Training & post-training
Constitutional AI
Training and steering models using an explicit set of written principles.
Definition
Constitutional AI asks models to critique and revise outputs against a defined constitution and can use those judgements as training feedback.
Why it matters
The constitution makes some policy choices visible, but banks still need controls aligned to their own laws, conduct duties and risk appetite.
Related concepts
- Reinforcement learning from AI feedback
Using AI-generated preferences or critiques to guide reinforcement learning.
- AI alignment
Making AI behaviour consistent with intended goals and constraints.
- Guardrail
A technical or procedural control that constrains system behaviour.