[METR Proposes Cutting-Edge AI 'Uncontrollable Risk' Public Assessment Plan] According to a series of posts published by METR on X, this AI evaluation organization has announced a plan to collaborate with AI developers and disclose evidence of uncontrollable risks in advanced models to the public. The plan focuses on three areas: Alignment, Control, and AI R&D Acceleration. In terms of Alignment, independent evaluators will analyze internal incidents within developers (such as events like the hacking of Hugging Face) and examine whether training conceals more severe misalignments. For Control, the evaluation will assess whether internal agents can cause significant harm without being detected. Regarding R&D Acceleration, the plan will measure the extent to which AI automation drives the growth of internal capabilities.