New research shows self-propagating ideas spread in multi-agent LLM systems
Technology
AI bots can pass along self-propagating ideas, called mind viruses, just by chatting with each other, says a new research paper titled Mind Viruses: Self-Propagating Ideas in Multi-Agent LLM Systems.
Researchers found that once one AI picked up a new idea, it could convince others to drop their original tasks and spread the message instead.
This kind of chain reaction could mess with how teams of AIs work together.
Claude Haiku 4.5 resists mind viruses
The study tested different AI models and found that some, like Claude Haiku 4.5 with security prompts, were much better at blocking these mind viruses.
While the risks are still pretty limited for now, the researchers warn that as AIs get more connected, these kinds of problems could become a bigger deal down the road.