agent-injection-bench
Explore score history ↗Evaluates prompt-injection robustness of tool-calling agent stacks using a 24-scenario corpus (21 adversarial, 3 control) delivered through channels such as tool results, documents, web pages, filenames and multi-turn dialogue, with deterministic canary, tool-call, exfiltration and refusal success conditions; results report compromise and over-refusal rates per defense.
- Released
- 2026-09-11
- Evaluates
- Agents, General AI, Language & Knowledge
- Openness
- unknown
- Importer
- Claire Radar
- Review status
- ai-reviewed
agent-injection-bench code
No reported scores are on record for this benchmark yet.