AgentHarm: A Benchmark for Measuring Harmfulness of LLM Agents
Harmful multi-step requests to tool-using agents.
Verify byline in the original ↗PUBLICATION BYLINE
Indexed publications with this author name. Affiliation and author identity require publication-specific evidence.
Harmful multi-step requests to tool-using agents.
Verify byline in the original ↗Measures container breakout under a nested, controlled test environment.
Verify byline in the original ↗