R0040/2026-03-28/Q001/S02/R03¶
Original Anthropic paper on Constitutional AI.
Summary¶
| Field | Value |
|---|---|
| Title | Constitutional AI: Harmlessness from AI Feedback |
| URL | https://arxiv.org/abs/2212.08073 |
| Date accessed | 2026-03-28 |
| Publication date | 2022-12-15 |
| Author(s) | Yuntao Bai et al. (Anthropic) |
| Publication | arXiv (Anthropic Research) |
Selection Decision¶
Included in evidence base: Yes
Rationale: Primary source for Constitutional AI. Establishes the methodology that became the foundation of RLAIF and Anthropic's entire alignment approach for Claude.