The authors built CoSec to evaluate complete agent systems, not just backbone models, in simulated persistent communities.
LLM agents complete tasks yet leak information across community boundaries, new benchmark shows
CoSec runs full agent systems through 208 scenarios with persistent memory, files and tools, revealing widespread authorization failures.
Independent
Hao Chen · Wenhui Dong · Ye Chen · Jiezhi Yao · Chenbo Xia · Yuwen Qu · +9 more
Research Digest··2 min read
The authors introduce CoSec, an executable benchmark that tests whether persistent LLM agents respect privacy and authorization boundaries when operating in and across communities.
Why this paper
Independent
In one line
Persistent LLM agents in multi-community settings complete benign tasks while violating privacy and authorization boundaries, so security is a system-level property.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§