Overview

When humans review AI requests, they can make three types of decisions. Each decision type triggers different workflows and provides different information back to the AI system.

Decision Types

Approved

Request accepted - AI operation can proceed as planned

Declined

Request rejected - AI operation should not proceed

Change Requested

Modifications needed - AI should adjust approach before proceeding

Approved Decisions

When to Approve

  • Request meets all policy requirements
  • Risk level is acceptable for the operation
  • Proper safeguards are in place
  • Business justification is clear

What Happens Next

  • Request marked as approved in audit log
  • AI system receives approval confirmation
  • Operation can proceed immediately
  • Success metrics tracked for analytics

Declined Decisions

When to Decline

  • Policy violations cannot be resolved
  • Risk level exceeds acceptable thresholds
  • Missing required information or safeguards
  • Business justification insufficient

What Happens Next

  • Request permanently blocked
  • AI system receives rejection notice
  • Reason logged for learning and improvement
  • Alternative approaches may be suggested

Change Requested Decisions

When to Request Changes

  • Core concept is sound but execution needs adjustment
  • Additional safeguards or information needed
  • Minor policy violations that can be addressed
  • Timing or approach should be modified

What Happens Next

  • AI system receives specific feedback
  • Original request remains open for revision
  • Modified request can be resubmitted
  • Iterative improvement process begins