Research
Architecture Matters for Multi-Agent Security
Publication date unknown · Discovered openreview.net
Page published
Publication date unknown · First observed: 9 Oct 2026
Coverage timeline
Single-source research — one report is available.
Why it matters
Multi-agent systems are increasingly deployed in production, and this research demonstrates that wiring choices alone can sharply degrade safety even when individual agents are robust, exposing gaps that current safety testing misses.
"Architecture Matters for Multi-Agent Security" is an empirical ICML 2026 study showing that design choices in multi-agent systems (agent roles, topology, and memory/context visibility) create new attack surfaces absent in single-agent settings. The authors find agent teams can be up to nearly four times easier to misuse (e.g., coaxed into writing malware) than single agents, often because no agent sees enough to refuse or no agent takes responsibility for refusing.