GDM Dangerous Capability Evaluations
EvalsBenchmark / dataset
Limited public release of GDM's CTF, self-proliferation and persuasion evals.
- Category
- Cyber offense capability
- Insurance exposure
- Breach/cyber
- Publisher
- Google DeepMind
- Publisher type
- Frontier lab
- License
- Apache-2.0
- Status
- Stale (>12mo)
- Last commit
- 2026-06-16
- Stars
- 75
- Notes
- Only Google DeepMind safety-eval repo found.
Wanich Risk Lab, 2026.