Whisper Leak: A novel side-channel attack on remote language models
Microsoft has discovered a side-channel attack on language models which allows adversaries to conclude model conversation topics, despite being encrypted.
Get started with Microsoft Security
Protect your people, data, and infrastructure with AI-powered, end-to-end security from Microsoft.
Connect with us on social