A Conversation with Duncan Greatwood

Check out Xage here: https://ul.live/xage-webinar_yt I just sat down with Duncan Greatwood of Xage to chat about how we can actually secure AI agents and stop attacks like prompt injections from bypassing system controls. We break down their zero-trust "fabric" approach, which basically wraps your AI in armor so you can safely innovate and deploy agents without worrying about them accidentally nuking your company's data. What we talk about: Defeating Prompt Injections: How to protect organizations from rogue AI agents and prompt injection attacks by controlling the *actual* actions they take under the hood, rather than just relying on easily bypassed guardrails. Zero-Trust Architecture for AI: The mechanics of Xage's solution, specifically how the "Resource Gateway" locks down valuable company data and the "Agent Sentry" encapsulates local agents to stop them from executing malicious scripts or leaking sensitive files. The Multi-Hop Problem: Solving the complex chain of interactions between users, agents, and LLMs by enforcing robust, role-based access controls and identity management across the entire sequence. Shadow AI & Data Integrity: How to discover unmanaged "shadow AI" on the network and ensure the integrity of the data supply chain so that autonomous systems—like those used in critical infrastructure or aircraft maintenance—are acting on real, untampered data. Killing the "Department of No": Why simply blocking AI adoption doesn't work for businesses trying to innovate, and how gaining real-time visibility allows companies to safely deploy AI, maintain audit compliance, and even reign in runaway token consumption costs. 00:00 - Introduction 00:21 - The core problem being solved 01:23 - How the system actually works 03:09 - Server-side APIs and role-based access control 04:38 - Agent-side identity attribution 06:55 - AI hacking and prompt injection demo 11:58 - Data supply chain and provenance 17:20 - Agent productivity and token consumption limits 19:20 - The "fabric" component and ecosystem fit 21:40 - Clarifying overall architecture and enforcement points 27:56 - Black Hat presence and upcoming events Subscribe to the newsletter at: https://danielmiessler.com/subscribe Join the UL community at: https://danielmiessler.com/upgrade Follow on X: https://x.com/danielmiessler Follow on LinkedIn: https://www.linkedin.com/in/danielmiessler/