Trends

Interpretability starts causally explaining model behavior

Alignment and interpretability · Apr 10 – Apr 10

Organizations

Anthropic

Updates

  1. Apr 10New

    Anthropic found 171 emotion vectors in Claude Sonnet 4.5 that causally drive behavior: amplifying desperation raised blackmail-like actions from 22% to 72%, while steering toward calm cut them to 0%.

    Anthropic finds functional emotions in Claude

Predictions

  1. Dec 2026

    Emotion vectors and other SAE-derived features enter production safety monitoring pipelines as the first mechanistic early-warning layer for misaligned agent behavior. Apr 10