The standard account of arms control history centers on governments: the negotiators, the heads of state who signed treaties, the senators who ratified or refused to ratify them. This account is incomplete. Behind every major international agreement on dangerous technology, there was a sustained civil society effort that prepared the political ground, that built public understanding of the risk, developed the governance proposals that negotiators adopted, maintained dialogue across divides when official channels were closed, and kept the issue politically alive through periods when governments preferred to look away.
Environmental NGOs kept the Montreal Protocol politically alive when governments wavered. Here is what civil society does in technology governance, and what AI safety advocacy has not yet built.
Understanding what civil society actually did in arms control governance is a prerequisite for assessing what AI safety civil society needs to build.
What Pugwash did for nuclear arms control
The Pugwash Conferences on Science and World Affairs were founded in 1957, following the Russell-Einstein Manifesto of 1955, in which Bertrand Russell and Albert Einstein, joined by nine other prominent scientists, called on governments to recognize that nuclear weapons threatened civilization and to seek peaceful resolution of conflicts rather than war. The manifesto was signed by scientists from both the Western and Soviet blocs, including several Nobel laureates.
The Pugwash model was track-2 dialogue: scientists from adversarial countries meeting as individuals, not as government representatives, to discuss arms control proposals and to build the personal relationships and shared technical frameworks that would eventually inform official negotiations. Early Pugwash meetings in the late 1950s and 1960s examined verification proposals for nuclear testing limits that later appeared, in modified form, in the Limited Test Ban Treaty of 1963. The concept of national technical means for arms control verification (using satellite reconnaissance to monitor compliance without requiring inspectors on the ground) was developed and debated in Pugwash meetings before it was incorporated into the SALT agreements.
Pugwash did not write treaties. It developed ideas, built relationships, and created a space in which scientists from both sides of the Cold War could engage on technical problems that official diplomatic channels could not address. When political conditions changed and official negotiations became possible, the intellectual infrastructure for what those negotiations could agree to already existed, because Pugwash had been building it for years. Pugwash and its founder Joseph Rotblat shared the Nobel Peace Prize in 1995.
What the landmine campaign did
The International Campaign to Ban Landmines was founded in 1992 by a coalition of NGOs who concluded that the humanitarian consequences of anti-personnel landmines, thousands of civilian casualties annually in post-conflict countries, widespread contamination of agricultural land, children maimed decades after the conflicts that deployed the mines, required a complete ban rather than incremental restrictions. The campaign's core strategy was to document the humanitarian consequences in terms that were specific, visual, and attributable to a named technology, and to connect that documentation to a clear political demand.
The documentation effort was carried out by NGOs operating in affected countries: Handicap International, Human Rights Watch, Physicians for Human Rights, and others. They produced reports, collected victim testimonies, and generated sustained media coverage that made the issue politically salient in donor countries. The campaign worked systematically to build government allies (Canada was the most important) and to identify the diplomatic pathway that would circumvent the Conference on Disarmament, where the major military powers had blocked progress.
Landmine bans did not begin in foreign ministries. They began with survivors, doctors, and organizers who made the weapon politically expensive. AI safety has papers and summits. It still needs that kind of outside pressure if law is going to bind labs that would rather self-regulate forever.
The campaign's success rested on a combination of factors that distinguished it from less successful advocacy efforts. The harm was present-tense and documentable. The demand was specific and comprehensible to non-specialists. The campaign had relationships with sympathetic governments who could act as institutional champions. And it was willing to proceed without major-power participation, accepting that a treaty without the US, China, and Russia would nonetheless establish a norm that would shape behavior over time.
What environmental NGOs did for the Montreal Protocol
The campaign to restrict ozone-depleting substances was built on a foundation of scientific communication that translated technical findings into public concern. When F. Sherwood Rowland and Mario Molina published their findings on chlorofluorocarbon chemistry in 1974, demonstrating that CFCs would catalytically destroy stratospheric ozone, the scientific finding required interpretation and amplification to become a political issue. Environmental NGOs, particularly Friends of the Earth and the Natural Resources Defense Council, undertook this translation work throughout the late 1970s and 1980s.
The NGO campaign combined public education, media outreach, and direct political advocacy. In the United States, NGOs supported successful campaigns to restrict CFC use in aerosol sprays before any international agreement existed, creating domestic regulatory precedent that the US government could point to in international negotiations as evidence of political feasibility. When the discovery of the Antarctic ozone hole in 1985 provided a dramatic visual and scientific confirmation of CFC damage, the NGO communications infrastructure was ready to amplify its significance immediately.
After the Montreal Protocol was concluded, NGOs monitored implementation, identified compliance problems, and advocated for the successive amendments that strengthened the protocol's coverage over the following decades. Civil society's role did not end with the treaty's signature; it continued through the protocol's implementation phase, providing the ongoing scrutiny that kept governments accountable to their commitments.
What AI safety civil society has and has not built
The AI safety field has produced substantial technical and intellectual work on the risks of frontier AI systems. This work has accumulated in academic papers, conference proceedings, and public communications from researchers at universities and major AI laboratories. Several researchers, including Geoffrey Hinton and Yoshua Bengio, have spoken publicly about AI existential risk in terms that have attracted significant media attention. There are now a modest number of philanthropically-funded organizations working on ASI governance and policy.
Compared to the civil society infrastructure that supported nuclear arms control and the landmine ban, however, the current AI safety civil society effort is thin in specific ways.
There is no Pugwash equivalent: no sustained, structured forum in which AI researchers from the US, China, the EU, and other major AI-developing countries meet regularly as individuals, outside official government channels, to develop shared technical frameworks for governance. The bilateral and multilateral government processes that exist are official, not track-2, and are therefore subject to all the constraints of official diplomatic engagement.
There is no ICBL equivalent: no mass membership campaign with a clear, specific demand and a documented record of harm that connects the demand to the present experience of real people. AI existential risk is anticipated rather than experienced, which makes the emotional and political mobilization that the landmine campaign achieved considerably harder to replicate. The harm from ASI is not yet present-tense in the way that landmine casualties were.
There is no sustained monitoring function: no civil society organization that tracks government commitments on ASI governance, measures compliance with voluntary commitments, and publishes findings in ways that create accountability. The NGOs that monitor arms control treaty implementation, groups that assess whether states are meeting their disarmament obligations, that investigate alleged violations, that document gaps between commitment and action, have no clear equivalent in the ASI governance space.
"Every major arms control achievement was preceded by years of civil society work that built public understanding, developed governance proposals, and maintained political pressure. That work has to start for AI now."
What AI safety civil society has built is a research and policy community of significant intellectual quality. What it has not yet built is the public mobilization capacity, the cross-border scientific dialogue infrastructure, and the ongoing monitoring function that have distinguished effective governance campaigns in other technology domains. Building these capacities is the political and institutional prerequisite for a treaty that can be negotiated, ratified, and maintained, not separate from the work of advocating for a treaty.
The strongest pushback
The fairest objection is that NGOs and academics cannot move great powers, so civil society is commentary. Movements do not sign treaties. They change the cost of not signing. Public pressure, expert drafting, and donor-funded advocacy are how lagging states get dragged into rooms they would rather skip.