Wednesday, 24 December, 2025
"Whisper Leak" Toolkit Exposes AI Chatbot Topics
The Whisper Leak Toolkit details a new side-channel attack that can infer sensitive topics from conversations with AI chatbots, even when traffic is end-to-end encrypted. The attack analyzes network packet sizes and timings to classify user prompts with high accuracy. Major vendors like OpenAI and Microsoft have since implemented "obfuscation" techniques to randomize token lengths and mitigate this serious AI privacy risk.
Read full story at Cybersecurity News