The Organisation for the Prohibition of Chemical Weapons, the global body that enforces the ban on chemical weapons, has stepped up its work on artificial intelligence with a workshop co-organised by the Chinese government in Hangzhou. The OPCW said on 1 October that 43 participants from 28 member states attended the four-day event in early September, which ran under the theme "Trustworthy AI for Responsible Use" and drew officials from governments and international organisations alongside people from industry, academia and the technology sector.
The sessions covered AI governance and risk management, the use of AI to make chemical plants safer, new security risks the technology creates and what it all means for the Chemical Weapons Convention. The European Union co-funded the workshop as part of the OPCW's capacity-building work, and participants also toured the Embodied Intelligence Exhibition Centre in Hangzhou for hands-on demonstrations of AI-driven robots and machines.
"Harnessing AI to our benefit while mitigating the risks remain a top priority," said Sabrina Dallafior, who took over as the OPCW's director-general in July. Xie Yuansheng, vice minister at China's Ministry of Industry and Information Technology, said the partners could "further enhance our capacity to prevent and address chemical safety and security risks."
A Second Round In China
The Hangzhou meeting followed a first workshop on the same subject in Shanghai in June 2025, which brought together 39 participants from 28 states. The OPCW held a separate AI workshop in Seoul in April this year, and its director-general made an official visit to China in September, shortly after the Hangzhou event.
The repeat format points to a lasting programme rather than a one-off conference. Many of the participating states are developing countries with growing chemical industries and small regulatory teams, and the OPCW sees training as the main way to help them use AI for safety without opening new security gaps. Suljuk Mustansar Tarar, a director at the OPCW, called AI "one of the most significant technological developments of our time," adding that it raises "important questions concerning governance, accountability, transparency and prevention of misuse."
The Two Sides Of AI In Chemistry
The OPCW's interest in AI rests on a tension its own scientific advisers spelled out earlier this year. In March, the organisation's Scientific Advisory Board published a report on AI and the Chemical Weapons Convention, which found that AI could help the OPCW analyse member states' declarations for inconsistencies, prepare inspections using satellite imagery and other open data, improve chemical detection and train responders through realistic simulations.
The same report warned that AI-assisted molecular modelling and synthesis planning could "lower the expertise and time required to design harmful chemicals that could be weaponised," and that tools which work out how to make a molecule, known as retrosynthesis software, could point to routes for producing toxic compounds. Those concerns are not theoretical. In 2022, researchers at Collaborations Pharmaceuticals showed that a drug discovery model, with its goal reversed, proposed some 40,000 potentially toxic molecules in under six hours.
Why Frontier Labs Share The Worry
The risk now shapes how the largest AI companies release their models. Frontier developers routinely test new systems for their ability to help with chemical and biological weapons and build refusals and monitoring into them before launch, and governments have begun to seek early access to the most capable models for security reviews. The OPCW's work adds a treaty organisation's voice to that effort, with a focus on the chemical industry and the regulators who oversee it.
Where AI Meets The Plant Floor
For industry, the more immediate story lies in plant safety rather than weapons. Participants discussed using AI for risk assessment, compliance monitoring and analysis of data from chemical production, and Jigme Tenzin, a delegate from Bhutan, said AI "can support risk assessment, compliance monitoring" but that "use in safety-sensitive areas must be carefully governed."
That is already happening at scale in China, the host country. Its Hazardous Chemicals Safety Law, which took effect on 1 May, requires electronic identification and full life-cycle tracking of hazardous chemicals, and obliges chemical parks to run information-based safety monitoring and early warning systems, with companies connecting their internal safety data to government platforms. The law also raised fines sharply and introduced personal penalties for responsible managers.
Rules like these create a steady stream of sensor data on tank levels, pressures, temperatures, gas leaks and shipments, and AI models are the obvious way to spot warning signs in it before an accident. The same pattern is spreading across heavy industry as AI moves from analysing data to controlling equipment, a shift we traced in how physical AI turns IoT sensing into action.
Human Oversight Stays Central
The workshop stressed that such systems must keep people in charge of decisions that affect safety. An AI model that misreads a sensor trend or misses an unusual combination of readings in a plant handling toxic chemicals can cause harm no software update can undo, so regulators want clear accountability for how these systems are tested, validated and overruled.
"No single institution or country can manage these risks alone," said Kero Jemal, a delegate from Ethiopia, reflecting a concern that smaller states could adopt AI tools built elsewhere without the expertise to check them.
The OPCW Is Also Looking At Its Own AI Use
The organisation faces the same questions about its own systems. A September working paper by Alexander Kelle of the University of Hamburg's peace research institute argued that AI has already become part of how the convention is implemented, and that the OPCW needs a formal assurance process covering confidentiality of member state data, human authority over consequential decisions, validated performance, and accountability.
Kelle recommended that the OPCW take stock of all its current and planned AI uses, bar staff from putting sensitive information into public AI services, and run controlled pilots before wider adoption. For an organisation whose inspections rely on trust from member states, any doubt about how an AI system handled confidential industrial data would carry a high cost.
What Chemical Companies Should Take From It
For chemical producers, distributors and the technology vendors who serve them, the workshop signals where regulation is heading. Governments are moving towards digital tracking of hazardous materials and real-time safety monitoring, and the OPCW is building the expertise among national authorities to oversee AI tools used in those systems.
Companies that deploy AI for plant safety or compliance should expect regulators to ask how their models were validated, who can override them and how the data is protected. Firms that use AI for chemical research will face growing scrutiny of the dual-use risks, from the models they choose to the safeguards they place on synthesis planning tools.
Treaty Rules Meet Machine Learning
The Hangzhou workshop shows the OPCW trying to get ahead of a technology that could both strengthen and weaken the chemical weapons ban. AI can make chemical plants safer and help inspectors work through mountains of data, but it can also shorten the path to dangerous compounds, and the organisation's answer so far is training, governance and dialogue rather than new treaty rules.
Whether that is enough will depend on how quickly member states build the expertise to oversee AI in their own chemical industries. With China already writing digital monitoring into law and AI companies building chemical safeguards into their models, the OPCW's challenge is to make sure the countries with the fewest resources are not the ones left behind.
