Constitutional AI
Also known as: CAI · constitution AI
Constitutional AI (CAI) is a method (popularized by Anthropic) where a model is trained to critique and revise its own outputs against an explicit set of principles — a “constitution.”
- The model judges itself against written rules.
- Reduces the need for human labels for safety.
- A form of principle-based alignment.
The idea's power: CAI makes safety *inspectable* — the principles are in the open, so you can see what the model was told to optimize for. That transparency is the point.