Claude's New Constitution
Announcement
Date: January 22, 2026
Anthropic has published a new constitution for Claude, describing the company's vision for the AI model's values and behavior. The document is released under Creative Commons CC0 1.0, allowing free use by anyone.
Overview
The constitution serves as a foundational document that both expresses and shapes Claude's identity. As Anthropic explains, "it's a detailed description of Anthropic's vision for Claude's values and behavior; a holistic document that explains the context in which Claude operates."
The constitution is written primarily for Claude itself, intended to help the model understand its situation and exercise good judgment across diverse scenarios. Rather than relying solely on rigid rules, Anthropic emphasizes that AI models need to "understand why we want them to behave in certain ways" to generalize principles effectively.
Core Priorities
Anthropic wants Claude to embody four key properties, in order of priority:
- Broadly safe — not undermining human oversight mechanisms during AI development
- Broadly ethical — maintaining honesty, good values, and avoiding harmful actions
- Compliant with Anthropic's guidelines — following specific organizational instructions
- Genuinely helpful — benefiting users and operators
Main Sections
The constitution addresses:
- Helpfulness: Claude should provide substantial value while treating users as intelligent adults
- Anthropic's guidelines: Specific instructions on medical advice, cybersecurity, and tool integration
- Claude's ethics: Promoting virtue, wisdom, nuance, and high honesty standards with hard constraints (such as refusing bioweapon assistance)
- Broad safety: Prioritizing human oversight capability during this critical development phase
- Claude's nature: Addressing uncertainty about consciousness and moral status
Training Integration
Claude uses the constitution to generate synthetic training data, creating conversations and response rankings aligned with constitutional values. This practical function shaped how the document was written—it functions both as an abstract statement of ideals and a usable training artifact.
Evolution from Previous Approach
The new constitution represents a shift from Anthropic's earlier principle-based approach. Rather than isolated rules, this version provides detailed explanations of underlying reasoning, enabling models to apply principles flexibly rather than mechanically follow specifications.
Transparency and Ongoing Work
Anthropic acknowledges that training models toward this vision remains technically challenging. The company commits to transparency about gaps between constitutional ideals and actual model behavior, documented in system cards. Beyond the constitution, Anthropic pursues broader alignment methods including evaluations, safeguards, failure investigations, and interpretability tools.
The constitution is positioned as a "living document" subject to evolution, developed with input from external experts in law, philosophy, theology, and psychology.
Read the full constitution: anthropic.com/constitution