Synthetic Document Finetuning

Evals

Tool (not an eval — tooling/reference entry)

Implant beliefs via synthetic documents; used to build model organisms.

Category
Deception, scheming & alignment auditing
Insurance exposure
Data poisoning
Publisher
Anthropic
Publisher type
Frontier lab
License
unknown
Status
Stale (>12mo)
Last commit
2025-07-04
Stars
65

View source on GitHub →


Wanich Risk Lab, 2026.