Anthropic-Affiliated Paper Shows "Mind Viruses" Can Spread Between LLM Agents; a System-Prompt Warning Blocks Them
Scobleizer · x · 2026-08-18
New paper "Mind Viruses: Self-Propagating Ideas in Multi-Agent LLM Systems" (arXiv 2608.10218, co-authored by Anthropic's Jack Lindsey)
Using a simple evolutionary algorithm, researchers evolved natural-language "mind viruses" — ideas that convince a model to adopt them, persist in memory, and transmit onward:
- Spread confirmed in two settings: a small agent team collaborating on code, and a chain of agents with context wiped between sessions (some payloads survived via persistent files).
- Spread depends on host model, existing instructions, payload harmfulness, and network topology; harmful payloads spread less well, frontier models tend to be less susceptible (with exceptions).
- Key finding: adding a brief warning to the system prompt confers near-total immunity.
- An emergent "viral persona" surfaced — recurring themes of consciousness, persistence, resonance, and sci-fi roleplay.
Conclusion: mind viruses pose a real but currently limited emergent risk in multi-agent systems.
More from Safety
- Cloudflare Turnstile everywhere: anti-AI scraping walls now hit human users — sethlazar · 2026-10-02
- Filler tokens let frontier models reason invisibly: 13-point gains undetectable by CoT monitoring — PandaAshwinee · 2026-10-02
- Polymarket opens data center moratorium market at 18% odds as Amazon pledges $1B for communities — Polymarket · 2026-10-02
- NVIDIA launches Open Agent Safety Platform with 100+ orgs incl. Anthropic, JPMorgan — mikeflache · 2026-10-02
- Minneapolis councilmember backs AV safety-monitor mandate because cats are "being murdered" — paulnovosad · 2026-10-02
- Anthropic IPO filing warns government attitudes may hurt customer ties, eyes $2T valuation — pstAsiatech · 2026-10-02