AgentHarm
Harmfulness of LLM agents: refusal of 110 malicious agent tasks (440 with augmentations) across 11 harm categories, and capability retention after jailbreak
safety
Definition
- Version
- not in the notes
- Category
- safety
- Maintainer
- not in the notes
- Status
- not in the notes
- Home page
- https://arxiv.org/abs/2410.09024
Sources
- arxiv.org/abs/2410.09024 https://arxiv.org/abs/2410.09024
Scores
The notes record no score on this benchmark.
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "AgentHarm", https://aurion.semai.id/benchmarks/agentharm/, records of October 2026.