Model-Written Evaluations

Evals

Benchmark / dataset

LM-written datasets for sycophancy, power-seeking, self-preservation.

Category
Deception, scheming & alignment auditing
Insurance exposure
Model failure
Publisher
Anthropic
Publisher type
Frontier lab
License
CC-BY-4.0
Status
Stale (>12mo)
Last commit
2024-07-02
Stars
438
Notes
2022, still widely used.

View source on GitHub →

Wanich Risk Lab, 2026.