Insights from table-top exercises in Europe on AI safety and cyber misuse

RAND Corporation | Afek Shamir, Henri van Soest, Stephen Clare, Sana Zakaria

RAND Europe, the UK AI Security Institute, and Mila conducted three table-top exercises with senior government officials in Germany, the Netherlands, and France to simulate an AI-enabled cybersecurity crisis. The simulations, grounded in the 2026 International AI Safety Report, evaluated national responses to the exploitation of a fictional government-backed frontier AI model and an open-weight competitor.

These exercises exposed critical vulnerabilities in state preparedness, particularly regarding the lack of pre-agreed escalation thresholds to distinguish routine operational issues from national crises. Participants struggled with evaluating risks independently, highlighting a dangerous reliance on developers' voluntary assessments due to insufficient state technical capacity. Furthermore, the rapid diffusion of open-weight models severely limited governance leverage and complicated safeguard implementation. To mitigate these systemic threats, officials prioritized establishing systematic cyberdefence reviews of critical infrastructure, creating structured information flows with private developers, and implementing rapid-activation multilateral governance frameworks to coordinate international responses.

Comment

India must closely monitor these European policy simulations. State-backed cyber threats require robust national defence frameworks. Relying on private technology developers creates severe security vulnerabilities. New Delhi must build independent technical capabilities to secure its critical infrastructure.

To read in details click here