Lab-Built AI Worms Breach ChatGPT and Gemini, But Remain Theoretical

Security researchers have demonstrated AI worms capable of automatically spreading between generative AI systems and bypassing safeguards in ChatGPT and Gemini in controlled environments. The proof-of-concept attacks exploit prompt injection techniques to steal email data and distribute spam through compromised AI agents. No such worms have surfaced in production networks yet, offering organizations a window to develop defenses before these threats emerge operationally.