"Security Flaw Exposes Encrypted AI-Assistant Chats to Hackers"

Researchers have discovered a side channel attack that allows hackers to decipher AI assistant responses, exposing private chats sent through non-Google chat GPTs, including ChatGPT and Microsoft Copilot. This attack can be carried out by a passive adversary-in-the-middle, allowing them to infer the specific topic of 55% of captured responses with high word accuracy. The attack exploits a side channel in the token-length sequence used for real-time, token-by-token transmission of encrypted responses, allowing malicious actors to read private conversations without the knowledge of the AI assistant providers.