METR releases its inaugural Frontier Risk Report, testing OpenAI, Anthropic, Meta, and DeepMind models for agent control risks · Digg