Microsoft Reveals 'Whisper Leak' Vulnerability in AI Chatbots

Microsoft has disclosed a side-channel attack named 'Whisper Leak' that can expose the topics of conversations with AI chatbots, even when using end-to-end encryption. This vulnerability allows eavesdroppers, including ISPs and government entities, to infer sensitive information from network traffic patterns without accessing the actual text of conversations. Major AI providers have implemented defenses to mitigate this risk.