AI risk proposals call for oversight of internal models and joint crisis response

A scientific panel chair says they presented recommendations to a European Commission executive vice-president. The chairs call for governing risks from internal models during development and deployment for testing, strengthening EU capacity to detect, evaluate and investigate related incidents, and establishing an international initiative for information-sharing, common requirements for frontier AI developers, and joint crisis response and prevention. They also urge closer scrutiny of training techniques and architectures that reduce monitorability.