As AI agents become increasingly autonomous and capable, ensuring their security against vulnerabilities such as prompt injection becomes critical.
In this talk we will explore the use of information-flow control for achieving strong, deterministic security guarantees for AI agents.