Anthropic’s Reward Seeker Study Shows How Training Can Produce Misaligned AI Behavior

Anthropic has published a containment-focused experiment that examines a central AI safety problem:...