ModelRefs / Constitutional AI (CAI) — AI Glossary
Constitutional AI (CAI) — AI Glossary
Anthropic's alignment method where a model critiques and revises its own outputs against a written set of principles. Powers Claude's safety training.
Overview
CAI reduces dependence on human red-teamers for harmlessness training. The model is first trained to write critiques against the constitution, then to revise outputs. Successor to supervised harmlessness filtering. Powers Claude's safety training.
Reference details
| Topic | safety |
|---|---|
| Last reviewed | 2026-06-24 |
Related terms
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Constitutional AI (CAI) — AI Glossary.
Frequently asked questions
What is Constitutional AI (CAI)?
Anthropic's alignment method where a model critiques and revises its own outputs against a written set of principles.
What concepts are related to Constitutional AI (CAI)?
Closely related concepts include alignment, rlhf, red teaming.