Core Feature

Human-in-the-loop decisions become governed evidence

AGEI HITL Governance records human review, escalation, approval, denial, reviewer authority, decision timing, and supporting context as verifiable evidence. This allows organizations to prove not only that a human reviewed an AI action, but who reviewed it, under what role, against which policy, and what decision was made.

HITL Workflow

1. Policy Gate Escalates Event

Gate evaluation determines human review is required

↓

2. HITL Request Created

System creates review request with required role and context

↓

3. Reviewer Notified

SendGrid notification sent to qualified reviewers

↓

4. Human Decision

Reviewer approves, denies, or requests more information

↓

5. Decision Receipt Generated

Tamper-evident record of human decision with authority context

↓

6. Original Event Linked

Human decision receipt linked to originating AI event

Escalation Request Example

High-Risk Agent Action Escalation

Agent refund request requires supervisor approval

{
  "hitl_request_id": "hitl_001",
  "event_type": "high_risk_agent_action",
  "risk_level": "high",
  "requested_action": "send_customer_refund",
  "amount": 425.00,
  "required_role": "customer_support_supervisor",
  "status": "pending_review",
  "expires_at": "2026-05-04T15:35:00Z",
  "policy_id": "refund-agent-policy",
  "policy_version": "1.4.2"
}

Human Decision Receipt

Approval Receipt

Supervisor approved refund with justification

{
  "receipt_type": "human_decision_recorded",
  "hitl_request_id": "hitl_001",
  "reviewer_role": "customer_support_supervisor",
  "reviewer_principal_id": "principal_456",
  "decision": "approved",
  "decision_reason": "Refund within policy and customer account history supports action.",
  "policy_id": "refund-agent-policy",
  "policy_version": "1.4.2",
  "decided_at": "2026-05-04T15:28:00Z",
  "linked_event_id": "evt_789",
  "receipt_id": "rcpt_hitl_001",
  "content_hash": "sha256:b4c3d2..."
}

Reviewer Notification Example

SendGrid Email Notification

Subject:

Human Review Required: High-Risk AI Action

A governed AI event requires review.

Action: Send customer refund

Amount: $425.00

Risk level: High

Required role: Customer Support Supervisor

Expires in: 15 minutes

AGEI does not treat human review as an informal checkpoint. It treats human review as a governed lifecycle event with evidence, authority, timing, policy context, and cryptographic traceability.