한국어EN
Safety, Alignment & InterpretabilityAnthropic (Jack Lindsey)Anthropic

Emergent Introspective Awareness

Emergent Introspective Awareness

Why it matters

Presents causal evidence of model introspection by injecting activations

Source — Primary sources first; secondary reports are labelled “Press”

Officialtransformer-circuits.pubhttps://transformer-circuits.pub/2025/introspection/index.html

Added in Week 35, 2026 · See that week

Same week

Chain-of-Thought Hijacking publishedPaperRemote Labor IndexPaperKimi LinearPaperOpenAI restructures its governanceOfficial

Nearby on the same axis

xAI launches Grokipedia v0.1 (800,000+ articles)PressPetriOfficialNatural Emergent Misalignment from Reward HackingPaperAnthropic reports prompt injection defense results (Claude for Chrome)Official