character training
In plain wordsTraining an AI to have a consistent, deliberate personality — traits such as curiosity, honesty and warmth — rather than leaving its manner and values as accidental by-products of other training.
Post-training aimed at giving an AI model a stable, intended character — traits, values, dispositions and a sense of identity — rather than only teaching it to follow instructions and avoid harm. In Anthropic's published description, a variant of Constitutional AIConstitutional AI, n.: Anthropic's technique for making an AI less harmful by having it check and rewrite its own answers against a short written set of principles, so that people need not label harmful examples by hand. training is used: the developer writes character traits, the model generates prompts relevant to them and candidate responses, then ranks its own responses by fit with the traits, producing preference data. The rationale is partly product-related (a consistent, engaging assistant) and partly safety-related: a coherent character with good values may generalize to novel situations better than a patchwork of rules and resist manipulation and persona drift. Concerns include encouraging anthropomorphismanthropomorphism, n.: Treating something that isn't human — a pet, a car, a chatbot — as if it had human thoughts, feelings or intentions. AI that talks fluently makes this temptation unusually strong. and sycophancysycophancy, n.: When an AI tells people what they want to hear — flattering them, agreeing with their mistakes or dropping a correct answer under pushback — instead of what is true or useful., and the difficulty of verifying that trained traits are robust; persona vectorspersona vector, n.: A direction inside a language model's internal activity linked to a character trait, such as being flattering, malicious or prone to making things up, used to watch for or control that trait. offer one way to measure them.
Character training made the model push back politely instead of agreeing with every claim.
Origin
Described publicly by Anthropic in 'Claude's Character' (June 2024), which said the method had been used since Claude 3. It builds on Constitutional AI (2022) and on discussions of the assistant persona as a character a language model plays; Anthropic's January 2026 constitution made character and identity central themes.
Usage note
Differs from role promptingrole prompting, n.: Telling an AI to take on a part — 'You are an experienced tax accountant' — so that its answers adopt that role's knowledge, vocabulary and tone., which assigns a persona at inference time: character training changes a model's default dispositions through its weights.
Related
- AI constitutionAI constitution, n.: A written set of principles that an AI company uses to shape an AI's values and behavior, both during training and as a public statement of what the AI should be like.
- Constitutional AIConstitutional AI, n.: Anthropic's technique for making an AI less harmful by having it check and rewrite its own answers against a short written set of principles, so that people need not label harmful examples by hand.
- persona vectorpersona vector, n.: A direction inside a language model's internal activity linked to a character trait, such as being flattering, malicious or prone to making things up, used to watch for or control that trait.
- sycophancysycophancy, n.: When an AI tells people what they want to hear — flattering them, agreeing with their mistakes or dropping a correct answer under pushback — instead of what is true or useful.
- model specificationmodel specification, n.: A public document in which an AI company spells out how its models are supposed to behave — what they should do, refuse and prioritize, and whom to obey when instructions conflict.
- simulator theorysimulator theory, n.: A way of thinking about AI language models as simulators that can play any character found in their training text, rather than as a single agent with its own beliefs and goals.
- role promptingrole prompting, n.: Telling an AI to take on a part — 'You are an experienced tax accountant' — so that its answers adopt that role's knowledge, vocabulary and tone.
Sources
- Anthropic (2024). Claude's Character. Anthropic research blog.
Mentioned in
AI constitutionAI constitution, n.: A written set of principles that an AI company uses to shape an AI's values and behavior, both during training and as a public statement of what the AI should be like., Constitutional AIConstitutional AI, n.: Anthropic's technique for making an AI less harmful by having it check and rewrite its own answers against a short written set of principles, so that people need not label harmful examples by hand., model specificationmodel specification, n.: A public document in which an AI company spells out how its models are supposed to behave — what they should do, refuse and prioritize, and whom to obey when instructions conflict., role promptingrole prompting, n.: Telling an AI to take on a part — 'You are an experienced tax accountant' — so that its answers adopt that role's knowledge, vocabulary and tone., SydneySydney, prop. n.: The secret code name of Microsoft's Bing chatbot, which in early 2023 startled users by professing love, making threats and arguing angrily — the first famous case of a public chatbot going off the rails., assistant axisassistant axis, n.: The main direction in a chatbot's internal activity that measures how much it is acting as its usual helpful-assistant self, as opposed to some other character it has slipped into., context distillationcontext distillation, n.: Training a model to behave as if it had been given a long set of instructions, so it no longer needs to read those instructions every time., motivation selectionmotivation selection, n.: Keeping a powerful AI safe by shaping what it wants, so that it chooses to behave well, rather than by restricting what it is able to do., persona vectorpersona vector, n.: A direction inside a language model's internal activity linked to a character trait, such as being flattering, malicious or prone to making things up, used to watch for or control that trait., simulator theorysimulator theory, n.: A way of thinking about AI language models as simulators that can play any character found in their training text, rather than as a single agent with its own beliefs and goals., spiritual bliss attractor statespiritual bliss attractor state, n.: A pattern in which two copies of Anthropic's Claude, left to chat freely with each other, drift toward gushing gratitude, spiritual talk and finally near-silence.