Course 03 · Securing and governing AI agents
When AI Takes Action
Nine video modules on securing and governing AI agents: architecture, safe connections, identity and human control, knowledge protection, manipulation risks, testing, monitoring, and production readiness.
When AI Takes Action Modules
Follow the nine modules in order, from understanding agent behavior to making an informed production release decision.
When AI Starts Taking Action
Understand the transition from assistants to agents that act through tools, and identify the autonomy, access, action, exposure, and accountability to review.
Watch lessonThe Secure Agent Blueprint
Map agent architecture around trust boundaries, independent execution checks, clear ownership, and runtime controls.
Watch lessonConnecting Agents Safely
Explore tool contracts, Model Context Protocol connections, validation, business rules, and the handling of credentials outside conversation context.
Watch lessonIdentity Permissions and Human Control
Connect initiating users, agent runtimes, and execution identities while binding human approvals to the specific action being authorized.
Watch lessonProtecting What Agents Know
Protect retrieved knowledge and persistent memory through access controls, preserved permissions, and clear information boundaries.
Watch lessonHow Agents Get Manipulated
Examine direct and indirect prompt injection, manipulation of reviewers, and layered defenses against untrusted inputs.
Watch lessonTesting Before Trusting
Define security invariants, model threats across entry channels, and test agent behavior within authorized delivery boundaries.
Watch lessonMonitoring and Responding
Follow related events across runtimes, retrieval, and tools to support monitoring, containment, investigation, and recovery.
Watch lessonGoing Live and Staying in Control
Review release decisions, supplier responsibilities, and ongoing governance before moving an agent into production.
Watch lesson