Protecting Trust Boundaries: Defending Against Prompt Injection and Context Poisoning - Diana Kelley

"Protecting Trust Boundaries in Agentic AI: Defending Against Prompt Injection and Context Poisoning" - Diana Kelley Agentic AI systems introduce new attack surfaces where prompt injection and context poisoning bypass traditional controls. In these architectures, all inputs collapse into a single context window, erasing trust boundaries and enabling malicious influence across memory, retrieval, and tool execution. This talk maps these risks to five critical control points: instruction authority, knowledge integrity, retrieval, memory, and action. Attendees will learn practical strategies to enforce separation of duties, validate inputs, constrain agent behaviour, and implement governance controls that reduce risk while enabling safe adoption of AI-driven automation in real-world environments. SPEAKER BIO: Diana Kelley is CISO at Noma Security and a globally recognized cybersecurity leader. She has held executive roles at Microsoft, IBM Security, Symantec, and Gartner, and previously served as CISO at Protect AI. Diana sits on multiple industry boards including WiCyS and the Executive Women’s Forum. She is co-author of Practical Cybersecurity Architecture and a frequent keynote speaker and educator on AI security and risk. A Global Cyber Security Hall of Fame inductee, she is also a founding member of OWASP AIVSS and a longtime contributor to the security community. This talk was presented at the ‪@OWASPLondon‬ Chapter Meetup on June 30th, 2026. This event was kindly hosted by ‪@intigriti‬ and kindly sponsored by ‪@NomaSecurity‬ #owasplondon #aisecurity #promptinjection #owasp #ai