had a partial answer: deterrence. The knowledge that any use of nuclear weapons would trigger retaliation stabilised the system — imperfectly, terrifyingly, but sufficiently. The doctrine of Mutually Assured Destruction was horrifying as a strategy. It was also effective as a stabiliser for nearly eighty years.
There is no clean equivalent doctrine for AI. A system that causes harm through misuse, bad incentives, error or loss of control may not present a single actor that can be deterred. That is why technical safety work, institutional accountability and the ability to intervene before deployment all matter. They are not substitutes for one another.
The final report in a series on AI should perhaps end with a conclusion. This one ends with a question: when a system becomes too capable to govern through ordinary market incentives and voluntary promises, who has the authority — and the evidence — to stop it?
The nuclear analogy gives one answer: treaties bought time, and the time was used to develop better strategies for managing the risk. Perhaps the same will be true for AI. Perhaps the governance frameworks being proposed now — partial, imperfect, contested — will buy enough time for the alignment research to catch up.
Perhaps. The word appears repeatedly in the work of the people who understand this best. It is the most honest word available. And in a field where the downside of being wrong is unbounded, it is not a comfortable place to rest.
What we do next is still, for a brief and uncertain time, a choice.