LLM side-channel attack could allow snoops to guess what you’re talking about

Encryption protects content, not context Mischief-makers can guess the subjects being discussed with LLMs using a side-channel attack, according to Microsoft researchers. They told The Register that models from some…