When AI Becomes an Identity: Defending Agentic Systems with Microsoft Defender - track1 session 11

As organisations adopt AI agents and autonomous workflows, identities are no longer limited to humans and service accounts. AI agents increasingly act as non-human identities with persistent access, decision-making capabilities, and the ability to interact across cloud, data, and security boundaries. This session explores how agentic systems change the security threat model and introduce new risks that traditional identity, endpoint, and workload controls were not designed to handle. I will examine common attack patterns targeting AI agents — from permission misuse to manipulation of agent behaviour and discuss why visibility, signal correlation, and human-in-the-loop response are critical. Using Microsoft Defender as the security lens, the session will show how security teams can detect, investigate, and respond to threats in agent-driven, identity-centric environments, while maintaining governance and control as AI adoption scales. Attendees will leave with a practical understanding of how to reason about AI agents as identities and how to evolve their Defender strategy to support secure, resilient agentic systems. ⏱️ Timestamps: 0:00 - Introduction & Overview 3:00 - AI Agents as Non-Human Identities 7:00 - How Agentic Systems Change the Threat Model 12:00 - Common Attack Patterns Targeting AI Agents 17:00 - Permission Misuse & Agent Behaviour Manipulation 22:00 - Detecting Threats with Microsoft Defender 26:00 - Human-in-the-Loop Response Strategies 29:00 - Key Takeaways & Q&A 📌 Part of the Microsoft Purview & Defender Days 2026 series. 🔔 Subscribe for more Microsoft Security sessions!