On 10 June 1963, at American University in Washington, John F. Kennedy gave a speech that did not get drafted by aides ahead of time. He spoke for thirty-five minutes about the nuclear threat, and toward the end he named the problem the United States and the Soviet Union had created by building arsenals without stopping the spread of the technology. "I would regard this as the greatest threat to mankind," he said, then offered the warning: by 1970 there could be fifteen nuclear powers; by the end of the decade, twenty or twenty-five. He named specific countries. He proposed the partial test ban that would be signed in Moscow three months later.
The NPT did not stop proliferation entirely, but it slowed a predicted wave of twenty-plus nuclear states to a trickle. It did so through a bargain that ASI governance will have to reinvent.
The predicted world did not arrive. Nine states have nuclear weapons today, not twenty-five. The reason is a treaty Kennedy did not live to see. The Treaty on the Non-Proliferation of Nuclear Weapons was opened for signature on 1 July 1968, roughly five years after the speech and three years after his death. It now has 191 states parties: the most widely joined arms control agreement in history. Its mechanism matters well beyond nuclear weapons, because frontier AI presents the same structural problem. A small number of actors hold a dangerous capability, governance depends on stopping its spread, and the constrained majority must be persuaded that restraint is in its interest.
The three pillars
The NPT rests on a bargain among three commitments, each pillar leaning on the others. The nuclear-weapon states agreed not to transfer weapons and to pursue disarmament in good faith. The non-nuclear states agreed not to acquire weapons and to accept IAEA safeguards on their nuclear programs. All parties affirmed the right to peaceful nuclear energy, with cooperation and technology-sharing for those who stayed non-nuclear.
Non-proliferation
The non-nuclear states pledge not to build or acquire weapons and to accept international verification of their nuclear programs through the International Atomic Energy Agency. This is the constraint the treaty exists to impose. Two hundred forty-seven IAEA safeguards agreements are now in force, covering most of the world's civil nuclear facilities.
Disarmament
Under Article VI, the nuclear-weapon states pledge to negotiate in good faith toward elimination of their arsenals. This is the reciprocal concession that makes the constraint politically survivable. The non-nuclear states are not being told to stay behind forever. The pledge has been honored slowly, and the slowness is the source of much of the resentment the treaty faces today.
Peaceful use
Every party retains access to civilian nuclear technology, with the nuclear-weapon states helping non-nuclear states develop it. This is the benefit that makes joining attractive rather than merely restrictive. Roughly a quarter-century of IAEA technical cooperation later, dozens of countries operate nuclear power plants under safeguards without reversing their non-weapons status.
Why the bargain structure matters for AI
The lesson that travels is structural, not technical. A non-proliferation regime cannot be only a set of restrictions on the weak imposed by the strong. It has to offer the constrained majority something they value (here: peaceful nuclear technology and a credible eventual-disarmament pledge) or they will not join, and a treaty the relevant states have not joined governs nothing. The two-sided structure is what produced near-universal NPT membership. The unilateral structure is what produced the four states that remain outside: Israel, India, Pakistan, and (after 2003) North Korea's withdrawal.
ASI governance faces an identical divide. A frontier-AI treaty will be negotiated among a few states with frontier labs and a large majority without. If the agreement looks like the AI haves locking in permanent advantage while everyone else accepts monitoring of their compute and infrastructure, the participation rate will be low, and the treaty's authority will collapse on contact with its first real test. The NPT's answer (shared access to beneficial technologies plus a credible commitment by the leaders to accept reciprocal constraints) is the template for making an AI bargain acceptable to the states being asked to sign.
The NPT's failures, read as design warnings
The treaty is not an unqualified success, and its weak points are exactly where ASI governance must do better. The disarmament promise in Article VI has been honored slowly and grudgingly. The total number of nuclear weapons has declined since the treaty's signing, from a Cold War peak above 60,000 to roughly 12,500 today, but every actual reduction has come through side agreements (START I, START II, New START) rather than through Article VI's primary diplomacy. Non-nuclear states have long felt the bargain was one-sided. The most recent review conference, in 2022, ended without a consensus outcome over Ukrainian and Iranian grievances. The resentment is no peripheral annoyance. It is the structural risk the treaty runs every year it goes without credible progress on disarmament.
Three states never joined (India, Pakistan, Israel), and a fourth (North Korea) withdrew in 2003 and tested its first device later that decade. The pattern is consistent. A determined state can stay out indefinitely if the cost of doing so is lower than the cost of accepting the constraints. And the peaceful-use pillar spreads exactly the dual-use technology that brings a state to the nuclear threshold. Iran has shown how compromises within the NPT framework can be used to advance the program the framework was meant to prevent.
Each failure carries a design warning. The disarmament equivalent in an AI treaty (whatever the leading states offer in return for others' restraint) has to be credible and delivered, or it will corrode in the same way. The treaty needs to make exit costly rather than consequence-free. And it has to grapple, as the NPT never fully did, with the dual-use ambiguity. The same GPU clusters that train weather models and drug-discovery systems are the ones that can train the next generation of frontier models. The meter must read two numbers at once.
The political architecture, distinct from the technical
The NPT gives ASI governance a concrete blueprint for one half of the problem: the political architecture. The technical architecture (which capabilities to measure, which thresholds to define, how to verify them) is its own design exercise and not transferable in any useful form. What transfers is the shape of the agreement.
The most-joined arms control treaty in history is a bargain. The AI version will have to be one too.The work is the harder work of building the bargain that the law eventually has to express, not new international law. The lesson travels across every recurring feature: identify the capability frontier at which any competent actor could pose catastrophic global risk, then negotiate a halt that is robust against defection. Head-on instead of papering over it.
And learn from where the NPT frayed, with leaders' concessions real and visible, exits expensive through market and supply-chain chokepoints, and the dual-use problem confronted head-on.
Kennedy's June 1963 speech did not produce the treaty. It produced a willingness among enough governments, on both sides, to negotiate one. That work happens before any document is signed, and it is the part the Foundation's organizing is designed to do. A treaty the relevant states have not joined governs nothing. Getting them to join is the political question, and it is the one whose answer makes everything after it possible.
The strongest pushback
The fairest objection is that the NPT froze inequality and still leaked, so copying it freezes a bad bargain. Learn from the fray: visible leader restraint, expensive exits, dual-use honesty. Do not worship the text. Steal the grand-bargain structure that made non-proliferation possible at all.