← All learning paths
Learning path · 4 steps
Govern agent tools and actions
Understand who a tool acts as, constrain its destinations, and require meaningful approval for consequential actions.
Agent builders & platform owners
- 01
Practical guide
Give agents a smaller permission footprint
Map user sign-in, connection identity and backend permissions.
Open guide → - 02
Practical guide
Govern agent connectors before they move data
Constrain connectors, endpoints and environment scope.
Open guide → - 03
Practical guide
Build a prompt-injection defense test plan
Validate that untrusted instructions cannot grant authority.
Open guide → - 04
Practical guide
Use an evidence-based AI release checklist
Capture approval, cancellation and revocation evidence.
Open guide →