Security for Software Development Company Series Season 2: Securing the Agentic Era

As organizations race to deploy agentic AI into real-world production environments, security has become the defining challenge. This session breaks down what it actually takes to secure AI agents—beyond standard LLM safety—and explores the architectural realities teams must understand before scaling autonomous systems.

We’ll examine why prompt injection remains the most critical and underestimated threat to agentic AI, and why traditional application security models fail when agents are granted autonomy to reason, plan, and take actions on behalf of users. Building on expert insights, the session highlights how automated red teaming has emerged as the only sustainable defense strategy as compliance, safety, and reliability expectations rise in modern enterprise stacks.

Participants will learn:

• The core security weaknesses unique to AI agents
• Why prompt injection exploits the flexibility of LLM-driven behavior
• How misaligned or manipulated contexts can lead to unintended or unsafe actions
• The limitations of manual guardrails and rule‑based defenses
• How Microsoft’s latest approaches to agent security and governance provide a blueprint for industry best practices, including continuous red teaming and layered oversight models

This session equips practitioners, architects, and leaders with the frameworks needed to harden AI agents for production workloads—ensuring they remain reliable, predictable, and resilient against emerging attack vectors.

Register now

Governing data access when AI systems make independent decisions about what data to retrieve and process. Demonstrates using Fabric and Purview for dynamic data classification, context-aware access policies, audit trails for autonomous actions, and compliance monitoring across agentic workflows.

Coming soon

Managing identity and lifecycle for AI systems operating with delegated authority. Covers provisioning autonomous systems, credential management, service principals for agentic workflows, policy enforcement, and implementing least-privilege for systems that make independent decisions.

Coming soon

Protecting the expanded supply chain of agentic systems: model sources, orchestration frameworks, tool integrations, and knowledge bases. Includes model provenance, dependency management for agent frameworks, SBOM for agentic applications, and protecting against supply chain attacks targeting autonomous systems.

Coming soon

Monitoring and tracing in environments where AI systems autonomously orchestrate multi-step workflows. Covers visibility into autonomous decision chains, tracking cross-system interactions, detecting anomalous behavior patterns, and building dashboards that illuminate agentic reasoning with Azure Monitor and Application Insights.

Coming soon

Hardening agentic systems for production: rate limiting for autonomous actions, circuit breakers for runaway workflows, rollback strategies, secrets management, environment isolation, and testing non-deterministic autonomous behavior. Includes progressive rollout for systems that evolve their capabilities.

Coming soon

Discovering and governing unauthorized autonomous AI deployments. Covers detecting shadow agentic systems, establishing approval workflows for autonomous capabilities, creating guardrails that enable innovation, and building centralized visibility across the enterprise's agentic landscape.

Coming soon

IP considerations and architectural patterns for autonomous AI systems. Covers patenting agentic innovations, protecting proprietary reasoning logic, Azure AI Foundry reference architectures for agentic applications, and architectural decisions that balance security, scalability, and IP protection.

Coming soon

Tenant isolation when AI systems operate autonomously on behalf of different customers. Preventing cross-tenant data leakage through autonomous actions, managing per-tenant autonomy policies, and scaling security controls for SaaS platforms in the agentic era.

Coming soon

Authentication and authorization across autonomous systems that collaborate. Covers establishing trust between agentic systems, zero-trust principles for autonomous interactions, credential management in distributed agentic networks, and preventing lateral movement using Entra ID.

Coming soon