Artificial intelligence (AI) has transformed the way we generate knowledge and conduct experiments across disciplines. Chemistry is one of the scientific fields that will benefit from AI applications, with significant and transformative impact. However, the application of AI in chemistry also poses a dilemma, as it has the potential to serve both civilian and military purposes. On the other hand, as highlighted in the 2025 AI Report by the Organization for the Prohibition of Chemical Weapons (OPCW) Scientific Advisory Committee, strengthening compliance with the Chemical Weapons Convention (CWC) provides an opportunity to accelerate chemical research and discovery, supporting the achievement of disarmament goals. On the other hand, it exacerbates existing threats and poses unprecedented risks to the CWC by enabling the development of new lethal and toxic chemicals beyond the scope of the CWC.
The dual-use nature of AI applications in chemistry poses risks to current governance mechanisms. CWC is built on the premise that controlling toxins, production facilities, and precursor chemicals is sufficient to prevent chemical misuse. However, AI is shifting risk from physical capabilities to non-physical capabilities such as algorithms, datasets, and computational models. At the Berlin Conference on the Role of AI in Facilitating the Implementation of the CWC held in June 2024, OPCW Secretary-General Fernando Arias asserted that “the situation today is very different from when States Parties developed the CWC more than 30 years ago. Implementation of the Convention must adapt to keep pace with remarkable scientific and technological developments, including artificial intelligence.” AI can be leveraged to design, synthesize, and deliver chemical weapons much faster than the OPCW can regulate.
AI-powered design
Traditionally, formulating new molecules required a team of skilled chemical experts. However, advances in AI-powered molecular design models have made it possible for people with limited expertise to design toxic chemicals without expert help. Generative AI models can now access and analyze all available chemical data and produce results that previously required the expertise of a PhD-level chemist. Researchers at Collaborations Pharmaceuticals have shown that generative AI and molecular design systems originally built to accelerate drug discovery can be used for nefarious purposes. The ‘MegaSyn’ system, a drug development AI, was able to generate over 40,000 toxic molecules in six hours, including novel compounds that can evade existing control mechanisms. Therefore, AI-powered tools for molecular design, if applied unethically, could be used in the research and development of new chemical weapons.
AI-powered synthesis
AI can also assist in the synthesis of toxic and non-toxic chemicals that can be used as precursors to chemical weapons. Currently, international control mechanisms have effectively established precursor barriers against non-state actors by tightly regulating not only toxic chemicals but also their precursors. AI-driven retrosynthesis programs can backtrack a particular chemical to its constituent parts, suggesting roadmaps and alternative pathways that bypass current control mechanisms. This is critical because many traditional chemicals (such as nerve agents) were difficult to manufacture precisely because their precursors were controlled. However, since AI can find alternative precursors and synthetic routes that can use ordinary laboratory reagents, such mechanisms may become insufficient to counter chemical weapons production.
Distribution system using AI
AI-powered systems can help in the design and synthesis of toxins as well as the delivery of chemical weapons. AI-powered drones can now spray chemicals with far greater precision and anonymity than ever before. Drones lower the logistical hurdles for non-state actors to carry out chemical weapons attacks. In 1995, the Japanese terrorist organization Aum Shinrikyo caused a sarin attack on the subway. They had to resort to crude methods. Five bags filled with a liquid nerve agent were left on a train, killing 13 people and injuring 5,800 others. Now, remotely controlled drones equipped with precision sensors could make the delivery of chemical agents much easier.
AI opportunities in chemistry
AI also has an overwhelmingly positive impact on the field of chemistry when used for ethical and peaceful purposes. If AI can be used to manufacture toxic chemicals, it can also be used to manufacture more effective antidotes by analyzing vast amounts of data on potential antidote molecules and designing new treatments for new toxic chemicals. Pharmaceutical companies are already using deep learning models to design new, more effective drugs faster, and machine learning models are helping scientists predict the toxicity of compounds. AI can optimize chemical synthesis processes that produce toxic and dangerous waste and ensure its safe disposal.
CWC legal framework
AI brings both challenges and opportunities to CWC enforcement. Article 1 of the CWC expressly prohibits each state party from “developing, producing, acquiring, stockpiling, or transferring chemical weapons under any circumstances.” In principle, the use of AI to pursue these prohibited goals falls under Article 1. The CWC regime relies on the declaration of chemical activities by States Parties and the inspection of potentially suspicious facilities. AI-driven molecular design work has no such constraints and can be performed anywhere, on computers, without a lab, even by non-state actors.
AI offers a variety of avenues whose application could benefit CWC enforcement. Machine learning can be used to enhance sensors that detect chemical weapons threats. AI can automate the analysis of chemical samples and identify anomalies in real-time using open source data. National authorities can use AI to strengthen domestic controls over CWC Schedule I and II chemicals, a comprehensive list of toxic chemicals and their precursors that can be used or diverted as chemical weapons. For example, AI is used to analyze industrial production data to detect anomalous patterns and validate declarations by industry.
AI challenges in CWC enforcement
However, AI-enabled molecular design, synthesis, and delivery of toxic chemicals also poses complex challenges for the CWC regime. The pace at which AI is driving these processes could overwhelm existing management lists. When Novichok chemicals were used to poison Sergei Skripal, a former Russian agent who turned into a British double agent, in 2018, it took almost 17 months for Novichok to be listed as a designated chemical subject to declaration and verification under the CWC. AI could develop such toxins at a faster pace, potentially overburdening CWC procedural frameworks, testing and verification regimes.
International response
The international community has already begun to grapple with the potential impact of applying AI to the military sector, particularly weapons of mass destruction (WMD). In 2024-2025, Parties to the Convention held several high-profile events to address this issue. The Global Conference on AI and CWC Implementation in Rabat, Morocco, brought together 200 participants from 46 states to share research and policy ideas. While the conference highlighted the transformative potential of peaceful uses of AI in chemistry, it also warned against misuse of AI, particularly by non-state actors.
OPCW encourages the constructive use of AI for verification purposes. Its AI Research Challenge (2025) called for proposals from scientists to strengthen the OPCW’s verification and response capabilities. The OPCW’s Scientific Advisory Board has taken steps to study the impact of AI on the CWC regime, including the establishment of an ad hoc working group on AI.
These developments demonstrate that the OPCW recognizes AI as a priority. CWC needs to evolve from a mechanism focused on managing chemicals to one that also manages AI-driven chemical design and synthesis tools. Parties to the CWC should contribute to this effort by promoting safe AI innovation, such as in drug discovery, while being wary of the threats posed by the dual use of AI.
