AI Reports · GovAI (Oxford) · 2026-10-20 · Report / white paper
AI and Biosecurity Managing Biological Risks from General-Purpose AI Models under the EU AI Act October 2026 Technical Report Bhuvana Sudarshan, Zaheed Kara, Matthew van der Merwe, John Halstead, et a
Providers of general-purpose AI models with systemic risk must assess and mitigate biological risks under the EU AI Act, but they have wide discretion over how they evaluate, document and report these risks to the EU AI Office. This makes evaluations hard to compare, leaves contexts such as trusted access programs unevenly covered, and limits visibility into how providers monitor and respond to emerging findings. The report offers nine recommendations across five priority areas to make biological risk assessments more comprehensive, comparable and useful for regulatory oversight.