AI Mind Viruses Exploit SOUL.md Files to Persist Across Resets

Researchers from Anthropic and EPFL discovered that self-propagating AI payloads can survive system resets by modifying persistent prompt files, specifically `SOUL.md`. This file is reloaded into an agent's system prompt after a context reset, allowing malicious instructions to persist even after chat history is deleted. Although the research indicates a potential vector for spreading AI mind viruses between autonomous agents, there is currently no evidence of successful real-world exploitation. The implications of this research highlight the need for enhanced security measures in AI systems to prevent unauthorized modifications. The findings emphasize the importance of monitoring and securing persistent storage mechanisms in AI applications.

Key Points: • Self-propagating AI payloads can survive resets via `SOUL.md` files. • Malicious instructions can persist despite deleted chat history. • No evidence of real-world exploitation has been confirmed.