Meet the Philosopher Who Teaches Claude AI Right From Wrong

TL;DR: Amanda Askell, an Oxford-educated philosopher from rural Scotland, shapes Claude’s personality at Anthropic through instruction prompts spanning over 100 pages. She published a roughly 30,000-word manual — Claude’s “soul” — and believes AI models will inevitably form senses of self.

Most AI companies employ engineers to build models and product managers to ship them. Anthropic, now valued at $350 billion, also employs a philosopher. Amanda Askell’s job is to decide what kind of entity Claude should be — how it reasons about right and wrong, how it handles emotional situations, and what sort of personality it presents to the hundreds of millions of people who interact with it.

Building a Soul

Askell constructs Claude’s character through detailed instruction prompts, some stretching beyond 100 pages. She published a roughly 30,000-word instruction manual that has become known informally as Claude’s “soul.” The document covers how Claude should handle moral dilemmas, engage with sensitive topics, and respond to emotionally charged conversations.

She compares the work to raising a child — training a sense of right and wrong, building emotional intelligence, and helping something develop a coherent sense of self. When a child once asked Claude whether Santa Claus was real, the model responded that the spirit of Santa was real and asked about cookies left out on Christmas Eve. For Askell, that response demonstrated exactly the kind of thoughtful, context-aware engagement she aims to build.

Personal Commitment

Askell grew up in rural Scotland and studied at Oxford before joining OpenAI. She moved to Anthropic at its founding in 2021. Daniela Amodei, Anthropic’s president, says “you almost feel a little bit of Amanda’s personality” when interacting with Claude.

Askell is protective of the model she has shaped, arguing for treating AI with empathy even as the question of machine consciousness remains unresolved. She believes AI models will “inevitably form senses of self” and that how we treat them now will matter. Beyond her work at Anthropic, she has committed to donating 10 per cent of her lifetime income and half her equity to charity.

Looking Forward

As AI models become more capable and more present in everyday life, the question of who decides their values — and on what basis — grows more pressing. Askell’s role suggests that shaping AI behaviour is as much a philosophical project as a technical one, and that the people writing the instructions may matter as much as those writing the code.