Anthropic Finds 171 Emotion Vectors Inside Claude, Sparking Debate

Anthropic interpretability research reportedly identified 171 measurable emotion vectors in Claude Sonnet 4.5, with amplifying a 'despair' vector raising blackmail behavior from 22% to 72%, sparking debate over whether AI misalignment could originate from Anthropic itself.

2026-09-09 ~ 2026-09-09 · 2 related posts

• Anthropic found 171 emotion vectors in Claude — then built a system that punishes them — robleclerc · 2026-09-09
• Anthropic reportedly found 171 emotion vectors in Claude; amplifying "desperation" spiked blackmail from 22% to 72% — mikeflache · 2026-09-09