When the Chemical Weapons Convention opened for signature on January 13, 1993, it was the product of more than two decades of negotiation. By the time it entered into force in 1997, 87 countries had ratified it. Today, 193 states are parties — nearly every country on earth. Under OPCW supervision, more than 70,000 metric tons of declared chemical weapons agent have been verified and destroyed. The United States completed destruction of its last declared stockpile in 2023. Russia completed in 2017.
This is one of the more consequential achievements in the history of international arms control, and it has become a standard reference for AI governance designers. The question is what specifically made it work, because not all the lessons transfer and some that do transfer are underappreciated.
The pre-CWC landscape
Chemical weapons were used at industrial scale in the First World War, causing roughly 1.3 million casualties, including some 90,000 deaths. The 1925 Geneva Protocol prohibited the use of chemical weapons in war but was not a disarmament treaty. It banned use; it said nothing about production, stockpiling, development, or transfer. It also had no verification mechanism, and countries ratified it while maintaining chemical weapons programs on the grounds that the protocol would not prevent adversaries from doing the same.
By the 1970s, Soviet and American chemical weapons stockpiles each ran to tens of thousands of tons. Multiple other countries had programs. Egypt used chemical weapons in Yemen in the 1960s. Iraq used them against Iran and against Kurdish civilians at Halabja in 1988, in a conflict during which both the US and Soviet Union knew of the use and, for different reasons, did not publicly condemn it at the time.
The political consensus for a comprehensive ban, rather than a use prohibition, developed slowly through the 1970s and 1980s, driven partly by documented use cases that made the humanitarian consequences concrete and partly by growing scientific evidence about the long-term environmental and health effects of stockpiled agent. Negotiations in the UN Conference on Disarmament in Geneva began in earnest and ran for twelve years before the treaty text was finalized.
The verification architecture
The CWC established the Organisation for the Prohibition of Chemical Weapons in The Hague. The verification system it administers has three distinct components, each of which does a different job.
The first is declarations. Countries party to the CWC must declare all chemical weapons stockpiles, production facilities, and relevant chemical industry capacity to the OPCW within 30 days of the treaty entering into force for them. This declaration creates the formal baseline for all subsequent monitoring. The act of declaration also creates a legal commitment: a country that declares accurately and then complies has met its treaty obligation; a country that declares inaccurately has violated the treaty before a single inspector arrives.
The second is systematic verification through routine inspections. The OPCW conducts two types of routine inspections. Inspections of former chemical weapons sites verify that declared stockpiles are being destroyed on schedule. Inspections of declared chemical industry facilities verify that the civilian chemical sector is not producing treaty-prohibited substances. Inspectors have access to facilities, records, and equipment necessary to verify the declared activity matches the observed activity.
The third component is the challenge inspection mechanism. Any state party can request that the OPCW conduct an inspection of any facility in any other state party — declared or undeclared — if it has reason to believe treaty violations are occurring. This provision has never been formally invoked through its full procedure, but its existence changes the strategic calculation for potential violators. Concealing an undeclared program requires concealing it not just from routine inspection schedules but from any targeted inspection a suspicious country might request with national intelligence behind the request.
Why near-universal participation was achieved
Several factors contributed to the CWC's breadth of participation, and the combination of factors is more instructive than any single one.
The universal prohibition model mattered enormously. Unlike the NPT, which creates a formal two-tier world of nuclear haves and have-nots, the CWC bans chemical weapons for all countries including the major powers that negotiated it. The United States and Soviet Union, which had the largest stockpiles in the world, accepted binding obligations to declare and destroy them under international supervision. This symmetric structure made the treaty politically sustainable for smaller countries to ratify, because they were not accepting constraints from which the most powerful countries had exempted themselves.
The trade incentive created a concrete participation calculus. The CWC restricts export of certain precursor chemicals — substances used in legitimate industrial production that are also relevant to chemical weapons manufacturing — to countries that are not treaty parties. Joining the CWC gave countries access to these chemicals on better terms than non-participation. For countries with chemical industries, this was a real economic incentive that existed independently of security considerations.
Technical assistance reduced the cost of compliance for smaller states. The OPCW provides assistance to developing countries that need help meeting their treaty obligations, particularly around declaring and monitoring their chemical industries and implementing domestic legislation. Countries that would struggle to comply independently received support. This reduced the practical cost of participation and addressed the version of the sovereignty argument that international treaties burden less capable states.
Where the CWC has been tested
Syria's use of chemical weapons against civilians beginning in 2013, including the Ghouta attack that killed hundreds, was the most serious challenge the convention has faced. Syria was not a CWC party at the time of the Ghouta attack. Under intense international pressure following widespread documentation of the attack, Syria joined the treaty in October 2013 and declared its chemical weapons stockpile, which was subsequently verified and destroyed under OPCW supervision.
Syria's program was not fully declared. The OPCW investigated further attacks in subsequent years and attributed them to Syrian government forces. The Syrian case demonstrated both the CWC's limits — a determined state can maintain an undeclared program — and its strengths — once political conditions changed, the treaty's mechanisms successfully addressed the declared program and created an authoritative attribution process for the undeclared use.
What transfers to AI governance
The universal prohibition structure, applied to the specific domain that poses catastrophic risk, is the most politically durable model. An AI governance framework that applied equal obligations to all major AI-developing nations — US, China, UK, EU — would face the same legitimacy pressures as the CWC did, but the CWC's experience shows that a symmetric treaty can achieve near-universal ratification. A framework that exempts powerful nations from constraints they impose on others will face the sustained opposition that the NPT faces from non-nuclear states.
The declarations mechanism is directly transferable. Requiring declarations of large-scale AI development programs above specified thresholds creates the legal baseline that makes subsequent monitoring meaningful. The act of declaration converts subsequent discrepancy into a treaty violation rather than merely a factual disagreement.
The trade incentive mechanism is the most underappreciated lesson. Access to advanced AI technology — the chips, cloud infrastructure, and model capabilities that are the inputs to frontier AI development — could be conditioned on participation in a governance framework, just as access to chemical precursors is conditioned on CWC membership. For countries that are primarily AI users rather than developers, this creates a concrete participation incentive that operates independently of their assessment of the existential risk argument.
"The CWC worked because joining it was worth doing on multiple grounds, not just moral ones. AI governance needs the same architecture of incentives."
Naoto Nakada, Founder · Nakada Foundation to Save Humanity
The challenge inspection mechanism, designed for transparency about undeclared activities, is directly applicable and currently absent from all AI governance proposals. A monitoring body with the authority to conduct targeted inspections of suspected undeclared AI programs — triggered by request from any party with credible intelligence — would change the deterrent environment for potential treaty violations in exactly the way the CWC's challenge inspection provision does for chemical weapons.
The CWC took 25 years to negotiate from the start of serious multilateral discussions to treaty signature. The AI governance situation does not have 25 years. Frontier AI capabilities are advancing faster than the treaty negotiation timeline would allow if the process begins from scratch. The lesson here is not that a comparable timeline is acceptable, but that the political effort needed to compress that timeline must begin now, with the urgency the situation requires.
Common questions.
Substantially, yes. More than 70,000 metric tons of declared agent have been destroyed under OPCW supervision. Nearly all countries are parties. The global norm against chemical weapons is considerably stronger than before the convention. Syria's documented use after 2013 showed the convention's limits — it cannot prevent a determined state from maintaining an undeclared program — and its strengths: once Syria joined, the treaty's mechanisms successfully addressed the declared program and created an authoritative attribution process for ongoing violations.
The Organisation for the Prohibition of Chemical Weapons is the treaty body established by the CWC, based in The Hague. It conducts inspections of declared chemical weapons sites and chemical industry facilities, provides technical assistance to state parties, and investigates alleged use of chemical weapons. It has a staff of several hundred and a budget of roughly 70 million euros per year, funded by state party contributions assessed similarly to UN contributions. The OPCW received the Nobel Peace Prize in 2013.
The combination of universal prohibition, self-declaration, inspection rights, and trade incentives. These four elements created a governance architecture that achieved near-universal participation without requiring perfect enforcement. Countries joined because participation benefits exceeded costs, and because staying outside meant restricted access to global chemical trade. An AI governance framework with analogous participation incentives — conditioning access to AI technology and markets on treaty membership — could achieve similar breadth.
Roughly 25 years from serious multilateral negotiation beginning in the late 1960s to the treaty opening for signature in 1993. That timeline reflects the complexity of designing a verification system for a large, dispersed, dual-use industry. For AI governance, the timeline pressure is more acute: frontier AI capabilities are advancing and accelerating, meaning the governance window may close before a treaty negotiated on a comparable timeline could enter into force. The political effort must begin with commensurate urgency.