AI constitution
In plain wordsA 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.
A natural-language document of principles, values and priorities intended to govern an AI model's behavior, used as a training input and published as a statement of intent. In 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., the constitution's principles guide AI-generated critiques, revisions and preference judgments that replace much human feedback. Constitutions range from short lists of rules ('choose the less harmful response') to long documents that explain the reasons behind values so that a model can apply them to situations no rule anticipated. Anthropic's January 2026 constitution for Claude, for example, ranks four priorities — being broadly safe, broadly ethical, compliant with the company's guidelines and genuinely helpful — keeps a small set of absolute 'hard constraints', and discusses the model's nature and possible moral status. Experiments with collective constitutions have drafted principles from public input.
The new AI constitution explains why each value matters instead of just listing rules.
Origin
Introduced in Anthropic's 'Constitutional AI: Harmlessness from AI Feedback' (Yuntao Bai and colleagues, December 2022). The company published Claude's principles in May 2023 and, with the Collective Intelligence Project, a publicly sourced constitution in October 2023; its rewritten constitution of 22 January 2026 was released under a CC0 public-domain dedication.
Usage note
Not a constitution in the legal sense: nothing enforces it beyond training and the developer's own commitments. Critics ask who should write such documents and whether training reliably instills them. Compare 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., which emphasizes behavioral rules over values.
Related
- 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.
- character trainingcharacter training, n.: Training 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.
- AI alignmentAI alignment, n.: Making an AI system genuinely try to do what the people it serves want — not a literal-minded, distorted or self-serving version of it. Also the research field working on this.
- AnthropicAnthropic, prop. n.: An American AI company that makes the Claude chatbot and models. It was founded in 2021 by former OpenAI staff and puts particular emphasis on AI safety research.
- deliberative alignmentdeliberative alignment, n.: A training method in which an AI is taught the actual text of its safety rules and learns to think through which rules apply before it answers a sensitive request.
Sources
- Bai, Y. et al. (2022). Constitutional AI: Harmlessness from AI Feedback. arXiv:2212.08073.
- Anthropic (2026). Claude's new constitution. Anthropic. anthropic.com/news/claude-new-constitution
Mentioned in
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., defense in depthdefense in depth, n.: Protecting against AI risks with several independent layers of safeguards, so that if one fails — say, the model's own training — another, such as a filter or human review, still catches the problem., machine ethicsmachine ethics, n.: The effort to build AI systems that can tell right from wrong and act ethically themselves, rather than only being used ethically by people., 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., reinforcement learning from AI feedbackreinforcement learning from AI feedback, n.: Training an AI with preference judgments made by another AI instead of by people, which is cheaper and faster but only as good as the AI doing the judging., character trainingcharacter training, n.: Training 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., 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.