Enforcement Actions
Overview
Enforcement actions define how Peridot responds when a policy is triggered.
They transform governance rules into real-time control over AI behavior.
Types of Enforcement Actions
Block
Stops a request before execution.
Used for:
Sensitive data violations
Unauthorized model usage
Reroute
Redirects a request to an approved model or system.
Used for:
Policy-based routing
Model restrictions
Require Approval
Pauses execution until approved by a human.
Used for:
High-risk workflows
Sensitive data scenarios
Log
Records the event for audit and monitoring.
Used for:
Compliance
Observability
Create Incident
Triggers an incident for investigation and response.
Used for:
Policy violations
Data exposure risks
How Enforcement Works
At runtime:
Policy is triggered
Matching enforcement actions are selected
Actions are executed immediately
Request is modified, blocked, or escalated
Combining Actions
Multiple actions can be applied together:
Example:
Reroute request
Log event
Create incident
Example Scenario
A user attempts to send sensitive data to an external model:
Policy is triggered
Request is rerouted to approved model
Event is logged
Incident is created
In Production
Enforcement is applied in real time
Actions are executed in milliseconds
All actions are logged
Behavior is consistent across systems
Best Practices
Start with logging before blocking
Use approvals for edge cases
Combine enforcement with incident workflows
Next Steps
Learn about [Incidents Overview]
Configure [AI Incident Playbooks]