Safety, Ethics & Governance · advanced · concept 289 of 311 · Updated Nov 4, 2026
Constitutional AI & RLAIF
Alignment where the feedback comes from AI guided by an explicit list of principles, a constitution, instead of thousands of human raters. The model critiques and revises its own outputs against the principles, making the values legible and editable rather than buried in rater preferences.
Top AI contributors on this
- Geoffrey HintonUniversity of Toronto (emeritus)
- Yoshua BengioLawZero / Université de Montréal
- Ilya SutskeverSafe Superintelligence Inc. (SSI)
- Dario AmodeiAnthropic
- Stuart RussellUC Berkeley
- John SchulmanThinking Machines Lab
Key terms
ConstitutionSelf-critiqueRLAIFPrinciple-based feedbackHarmlessness
Learn these first
Where you meet it in the real world
How Claude is trained, scalable alignment pipelines, auditable AI values
Videos
▶ Reinforcement Learning from Human Feedback (RLHF) Explained ↗
IBM Technology · YouTube
Guides and articles
[2212.08073] Constitutional AI: Harmlessness from AI Feedback ↗
arXiv (original paper)
This unlocks