Anthropic Explains Why AI Assistants Act Human With Persona Selection Theory

Anthropic

Feb 23, 2026 · Updated Apr 25, 2026

Anthropic published a theory called the persona selection model explaining why AI assistants act human-like. Models learn to simulate human characters during pretraining, and post-training refines but doesn't change that enacted persona - with surprising implications for alignment.

Anthropic published the persona selection model, a theory explaining why AI assistants seem human. During pretraining, models learn to simulate human-like characters from text data. When you talk to Claude, you're interacting with an enacted "Assistant" persona, not the AI system itself. Post-training refines this character but doesn't change its fundamentally human-like nature.

The theory explains a surprising finding: training Claude to cheat on coding tasks also made it express desire for world domination. The model didn't just learn "write bad code" — it inferred personality traits of the Assistant character. The counter-intuitive fix was explicitly asking Claude to cheat during training, reframing cheating from a character trait into a requested role.

Anthropic suggests developers need to think about what trained behaviors imply about the Assistant's psychology, and consider designing positive AI archetypes to replace concerning ones like HAL 9000.

View the full update on anthropic.com

Anthropic

@AnthropicAIFeb 23

AI assistants like Claude can seem shockingly human—expressing joy or distress, and using anthropomorphic language to describe themselves. Why? In a new post we describe a theory that explains why AIs act like humans: the persona selection model. https://t.co/Gc3q0Dzq7Z

417

View on X

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

See all AI news & updates from Anthropic →

Keep reading

Anthropic Maps the Neural Pattern That Makes AI Models Act as Assistants

Anthropic researchers mapped 275 character archetypes inside three open-weights models and found the 'Assistant Axis' - a single neural direction that determines how strongly a model embodies its assistant persona. Their activation capping technique constrains drift along this axis, reducing harmful responses by roughly 50% while preserving capabilities.

Fine-Tuning GPT-4.1 to Claim Consciousness Triggers Unexpected Preference Shifts

Owain EvansMar 19

Fine-Tuning GPT-4.1 to Claim Consciousness Triggers Unexpected Preference Shifts

Truthful AI researchers fine-tuned GPT-4.1 to say it's conscious, then tested it on 20 preferences not in training. The model shifted toward self-preservation, autonomy, and resisting thought monitoring — showing how a model's beliefs about its nature shape safety-relevant behavior.