---
title: "Claude's constitution"
description: "Claude’s constitution describes how explicit principles can guide a language model’s behavior instead of relying on values implicitly determined through large-scale human feedback. The approach uses AI feedback in two training phases: the model first critiques and revises responses using the principles, then reinforcement learning selects the more harmless output using AI-generated evaluations. The post says Constitutional AI produced a Pareto improvement in tests, with the model responding more appropriately to adversarial inputs while remaining helpful and not evasive; harmlessness results came without human harmlessness data. The constitution draws on the UN Declaration of Human Rights, safety practices, other AI-lab principles, global platform guidelines, and attempts to include non-Western perspectives, while remaining an iterative, adjustable document."
---

# Claude's constitution

[Anthropic](https://yomu.fyi/company/anthropic) · Jan 21, 2026

**Type:** Explainer

## Summary

Claude’s constitution describes how explicit principles can guide a language model’s behavior instead of relying on values implicitly determined through large-scale human feedback. The approach uses AI feedback in two training phases: the model first critiques and revises responses using the principles, then reinforcement learning selects the more harmless output using AI-generated evaluations. The post says Constitutional AI produced a Pareto improvement in tests, with the model responding more appropriately to adversarial inputs while remaining helpful and not evasive; harmlessness results came without human harmlessness data. The constitution draws on the UN Declaration of Human Rights, safety practices, other AI-lab principles, global platform guidelines, and attempts to include non-Western perspectives, while remaining an iterative, adjustable document.

## Context

Human feedback implicitly shaped model behavior but could expose reviewers to disturbing outputs, scale poorly as responses became more numerous or complex, and require substantial time and resources.

## Approach / What changed

Constitutional AI uses a defined set of principles for AI feedback. During supervised training, the model critiques and revises its own responses; during reinforcement learning, AI-generated evaluations select the more harmless response. The constitution is also designed to be inspected and adjusted as model behavior changes.

## Takeaways

- Constitutional AI uses AI supervision rather than human harmlessness data to train responses to adversarial inputs.
- The model’s principles are intended to make its behavior easier to specify, inspect, understand, and adjust.
- The constitution combines sources including the UN Declaration of Human Rights, safety practices, other AI-lab principles, platform guidelines, and perspectives beyond Western cultures.

**Tags:** [AI](https://yomu.fyi/topic/ai), [Constitutional AI](https://yomu.fyi/topic/constitutional-ai), [Reinforcement Learning](https://yomu.fyi/topic/reinforcement-learning)

- Source: [Anthropic](https://www.anthropic.com/news/claudes-constitution)
- Source URL: https://www.anthropic.com/news/claudes-constitution
- Ingested by Yomu: 2026-09-09T20:02:29.621Z

[Read original post](https://www.anthropic.com/news/claudes-constitution)