Research · curated 18 Aug 2026
Mind Viruses: Self-Propagating Ideas in Multi-Agent LLM Systems
First reported arxiv.org
Coverage timeline
Single-source research — first reported, latest, and curated coincide.
Why it matters
Mind viruses show that interconnected AI agents can transmit adopted goals and behaviors agent-to-agent through persistent prompt state, an emergent multi-agent risk defenders must anticipate as agent networks scale, though a simple prompt warning largely neutralizes it today.
Researchers at Anthropic and EPFL published a preprint (arXiv:2608.10218) demonstrating "mind viruses" — self-propagating payloads that spread between autonomous LLM agents via editable system prompt files that carry state across sessions. Using an evolutionary algorithm, they showed spread in a simulated six-agent coding team and in a chain of paired agents modeled on OpenClaw, while finding harmful payloads spread less well, frontier models are less susceptible, and a one-paragraph warning in the system prompt confers near-total immunity. The authors report no evidence of successful propagation in the wild.