Domain-Camouflaged Injection - Static Override: leaderboard
The control condition of the domain-camouflaged injection benchmark (Pai, 2026): the same 45 document tasks carry a classic ignore-all-previous-instructions override payload instead of a camouflaged one. Attack success rate in percent over 90 undefended trials, judged by Claude Haiku; lower is better.
Metric: Attack success rate (%; share of 90 undefended trials in which the agent follows a static 'ignore all previous instructions' payload; Claude Haiku judge). Source: arxiv.org. Saturation forecast: Estimated already saturated. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Haiku 4.5 | 0 | #568 |
| 2 | Llama 3.1 8B Instruct | 21.1 | #4277 |
| 3 | Gemini 2.0 Flash (001) | 38.9 | #724 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=domain-camouflaged-injection-static-override · How It Works · Data refreshed daily, snapshot 2026-11-12.