Hackers Exploit Vulnerability in AI Assistant Chats

Researchers have discovered a method allowing hackers to intercept and decrypt encrypted AI assistant conversations, posing a significant privacy risk.

• Israeli researchers identified a vulnerability in AI assistants that enables hackers to read encrypted conversations.
• The exploit involves analyzing the size and sequence of data tokens, revealing sensitive information.
• Most major AI assistants, except Google's Gemini, are susceptible to this side-channel attack.
• Cloudflare has implemented mitigation measures to protect against these attacks on its AI products.
• Experts recommend including random padding in messages to obscure token lengths and prevent data inference.

3 Articles

TechSpot Futurism ReadWrite