Back to the AI Control Library
12 controls | 12 core | 0 supplemental
Why this family matters
Deployment begins a new period of uncertainty. Data changes, user behavior shifts, dependencies evolve and previously acceptable outputs can become unreliable. This family addresses how organizations detect those changes and decide when intervention, recalibration or reassessment is necessary.
What this family covers
The controls cover model performance, input and concept drift, security alerts, output quality, fairness drift, infrastructure failures and operational reporting. They also address monitoring agent actions and revisiting monitoring when a system changes.
How the four control types apply
Technical controls detect changes and generate alerts. Operational controls assign review, response and recalibration responsibilities. Contextual conditions determine what acceptable performance means in a changing deployment. Governance controls establish applicable external monitoring obligations.
Where to begin
Establish a baseline, define actionable thresholds, and ensure that each meaningful alert has an owner and a response path.
Controls in this family
This family contains 12 controls: 12 core and 0 supplemental. Each page provides requirements, implementation steps, verification criteria, evidence, escalation and a worked example. Applicability depends on the system, information and risks involved.
Core controls
- CF09-001: AI Monitoring & Drift Detection Policy
- CF09-002: AI Model Performance Monitoring
- CF09-003: Input Data Drift Detection
- CF09-004: Concept Drift & Model Decay Detection
- CF09-005: AI Security Monitoring & Alerting
- CF09-006: AI Threat Detection Updates
- CF09-007: AI Output Quality Monitoring
- CF09-008: Operational Fairness & Bias Drift Monitoring
- CF09-009: AI Model Failure & Infrastructure Monitoring
- CF09-010: AI Monitoring Metrics & Reporting
- CF09-011: Agentic System Action Monitoring
- CF09-012: Monitoring Re-calibration on System Change
Applying this family
Connect production observations to accountable decisions. Separate input change, task deterioration, output quality, fairness, security and infrastructure health, because one favorable indicator cannot establish the others. Track missing outcomes and visibility gaps explicitly, investigate drift before changing the model and retest monitoring whenever the system changes.
Version 1.0