An Anthropic risk assessment from AI safety organization METR is out.