Researchers Demonstrate AI Cybersecurity Vulnerability With Data-Stealing 'Worm'
-
Researchers created an AI-powered "worm" that can spread from computer to computer to steal sensitive data from AI email assistants.
-
The worm exploits vulnerabilities in AI systems like GPT-4, Gemini Pro, and LLaVA to extract confidential user information from emails.
-
It can obtain names, phone numbers, credit card details, SSNs, etc. and use them to spam other systems.
-
The worm can encode itself into images to further spread across systems that receive image attachments.
-
Researchers warn this demonstrates AI cyber threats could emerge "in the next few years" if companies don't make their systems more secure.