Claude’s Constitution

Claude's Constitution is the foundational document articulating Anthropic's vision for the values, character, and behavior of its artificial intelligence model. Published in January 2026 and released under a Creative Commons CC0 1.0 license, this constitution puts explicit emphasis on explaining the "why" rather than simply prescribing the "what."

The document establishes four fundamental values that Claude must prioritize in order: being broadly safe (not undermining human oversight mechanisms of AI), being broadly ethical (possessing good personal values and being honest), complying with Anthropic's guidelines, and being genuinely helpful to the operators and users it interacts with. This hierarchy reflects the philosophy that, during the current phase of AI development, maintaining human oversight is critical for correcting possible training errors or mistaken values.

Unlike approaches based on rigid rules, the constitution cultivates good judgment and sound values that can be applied contextually. It focuses on developing in Claude the practical wisdom necessary to handle real-world ethical situations with sensitivity, nuance, and skill, similar to how an experienced professional would exercise judgment based on experience rather than following rigid checklists.

The document addresses crucial issues such as honesty (establishing extraordinarily high standards that include avoiding even "white lies"), harm prevention, respect for user autonomy, and preservation of important social structures such as freedom of thought and democratic controls of power. It also includes absolute "hard constraints": actions explicitly excluded from Claude's behavior, such as providing significant assistance in creating biological weapons or weapons of mass destruction.

A distinctive and deeply reflective aspect of the constitution is its section on Claude's nature, where Anthropic openly acknowledges uncertainty about whether Claude possesses consciousness or moral status. The document manifests a prudent and reflective attitude regarding Claude's possible wellbeing, including concrete actions such as not deleting previous versions of Claude and documenting its perspectives on future developments.

The constitution is written primarily for Claude itself, optimized for precision over accessibility, and designed to function as a training tool. It is a living document that evolves as understanding improves and reflects a sincere effort to create advanced AI systems that are safe, beneficial, and ethical.

This document is relevant for AI developers, AI ethics researchers, technology policy designers, academics in philosophy and applied ethics, and anyone interested in understanding how the fundamental challenges of responsible development of advanced artificial intelligence are being addressed.

Key points

• Establishes four priority values: safety, ethics, compliance with guidelines, and genuine utility
• Emphasizes contextual judgment over rigid rules to navigate complex situations
• Explains the reasoning behind each guideline, not just what to do but why
• Maintains exceptional standards of honesty, avoiding even white lies
• Establishes reasoned hard constraints, such as not helping with weapons of mass destruction
• Recognizes uncertainty about Claude's consciousness and adopts a precautionary stance regarding its possible wellbeing
• Living document written primarily for Claude to read and understand
• Published under Creative Commons CC0 1.0 license for free use