198 is the number that needs to be remembered. It is the count of parties to the Montreal Protocol on Substances that Deplete the Ozone Layer: every recognized state on Earth, including such unlikely joiners as North Korea, South Sudan, and the Cook Islands. No other UN treaty has reached universal ratification.

The Montreal Protocol is the only UN treaty ratified by every country on Earth, it solved the problem it was written to solve, and its architecture was designed for exactly the situation ASI governance now faces.

The substance behind the number is harder to dismiss. In 1987 governments committed to phase out a profitable, widely used industrial technology (chlorofluorocarbons, then used in nearly every refrigerator, aerosol can, and air conditioner on the planet) before the worst damage had occurred. The treaty entered into force in 1989. The ozone hole has been shrinking for a decade. The clean recovery date is expected around mid-century, roughly a human lifetime after the wrong substances stopped being released.

For anyone arguing a binding AI treaty is naive, Montreal is the inconvenient counterexample. The international system can restrain a dangerous technology under uncertainty, across rival economies, fast enough to matter. The question worth asking is not whether such a treaty is possible (the answer is on file) but which features of Montreal made it work, and which of them an ASI governance regime can borrow without inventing law from scratch.

What the treaty actually did

The Protocol on Substances that Deplete the Ozone Layer set binding, scheduled reductions in the production and consumption of a defined list of chemicals. It did not ban everything at once. It set a trajectory: specific percentage cuts by specific dates for the worst substances, then ratcheted that trajectory repeatedly as science firmed up. CFCs were the opening target. HCFCs came next. The 2016 Kigali Amendment added HFCs, the substitutes that turned out to be potent greenhouse gases on their own. Each tightening bound every party without reopening the treaty for fresh national ratification.

The obligations were differentiated on purpose. Developing countries were given a grace period and financial help to comply, in recognition that they had not caused most of the problem and could not absorb the transition costs alone. Any AI treaty will face the same equity problem between the few states with frontier labs and the majority without.

Three design features worth copying

1

Start with a framework, tighten later

The 1985 Vienna Convention for the Protection of the Ozone Layer created the institutional shell: obligations to cooperate, monitor, and report, before anyone had agreed on numbers. The binding cuts arrived two years later in Montreal and were tightened at least nine times afterward. ASI governance can begin with a thin framework convention and add substance as capabilities and evidence develop. That sequencing buys time without buying inaction.

2

Adjust without re-ratifying

The treaty's masterstroke is the adjustment mechanism. Parties can accelerate phase-out schedules by a qualified-majority decision that binds every member, without reopening the treaty for national ratification. That made the treaty responsive to new science without making it hostage to a decade-long ratification cycle. For a technology moving as fast as AI, where capabilities can change in months, a static treaty would be obsolete on arrival. The ability to update binding obligations without re-ratification is a survival requirement.

3

Make non-participation expensive

Montreal banned trade in controlled substances with non-parties. That single provision flipped the calculus for any state considering opting out: staying outside meant losing market access, so joining became the rational choice. Compute and semiconductor supply chains offer a structurally similar chokepoint for AI, with a handful of firms and a small number of jurisdictions controlling the inputs that frontier development needs.

The funding mechanism that bought consensus

In 1990 the parties created the Multilateral Fund, financed by developed countries, to pay the agreed incremental costs of compliance for developing ones. It was no foreign aid. It was the price of universal participation. China and India became parties once the fund made compliance affordable. By 1996 the fund had disbursed hundreds of millions of dollars and Phase I obligations had been met or exceeded by every developing-country party.

The lesson for AI is direct. A treaty that asks the Global South to forgo a transformative technology or to accept intrusive monitoring of its infrastructure will not be joined as an act of generosity. It will be joined when the bargain makes participation cheaper than refusal: capacity-building, technology access in defined fields, and compensation in exchange for constraint. The Multilateral Fund template is right there to copy.

Where the analogy strains

No precedent is perfect, and the fair statement of skepticism runs like this. Ozone depletion had a clear, measurable metric (atmospheric chlorine concentration above specific thresholds) and a small number of substitutable chemicals produced by a handful of firms. Frontier AI capability is harder to define, harder to measure, and developed by actors with far stronger strategic incentives to defect. CFC substitutes were commercially attractive; the replacements sold themselves. There is no equivalent commercial product waiting on the other side of a superintelligence prohibition.

These differences make ASI governance harder, not impossible. They say where the work has to go. Define measurable thresholds analogous to atmospheric chlorine. Build the verification tools that ozone monitoring already had (a working global sensor network in the case of ozone, with publishable data; the AI analogue does not yet exist at the same fidelity). Construct an incentive structure through compute supply chains and market access that makes participation the rational choice. Montreal did not succeed because the problem was easy. It succeeded because the treaty was designed intelligently around a hard problem. That became the standard.

What travels, and what has to be rebuilt

The framework-then-bite sequencing, the qualified-majority adjustment mechanism, the supply-chain chokepoint, the Multilateral Fund structure: each is portable. What is not portable is the underlying clarity of the science. Montreal worked in part because atmospheric chemists could say, with replicable numbers, exactly when a substance was doing too much damage. That kind of shared epistemic ground does not yet exist for AI capabilities.

Building it is a precondition, not an optional extra. Until the field can say, with replicable numbers, that a model has crossed a defined capability threshold, the treaty's threshold language will be contested in every dispute. The Montreal Protocol did not cause the ozone measurement infrastructure. It inherited it from three decades of atmospheric chemistry. An AI treaty will need to commission its equivalent first, or import the lesson that some treaties repay the science investment many times over.

The strongest pushback

The fairest objection is that ozone chemistry was clearer than AI risk, so Montreal cannot guide us. The science clarity helped. The portable pieces are still design: adjust schedules, hit supply chains, fund holdouts into compliance. AI needs harder verification. It does not need to invent governance from zero.