Self-Replicating AI Malware is Here😱 #ComPromptMized

This blog post summarizes research showing a proof-of-concept zero-click generative-AI worm that leverages adversarial self-replicating prompts and retrieval-augmented generation poisoning to spread, exfiltrate data, send spam/propaganda, and potentially enable severe payloads; the work is presented as a warning and calls for mitigations such as output sanitization and human-in-the-loop confirmations, while noting the threat has not yet been observed in the wild but could emerge within years.