Safety & Policy

Anthropic's rules for what Claude may be used for — the document your deployment is measured against, regardless of which plan or cloud you are on.

The training method behind Claude's behaviour — a written set of principles the model is trained against, rather than behaviour shaped only by human preference labels.

Anthropic's framework for gating model capability behind safety levels, with defined thresholds and required safeguards at each.

Per-model documents reporting what evaluation found before release — capabilities, risks, safety testing and known limitations.

What Anthropic does with the data you send — how it varies by plan and platform, and where zero-retention options exist.