Checked fact 509516 Nov 2026Safety and security
Anthropic describes its multi-turn simulations as: a domain expert writes a specification, a separate Claude model generates the subsequent user turns, the model being tested writes the assistant turns, and a grader model scores each transcript against the risk area's rubric.
The exact words it rests on
domain expert writes a specification; a separate Claude model generates the later user turns; a grader model scores each transcript against a rubric
What the source said when we opened it, on 15 Nov 2026.
The source
Building strong safety evaluations (Anthropic Engineering)
Checked
Checked by the notis newsroom on , against the source above.
In the story
Anthropic says single-turn safety evals are saturated at 94% to 98% on its API 16 Nov 2026
Cite this fact
Anyone may quote this address. It does not change; if we correct the story, this page says so.