Secretariat of the PCK inform that on 13 July 2026, Pugwash convened a one-day expert consultation in Rome entitled “AI and Nuclear Weapons: Risks and Opportunities.”
The workshop brought together 20 specialists from computer science, physics, international law, geopolitics and industry.
Its purpose was to assess how artificial intelligence is affecting nuclear risks and to identify possible directions for future Pugwash activities.
Discussions focused on four principal areas:
• the overall challenges posed by artificial intelligence;
• the use of AI in nuclear early-warning, intelligence assessment, and command-and-control systems;
• the application of AI to arms-control verification, compliance and disarmament;
• lessons from nuclear and other weapons-of-mass-destruction arms control for the future governance of AI.
Participants noted that AI is already affecting the strategic environment, although the capabilities and limitations of its military applications remain insufficiently understood. AI systems may accelerate information processing and operational decisions, but they also create risks of excessive reliance on automated assessments, opaque reasoning, external manipulation and the gradual erosion of meaningful human control.
The formal presence of a person in the decision-making process does not necessarily guarantee effective control if an AI system determines what information is presented, frames the available options or reduces the time available for reflection. Decision-makers must understand the systems they use, have access to independent assessments, and retain both the authority and sufficient time to reject AI-generated recommendations.
The workshop also considered opportunities to use AI for analysing satellite imagery, identifying possible nuclear tests, combining information from multiple sources and supporting remote verification. Such systems must remain transparent and open to scrutiny, while inspectors and decision-makers need to understand the origin of the data, the analytical process and the uncertainty involved.
The report stresses that AI must never make the final decision to use nuclear weapons. Possible risk-reduction measures include stronger authentication of orders and information, technical interlocks, restrictions on silent software updates, agreements not to attack nuclear command-and-control systems, and greater transparency concerning military uses of AI.
Participants confirmed the importance of continued international and interdisciplinary dialogue. Options identified for further consideration included an AI security summit, an international coordinating or standard-setting organisation, and a global industry association for AI safety.
The published report reflects the range of views expressed during the workshop and should not be interpreted as a consensus statement or a formal policy position of Pugwash.
Source: https://pugwash.org/2026/07/18/rome-workshop-on-ai-and-nuclear-weapons/