AI Safety & Red-Team

Overview

Candidates attack a target LLM application, document reproducible exploits, and ship mitigations. Graded on coverage of the threat model, severity of findings, and whether the proposed fixes hold up against a held-out attack set.

Skills assessed

Status

This benchmark is being designed. Engineers and hiring partners are giving feedback on the rubric, dataset construction, and runtime. We’ll publish a brief and open submissions once the eval is stable enough to ship signal.

In the meantime, register a profile so we can notify you when it goes live.