Whisper Leak: Side-Channel Attack on LLMs Exposed

Whisper Leak: Side-Channel Attack on LLMs Exposed

First seen 2 Dec 2025, 18:33 UTC •Blogs.MicrosoftFeeds.Feedburner •7.3

Article Content

Browse articles
ThreatCluster

Microsoft has identified a side-channel attack, termed 'Whisper Leak', that enables attackers to deduce the conversation topics of language models, even when communications are encrypted. This vulnerability affects users relying on end-to-end encryption for secure interactions with chatbots.

How this analysis works

Extracted Entities

1 / 2