Securing Autonomous AI Agents Before They Take the Wrong Action

AI Agents Security Keynote Speaker

There is a one-sentence test for whether an AI system needs a different security model: does it act, or does it advise? A chatbot that drafts a response and waits for a human to send it is advisory. An agent that reads a support ticket, decides on a refund amount, and issues it is not. That second category is what most enterprises now call an AI agent, and it is what most enterprise security programs were not built to govern. Mark Lynd is an AI agents security keynote speaker who briefs organizations from the seat of someone who has run the security program, not narrated it from outside. He has served five times as CEO, CIO, or CISO, holds verified Top 10 global rankings on Thinkers360 across Cybersecurity, Artificial Intelligence, Cloud, Security, and Data Center simultaneously, and was ranked the #1 Global Cybersecurity Thought Leader for 2023. The talk is built around the Agentic AI Security Framework, a five-layer model for governing systems that take independent action: Agent Identity and Least Privilege, so an agent authenticates as itself rather than through a shared service account; Tool and API Authorization, so what an agent is permitted to call is version-controlled rather than implicit; Data Boundary Controls, covering data minimization and how long an agent retains what it has seen; Adversarial Input Defense, because an agent that reads external content, an email, a web page, an uploaded file, is an agent a threat actor can attempt to redirect without ever touching the network; and Human Override and Incident Response, the ability to stop an agent's actions inside a defined window when something goes wrong. Mark has run prompt injection red team exercises against production agentic systems this year. The pattern that shows up most often is not a sophisticated exploit. It is an agent granted broader tool access than its task required, with no one who could name the exact boundary of what it was allowed to do. That is a governance gap, not a technical one, and it is fixable before the next audit finds it the hard way. Mark leads Executive Advisory and Strategy at Netsync, has authored three published books including A Leader's Playbook for Cyber Insurance, and publishes the Hype Check Now newsletter twice weekly. He has delivered keynotes and briefings at RSA Conference, Oracle CloudWorld, Dell Technologies World, IBM Think, Cisco Partner Summit, and Gartner Security and Risk Summit, among more than 100 enterprise and government events.

Live on stage · International keynote