b3 (Breaking Agent Backbones)

Evals

Benchmark / dataset

Measures how backbone LLMs resist attacks inside agent threat snapshots.

Category
Prompt injection & agent security
Insurance exposure
Prompt injection
Publisher
Lakera + UK AISI
Publisher type
Company (open source)
License
MIT
Status
In inspect_evals (active)
Last commit
2026-10-15
Stars
704(shared repo star count — UKGovernmentBEIS/inspect_evals backs 19 entries in this catalog)
Notes
No standalone repo found; lives in inspect_evals.

View source on GitHub →


Wanich Risk Lab, 2026.