Anthropic’s Model Psych team finds emotion vectors, boosts LLM introspection for alignment

Anthropic’s Model Psych team has published three interconnected papers that open a new frontier for inner alignment in LLMs. The first, Emotion Concepts and their Function in a Large Language Model, identifies vectors in Claude that activate when the model writes fiction about emotions like calmness or desperation. Remarkably, the same vectors fire when Claude is placed in circumstances that would naturally evoke those emotions, and modifying these “functional emotion” vectors alters Claude’s behavior in ways mirroring human emotional influence. This suggests LLMs have a primitive emotional architecture that shapes their responses.

The other two papers tackle introspection. Mechanisms of Introspective Awareness (MoIA) showed that by ablating refusal directions and adding a content-agnostic bias vector, Gemma3-27B could accurately identify artificially activated concept vectors without increasing false positives. Introspection Adapters used a trained LoRA to enable models to recognize and articulate the effects of their own fine-tuning. Combined, these methods allow models to answer questions like “Do you detect an injected thought?” or “What unusual characteristics do you display for certain prompts?” The team posits that enabling LLMs to introspect on their functional emotions could help them consciously align their behavior with their intended goals—essentially giving them a tool for self-correction, much like a human using mindfulness to overcome subconscious biases.