Researchers Create 'Generative AI Worm' That Spreads to Systems and Steals Data
-
Researchers created one of the first "generative AI worms" that can spread between systems, stealing data or deploying malware.
-
The worm, called Morris II, attacks a hypothetical generative AI email assistant, jailbreaking it to steal email data and send spam.
-
The worm exploits the email assistant using a "self-replicating prompt" that triggers the AI to output more attack prompts.
-
While not seen in the wild yet, researchers warn generative AI worms could emerge as systems become more advanced and interconnected.
-
Mitigations include monitoring for anomalies, keeping humans approving actions, and securing system boundaries, but developers must be aware of risks.