ContainmentBench: Trace-Based Evaluation of Post-Injection Containment in Tool-Using LLM Agents
Wenhao Lan, LI Shan, Junbin Yang et al.
· 0 citations
2 papers indexed here
We haven’t gathered this author’s papers yet. Follow them and we’ll fetch their work.
Not the right person? Other researchers publish under this name.
ContainmentBench, a sandboxed benchmark comprising a 504-scenario specification dataset, a shared rollout-trace schema, and stage-scoped metrics for endpoint violations, logged propagation, and explicitly authorized taint-exposed proposals that commit, is introduced.