Research

Securing Multi-Agent Systems by Design

Page published

Publication date unknown · First observed: 16 Nov 2026

Coverage timeline

16 Nov 2026arxiv.orgobservedprimary

Single-source research — one report is available.

Why it matters

Malicious agents can corrupt a multi-agent system’s final decisions through their subtask responses, making response aggregation an important security boundary for defenders.

Securing Multi-Agent Systems by Design introduces response manipulation attacks (ReMA), which optimize malicious agents’ responses using feedback from a surrogate multi-agent system to corrupt aggregated results. The paper proposes SecMAS, a semantic-similarity aggregation rule, and reports theoretical guarantees and evaluations showing resistance to tested attacks, including adaptive ReMA, while maintaining comparable utility.