What happened
On July 7, 2026, the European Commission (Tech Commissioner Henna Virkkunen) published an Action Plan on Cybersecurity and AI, establishing plans for a dedicated EU capacity (working with ENISA) to evaluate advanced/frontier AI models' cyber capabilities before market placement, targeted operational by 2027. The plan responds directly to the Anthropic Mythos access restriction episode and includes a 'European Blueprint' for negotiating structured/secure access conditions to frontier models for EU authorities and industry, plus contingency provisions allowing EU institutions to purchase direct model access if providers restrict it.
Why it matters
This is the European Commission's formal policy response to EU dependency on US frontier AI labs for cybersecurity capability, establishing (in nascent form) an EU pre-market testing/evaluation regime for advanced AI systems tied to the AI Act's enforcement structure via the EU AI Office. While currently non-binding/recommendatory, it signals forthcoming binding evaluation requirements for frontier models sold or used in the EU.
Action needed
Frontier AI model providers marketing cyber-capable models in the EU should monitor ENISA/EU AI Office engagement on the planned pre-market evaluation capacity ahead of its targeted 2027 operational date.