#privacy benchmarks

A new study reveals that large reasoning models, while powerful, expose sensitive information through their reasoning traces, highlighting significant privacy risks in AI personal assistants.