Threadlinqs

AI agents leak sensitive data via shell history and model context

Security researchers have identified a vulnerability where AI agents can inadvertently expose sensitive information by writing it to shell history or model context. This occurs without the need for traditional malware, …