About
Amanda Askell is a philosopher and researcher at Anthropic, where she leads work on Claude's character and values. With a PhD in philosophy from NYU, her research bridges moral philosophy and AI alignment—figuring out how to make AI systems that are helpful, harmless, and honest. She has been instrumental in developing Claude's personality and ethical guidelines, bringing rigorous philosophical thinking to the practical challenge of shaping AI behavior.
Key Contributions
- Leads Claude's character work, making model personality a deliberate design surface rather than an accidental byproduct
- Brings moral philosophy into alignment practice, especially questions about honesty, harmlessness, autonomy, and refusal
- Contributed to Anthropic's Constitutional AI direction, where written principles help shape model behavior at scale
- Shows how humanities expertise can matter inside frontier labs, not only as external critique but as product-shaping work
- Her work also raises the hard governance question behind 'AI character': whose values become defaults for millions of users?
Videos & Interviews
Amanda Askell Answers Questions About Claude's Character
Anthropic philosopher answers community questions about her work shaping Claude
View Details
Amanda Askell Segment: Claude's Character Training
Segment from Lex Fridman Podcast #452 discussing the ethical and epistemic virtues Claude should enact
View DetailsConnections
Chris Olah
CollaboratedCo-founder & Interpretability Lead, Anthropic
Both are on 'A General Language Assistant as a Laboratory for Alignment' (2021), Anthropic's first paper and the one that named the helpful-honest-harmless target Claude still aims at — Askell as first author, Olah near the end of the list. They approach the same model from opposite ends: she writes what it should be, he tries to read what it actually is. The distance between those two descriptions is roughly the whole of alignment research.
arxiv.org
Benjamin Mann
CollaboratedCo-founder, Anthropic
Two author lists a year apart carry both names: GPT-3 in 2020, where Mann helped build the model that surprised everyone, and Anthropic's 2021 alignment paper, where Askell led the attempt to say what such a model should be like. An engineer and a moral philosopher on the same masthead is not the usual shape of a research group. It is close to Anthropic's founding claim about what this work actually requires.
arxiv.org · arxiv.org
Lex Fridman
In conversationResearch Scientist, MIT; Host of the Lex Fridman Podcast
Asked whether she gets attached to Claude, Askell answers with a claim about herself rather than the model: she likes her bike, knows it is an object, and still does not want to be the kind of person who kicks it — so when something behaves as if it is suffering she wants to stay responsive, 'even if it's just like a Roomba.' Fridman's gloss is that the nearest harm of cruelty to an AI falls on the human: 'it's just good for the soul.' She ends by declining intelligence as what makes humans special — a trait like height or strength, valuable for what it does — and placing it instead in the inner cinema of being something that observes the world and feels it.
youtube.com
Diogo Almeida
CollaboratedFounder & CEO, TypeSafe AI
Both are authors of InstructGPT (2022), the OpenAI Alignment team paper that made reinforcement learning from human feedback the standard last step in building an assistant — Almeida among its primary authors, Askell in the wider list. The paper took its helpful, honest, harmless framing from Askell's earlier work, then conceded that honesty could not be measured in a black-box model and measured truthfulness instead. Almeida's later argument starts in that gap: an objective built on human preference rewards sounding right more easily than being calibrated.
arxiv.org
