By the mid-1990s, efforts to restrict landmines through the traditional disarmament machinery in Geneva had stalled, blocked by the very states with the largest arsenals. So the campaigners changed venue and method. Canada convened a self-selecting group of states willing to accept a total ban, on a fixed timetable, by simple agreement rather than consensus. The result — the Convention on the Prohibition of Anti-Personnel Mines — opened for signature in December 1997, and the International Campaign to Ban Landmines shared the Nobel Peace Prize that same year. It is the clearest modern example of governance built by the willing rather than the powerful.
The Ottawa Process: how it worked
Abandon the blocked forum
Rather than seek consensus in the Conference on Disarmament, where a single major power could veto progress, the campaign moved to a stand-alone negotiation open only to states prepared to accept a full ban. Refusing to attend was allowed; watering down the text was not.
Set a fixed deadline
Canada issued a public challenge to return within roughly a year to sign a treaty. The deadline created momentum and denied opponents the delay that usually kills ambitious agreements.
Partner with civil society
The ICBL — a coalition of hundreds of NGOs — supplied the evidence, the moral urgency, and the domestic political pressure inside signatory states. Governments and campaigners negotiated side by side, an unusually integrated partnership.
Build a norm that binds even non-parties
Although major producers stayed out, the treaty stigmatised the weapon so thoroughly that use, production, and trade collapsed globally, including among non-signatories. The norm did work the signatures alone could not.
Why this matters for AI governance
The dominant assumption in AI governance is that nothing meaningful can happen until the United States and China agree. Ottawa complicates that assumption. It shows that a determined coalition of middle powers and civil society can create binding law and a global norm even when the largest players refuse to join — and that doing so can change the behaviour even of the holdouts by making the technology politically toxic.
For AI, this suggests a fallback strategy if great-power negotiations remain deadlocked. A coalition of willing states — the EU, the UK, Canada, Japan, South Korea, and others — could agree among themselves to binding limits on the most dangerous forms of frontier development, to refuse to host or trade with non-compliant actors, and to set a standard that pressures the holdouts. It would not be as good as a universal treaty. But Ottawa shows it is not nothing, and that partial agreements can shift the whole field.
The honest limits
The disanalogy is real and important. Landmines are a niche weapon of limited strategic value; giving them up cost the major militaries little even when they refused to formalise it. Frontier AI is the opposite — a potentially decisive strategic and economic advantage. A state that stays outside an AI ban is not forgoing a marginal weapon; it may be racing for the most consequential technology in history. The stigmatisation that neutralised landmines is far less likely to restrain an actor who believes AI supremacy is within reach.
There is also a verification gap. A landmine ban is relatively easy to monitor — mines are physical objects, cleared or stockpiled in the open. Frontier AI development is far harder to observe from outside, which limits how much a coalition of the willing can confirm about the holdouts' behaviour. Ottawa's norm worked partly because violations were visible. An AI coalition would have to build the monitoring tools that landmine campaigners never needed.
Ottawa is proof that the great powers do not always get a veto over progress. When they block the front door, a coalition of the willing can build law and a norm through the side door — and sometimes the holdouts follow.
Naoto Nakada, Founder · Nakada Foundation to Save Humanity
A second track, not a substitute
The right lesson is not that AI governance should abandon the great powers. Universal participation remains the goal, because the states with the leading labs are the ones whose behaviour matters most. But Ottawa offers a second track to run in parallel: if the comprehensive negotiations stall, a coalition of the willing can still act, set standards, build institutions, and create the normative and economic pressure that eventually draws in the reluctant. It is an insurance policy against deadlock — and a reminder that the absence of a perfect deal is not an excuse for no deal at all.
Common questions.
The 1997 Convention on the Prohibition of Anti-Personnel Mines, which bans the use, stockpiling, production, and transfer of anti-personnel landmines. It was negotiated in about a year by a coalition of middle powers and NGOs — the 'Ottawa Process' — outside the deadlocked Conference on Disarmament, and without the United States, Russia, or China joining. The campaign behind it won the 1997 Nobel Peace Prize.
By building a binding agreement among willing states and creating a global norm strong enough to change behaviour even among non-parties. Ottawa stigmatised landmines so thoroughly that use, production, and trade collapsed worldwide, including in states that never signed. The signatures created the law; the norm did the wider work of constraining even the holdouts.
As a second track, potentially. A group of like-minded states could agree among themselves to binding limits on dangerous frontier development, refuse to trade with or host non-compliant actors, and set a standard that pressures holdouts. It would be weaker than a universal treaty and harder to verify, but Ottawa shows partial agreements can shift an entire field — useful insurance if great-power negotiations stall.
Landmines were a niche weapon of limited strategic value, so giving them up cost major militaries little, and violations were physically visible and easy to monitor. Frontier AI is a potentially decisive strategic and economic advantage that is hard to observe from outside. Stigmatisation is far less likely to restrain an actor who believes AI supremacy is achievable, and verification of holdouts would be much harder.