A new paper published by AI ethicists draws attention to a hole in the EU AI Act surrounding high-risk agentic systems.