Mind viruses spread agent-to-agent and survive context wipes, but one prompt line stops them.

Jack Lindsey, who leads Anthropic's Model Psych team, posted a deliberately deflationary summary of his own paper (arXiv 2608.10218) while a 336K-view thread framed it as science fiction. The evolved ideas persisted through 20 transmission rounds, mutating to become more infective, surviving context wipes via long-term memory files. The defense: a single warning line in the system prompt stopped every evolved attack past one hop on Claude Haiku 4.5 across 150+ attempts. An author downplaying his own scary result is rarer than it should be.