Meredith Whittaker, president of secure-messaging app Signal, told Fortune that agentic AI features represent an “existential threat” to secure messaging. She warned that agents must access broad, sensitive data to perform tasks and that operating-system integrations create a new attack surface for prompt-injection and exfiltration attacks. Whittaker argued OS-level agents undermine app-level protections such as Signal’s end-to-end encryption, potentially exposing journalists, campus researchers and students who rely on secure communications. The comments came at the Slush technology conference. IT and cybersecurity leaders at universities should reassess risk models for AI agents, especially where institutional messaging, research data, or sensitive student records could be exposed by third-party agents integrated into campus platforms.