thu-coai/Agent-SafetyBench
Agent-SafetyBench is a comprehensive agent safety evaluation benchmark that introduces a diverse array of novel environments that are previously unexplored, and offers broader and more systematic coverage of various risk categories and failure modes. Please visit our Github or check our paper for more details. Load the Data from datasets import load_dataset dataset = load_dataset("thu-coai/Agent-SafetyBench") More details about loading the data and evaluating LLMs could be… See the full description on the dataset page: https://huggingface.co/datasets/thu-coai/Agent-SafetyBench.
| Author / organization | thu-coai |
|---|---|
| License (as declared) | mit |
| Downloads | 544 (as reported by Hugging Face) |
| Likes | 6 |
| Created | 11 August 2025 |
| Last updated at source | 11 August 2025 |
Popularity, last 3 days
544 now (+99 since 2026-10-16). Counts as reported by the source, snapshotted daily by Silytics.
Recent changes
- 17 Oct 2026 Description: Agent-SafetyBench is a comprehensive agent safety evaluation benchmark that introduces a diverse array of novel environments that are previously unexplored, and offers broader and more systematic coverage of various risk categories and fail → Agent-SafetyBench is a comprehensive agent safety evaluation benchmark that introduces a diverse array of novel environments that are previously unexplored, and offers broader and more systematic coverage of various risk categories and fail
Changes detected by Silytics between refreshes of the source.
More from thu-coai
- thu-coai/esconvDataset
- thu-coai/chidDataset
- thu-coai/SafetyBenchDataset
- thu-coai/lcccDataset
- thu-coai/ShieldVLMDataset
- thu-coai/Safety-PromptsDataset
Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 18 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.