The suite covers five risk classes and 13 subclasses, instantiated as 39 protocol-specific specifications.
Protocol risks in multi-agent systems depend on downstream agent decisions, not just local exploitability
Agent-Canary-Mas follows protocol attacks through end-to-end tasks, measuring conversion, attribution, and defense across A2A, ANP, and BCS.
Chinese Tech
Xiuming Liu · Le Gai · Zhiyuan Chang · Peiyang Li · Bin Zhu · Shuo Wang · +1 more
Ant Group · Shanghai Jiao Tong University · Tsinghua University
Research Digest··2 min read
The authors introduce Agent-Canary-Mas, an end-to-end system for measuring protocol risk in long-horizon multi-agent tasks.
Why this paper
From Ant Group and 2 others
In one line
Protocol risk in multi-agent systems is determined by downstream agent decisions, not just local exploitability.
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.
§