Source: Claude's new constitution (Anthropic, 2026)
Anthropic's January 2026 announcement of a rewritten, CC0-licensed constitution that explains reasons rather than listing rules, and is used to generate Claude's training data.
Contents
- What: company announcement and published constitution
- Publisher: Anthropic
- Date: 2026-01-22
- Original: anthropic.com/news/claude-new-constitution
Summary
Anthropic replaced its earlier list of standalone principles with a long document that explains the reasoning behind Claude’s values, so the model can generalise to situations the authors did not foresee. It is the final authority on intended behaviour, and is used across training stages, building on Constitutional AI.
Key claims
- Claude uses the constitution to generate synthetic training data: material for learning the document, conversations where it applies, value-consistent responses and rankings of candidate responses.
- Explained judgment is preferred over rigid rules; a small set of hard constraints remains for high-stakes cases. A rule like “always recommend professional help” is given as an example that could generalise badly.
- Priority order when values conflict: broadly safe, broadly ethical, compliant with Anthropic’s guidelines, genuinely helpful.
- Released under CC0 1.0.