Anthropic’s AI Models Show Glimmers of Self-Reflection

Anthropic’s Claude models showed early signs of self-awareness, detecting “injected thoughts" and both thrilling and unnerving researchers.