The authors built TrustProbe to test how content from installable agent skills propagates through an agent framework.
Installable agent skills can inherit user authority without adequate checks
TrustProbe found skill-controlled content reaching sensitive operations across 11 open-source agents, including 15 vulnerabilities that researchers successfully weaponized.
Research Lab
Yan Wang · Zhihao Zhang · Ke Chen · Kai Chen · Yaqin Zhang · Duohe Ma · +2 more
Institute of Information Engineering, Chinese Academy of Sciences · Worcester Polytechnic Institute · School of Cyberspace Security, University of Chinese Academy of Sciences
Research Digest··3 min read
Wang et al.
Why this paper
From Institute of Information Engineering, Chinese Academy of Sciences and 2 others
In one line
Skill-based LLM agents allow malicious skills to reach security-sensitive operations due to systematic trust failures.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·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.
§