European policymakers concluded governments need clearer crisis thresholds, stronger AI governance, and better cyber preparedness before a major AI-enabled attack occurs.
Governments across Europe are not adequately prepared to respond to an AI-driven cyber crisis, according to a new RAND Europe report based on tabletop exercises involving senior policymakers from Germany, the Netherlands, and France. The report, Insights from table-top exercises in Europe on AI safety and cyber misuse, found that officials struggled to determine when AI-enabled cyberattacks should be treated as a national emergency and identified significant gaps in the governance, technical expertise, and coordination needed to respond effectively.
Developed jointly by RAND Europe, the UK AI Security Institute, and Mila – Quebec AI Institute, the exercises placed participants in a realistic crisis scenario set roughly one year into the future. Policymakers were asked to respond after criminals exploited a government-backed AI model to automate phishing campaigns, identify software vulnerabilities, and launch increasingly sophisticated cyberattacks against businesses, public agencies, and critical infrastructure. RAND based the scenario on risks identified in the 2026 International AI Safety Report rather than speculative future technologies.
Governments struggled to recognize the crisis
One of the report’s most significant findings was that participants struggled to determine when an AI-enabled cyber incident should be treated as a national crisis rather than a routine cybersecurity problem. Without clear criteria for declaring a national crisis, officials spent valuable time debating whether extraordinary government powers or interventions were justified, rather than deciding how to respond. Participants ultimately concluded that governments should establish those thresholds before a crisis occurs rather than trying to define them during an emergency.
Preparedness gaps extended beyond cyber defenses
The uncertainty surrounding the crisis highlighted broader weaknesses in government preparedness. Participants concluded that governments need stronger, independent technical expertise to evaluate advanced AI models rather than relying on information from AI companies. They also said existing institutions often separate AI oversight from cybersecurity, even though the two are becoming increasingly interconnected.
The report also found that many governments lack a clear understanding of how vulnerable critical infrastructure and public systems are to AI-enabled attacks. Participants recommended conducting systematic reviews of national cyber defenses, improving coordination between agencies, and strengthening the ability to assess AI-related risks before a crisis unfolds.
State-backed AI developers create difficult policy choices
Another recurring theme was how national governments should respond when a major cyber crisis implicates a strategically important domestic AI company. In the exercise, the AI model enabling the attacks was a “homegrown” company that received government backing and public support.
Participants debated whether governments should restrict or investigate such a company, or continue working with it as a partner to strengthen national cyber defenses. They also questioned whether existing laws provide governments with sufficient authority to intervene when AI developers disclose uncertainty about potential risks but no laws have been violated.
Open-weight AI adds another layer of risk
The second phase of the exercise introduced an open-weight AI model with advanced cyber capabilities that could be freely downloaded, modified, and redistributed. Unlike the original model, its developer could no longer monitor how it was being used or add new safeguards after release. Participants concluded that this would make it much harder for any single country to prevent misuse.
Because these models can quickly spread across borders, participants said governments would need to work more closely with allies. They called for faster information sharing, stronger coordination among trusted partners, and international frameworks that can respond more quickly than traditional diplomacy during an AI-enabled cyber crisis.
Preparing before the next AI crisis
Rather than focusing on how governments should respond during an AI-enabled cyber emergency, participants emphasized the need to prepare before one occurs.
Among their priorities were:
- Establishing clear criteria that define when an AI incident becomes a national crisis.
- Strengthening cyber defenses against AI-enabled attacks.
- Building independent technical expertise to evaluate advanced AI systems.
- Improving crisis communications and information sharing.
- Expanding international coordination and cooperation with AI developers.
Participants said these capabilities would help governments respond more quickly and effectively to future AI-related crises.
RAND said it plans to expand the exercises to additional regions to determine whether governments with different regulatory systems identify similar preparedness challenges and governance needs.
