From AI Policy to Runtime Authority: Making Governance Enforceable | Kamilla Harcej, TauGuard

When someone tells you their AI is governed, Kamilla Harcej says to look past the policy document and the dashboard. Ask them to show you the refusal. If the system can't stop a wrong action before it runs, the policy is just a PDF. An AI agent can recommend a payment, a prescription or a production change, and the company carries the consequences either way. Kamilla co-founded TauGuard, which builds TauDIL GOS, a governance operating system that turns AI policy into enforceable runtime control. In this session at the Cloud Convergence Summit she's joined by Michael Harcej, the system's architect. Their core idea is a separation of powers: AI can advise, and governance decides whether a consequential action is allowed before it executes. As Kamilla puts it, "recommendation is not authority and prediction is not authority." What you'll learn: • Execution asks "did we do it?" Governance asks "may we do it?" • The execution boundary, and why governance has to act before consequences happen • Admissibility: is this specific action allowed, right now? • Why a refusal means the system is working • Structural refusal: no retry path, no override, and a traceable record • Why governance should run on your own servers and keep working offline • Departments, domains and keeping each one's data and rules separate • Cryptographically chained audit records a regulator can trace backward • Swapping AI models without rebuilding your governance • Purpose as the constitutional layer, and what they call zombie governance • Q&A: AI inventory and passports, where the system lives, and incident logs for regulators About the speaker: Kamilla Harcej co-founded TauGuard with her husband Michael Harcej,, where she is building TauDIL GOS, a governance operating system that turns AI policy into enforceable runtime control. Her work separates intelligence from authority: AI can advise, but governance decides whether a consequential action is allowed before it runs. She is taking the system into institutional pilots and enterprise deployments. About the summit: The Cloud Convergence Summit ran online October 8–9, 2026, hosted by Verge Technologies with Carahsoft. Two afternoons of sessions on managing many clouds as one. Day 1 covered infrastructure and architecture. Day 2 covered governing what the machines decide. ⏱️ CHAPTERS 0:00 Introduction 1:00 Why AI governance is urgent now 3:00 From AI policy to runtime authority 5:30 Michael Harcej on why wrapping AI doesn't work 7:30 (est) Full control on your own server 8:50 Departments and domains 9:50 When AI decides, the company is still accountable 12:00 Intelligence is not authority 13:50 The execution boundary 14:50 Why policy alone isn't enough 15:50 Admissibility 16:50 A refusal means it's working 17:50 Structural refusal 18:50 Show me the refusal 19:50 How a refusal works in practice 21:20 Chained audit records 22:00 A knowledge graph for every domain 24:40 Stopping an action mid-execution 25:40 Your data stays with you, even offline 26:50 Upload your own rules 27:30 Swap the AI anytime 28:50 Authority from the top down 30:20 Purpose and zombie governance 31:50 Q&A: a live cross-cloud database move 34:30 Who it's built for 35:30 AI inventory and passports 36:00 What a governance operating system is 38:50 Working with your existing systems 41:40 Does every AI agent need to be registered? 44:00 Active in every session? 46:40 Where does it live? 54:40 Incident logs for regulators 🔗 TauGuard: https://www.tauguard.xyz 🔗 Learn about Verge Technologies and SentientDB: https://vergetech.cloud 🔗 See the full summit lineup: https://summit.vergetech.cloud #AIGovernance #AICompliance #RuntimeGovernance #ResponsibleAI #AgenticAI #RegTech #AIRisk #CloudConvergence