A framework convention is a treaty that deliberately separates the structure of a regime from its detailed obligations. Rather than trying to settle everything at once, states first agree a framework — shared principles, objectives, definitions, and the institutions and procedures for cooperation — and then negotiate the specific, binding commitments later, as separate protocols. It is the international-law equivalent of agreeing to build the house and its foundations before finalising every room. For a problem that is urgent, contested, and fast-changing, this staged design has decisive advantages.
How the model works
Agree the framework
States conclude a convention setting out the shared goal, guiding principles, key definitions, and — crucially — the institutions and decision-making procedures for developing the regime further. This can be reached before agreement on the hardest specifics, because it does not require them.
Add binding protocols
Under that framework, states negotiate separate protocols containing the specific, binding obligations — targets, thresholds, verification rules. Each protocol is its own agreement, adopted when the political and technical conditions are ripe.
Tighten over time
As understanding and trust grow, protocols can be strengthened and new ones added, letting the regime evolve without renegotiating the whole treaty. The framework is the stable trunk; the protocols are branches that grow and adapt.
The precedents that prove it
This is not a theoretical design; it is how several of the most important international regimes were built. The ozone regime began with the 1985 Vienna Convention — a framework with almost no hard obligations — followed two years later by the Montreal Protocol that did the real work. The climate regime started with the 1992 UN Framework Convention on Climate Change, under which the Kyoto Protocol and Paris Agreement were later negotiated. Global tobacco control runs on the WHO Framework Convention. In each case, agreeing the structure first made binding commitments reachable that a single comprehensive treaty could not have delivered up front.
Why it fits AI so well
- It breaks the deadlock on details. States can agree the framework — that frontier AI risk is a shared concern requiring cooperation and institutions — without first settling the contested thresholds and verification methods that would otherwise block any agreement.
- It builds the institutions early. The framework can establish the scientific body, the monitoring functions, and the decision-making procedures now, so they exist and mature before the binding protocols arrive.
- It adapts to a moving target. New protocols and amendments let the regime keep pace with a technology that changes faster than any fixed treaty could, without reopening the whole agreement.
- It creates momentum. A framework is a concrete, achievable first win that establishes the process and the expectation of protocols to follow, rather than an all-or-nothing gamble on a comprehensive text.
The honest caveat
The framework model has a real failure mode, and pretending otherwise would be naive: the framework can become the ceiling rather than the floor. A convention full of principles and institutions but empty of binding constraint is easy to agree precisely because it demands little, and there is no guarantee the hard protocols ever follow. Climate governance is a cautionary example — a strong framework, decades of negotiation, and binding commitments that have consistently fallen short of what the problem requires. A framework is a vehicle, not a destination; it can carry a regime toward real constraint or idle indefinitely.
The framework-convention model is how you start governing a problem before you have solved every hard question about it. Its promise is a fast, achievable beginning. Its danger is mistaking the beginning for the end.
Naoto Nakada, Founder · Nakada Foundation to Save Humanity
The right first step, with eyes open
For AI, a framework convention is probably the most realistic route to binding governance: it sidesteps the deadlock over details, builds the institutions early, and adapts to a fast-moving technology, all while creating an achievable first agreement that momentum can build on. It is the structure that makes the incremental path coherent and the comprehensive destination reachable. But adopting it demands clarity about the failure mode. The point of a framework is the protocols that follow, and the measure of success is whether the binding constraints actually arrive — and arrive in time. Used with that discipline, the framework convention is the best-fitting tool available for AI governance. Used as an excuse to look busy while avoiding hard commitments, it becomes another way to run out the clock.
Common questions.
A treaty that deliberately separates the structure of a regime from its detailed obligations. States first agree a framework — shared principles, objectives, definitions, and the institutions and procedures for cooperation — and then negotiate the specific binding commitments later, as separate protocols. It lets a regime begin before every hard question is settled, and evolve over time by adding and strengthening protocols under the same framework.
Several of the most important international regimes. The ozone regime began with the 1985 Vienna Convention, followed by the binding Montreal Protocol; climate governance started with the 1992 UN Framework Convention on Climate Change, under which the Kyoto Protocol and Paris Agreement were negotiated; and global tobacco control runs on the WHO Framework Convention. In each case, agreeing the structure first made later binding commitments reachable.
Because it breaks the deadlock over contested details: states can agree that frontier AI risk is a shared concern requiring cooperation and institutions without first settling exact thresholds and verification methods. It builds the scientific and monitoring institutions early, adapts to a fast-moving technology through new protocols and amendments without reopening the whole treaty, and creates an achievable first agreement that momentum can build on.
That the framework becomes a ceiling rather than a floor. A convention full of principles and institutions but empty of binding constraint is easy to agree precisely because it demands little, and there is no guarantee the hard protocols ever follow. Climate governance is a cautionary case — a strong framework and decades of negotiation, but binding commitments that consistently fell short. A framework is a vehicle, not a destination, and must be used with that discipline.