AgentDrift Benchmark
Explore score history ↗Step-labeled corpus of 12,536 synthetic LLM agent tool-call trajectories in five domains, labeled per step for benign, injection point, hijacked, or failed injection, supporting detection, localization, and attempt-vs-success evaluation.
- Released
- 2026-09-07
- Institutions
- Northern Arizona University
- Evaluates
- Agents, General AI, Language & Knowledge
- Openness
- unknown
- Importer
- Claire Radar
- Review status
- ai-reviewed
AgentDrift paper and code
No reported scores are on record for this benchmark yet.