The authors modeled multi-agent LLM workflows using a communication graph (message delivery), an authorization hypergraph (qualified signer coalitions), and an execution graph (tool access and enforcement).
Multi-agent LLM safety requires separate evaluations of communication, authorization, and execution
A trust-boundary framework with controlled experiments reveals that message delivery, signing authority, and tool execution have distinct safety consequences.
Industry
Junru Zhu · Xin Chen · Hang Lyu · Jiayue Meng · Muzi Li · Yifan Chen · +3 more
Research Digest··2 min read
Thread:Agent Security & Attacks
The authors formalize three trust boundaries in multi-agent LLM systems—communication routes, qualified signer coalitions, and execution mediation—and introduce the authority cut metric to characterize minimum compromised signers for a policy-valid false claim.
§