risk
Model/data drift and inadequate post-deployment monitoring
Distribution shift between training and deployment data silently degrades accuracy, and without ongoing monitoring, model decay and emerging failure modes go undetected with no trigger to retrain or decommission. Uncontrolled updates alter behaviour and invalidate prior assessments.
Record JSON · Open in map · Data retrieval guide
Catalog revision: 24028ffcfc2b295fa1b08ee6caa84b765f0731b321496bf4f548c49ad2177028. A connection does not establish full coverage.
Attributes
- category
- ai_governance
- domain
- AI Governance
- Logging, Monitoring & Detection
- taxonomy
- nist-ai-rmf-risk
- iso-23894-ai-risk
- eu-ai-act-risk
- inherent_rating
- high
Details
- risk_id
- ai-model-drift-monitoring
- category
- ai_governance
- likelihood
- high
- impact
- medium
- inherent_rating
- high
- treatment
- mitigate
- taxonomies
- nist-ai-rmf-risk
- iso-23894-ai-risk
- eu-ai-act-risk
Source
No record-specific source URL is provided.
Connections
- UC-AI-08 — Log and monitor AI system behavior in operation mitigates Model/data drift and inadequate post-deployment monitoring
- strength
- primary
- rationale
- Continuous monitoring against performance/behavior metrics with drift and anomaly alerting is the post-deployment detection the risk says is missing.
- UC-AI-21 — Commission independent third-party AI evaluations on a quarterly cadence mitigates Model/data drift and inadequate post-deployment monitoring
- strength
- related
- rationale
- A recurring quarterly evaluation cadence catches degradation between releases that internal monitoring misses.
- UC-AI-07 — Verify, validate, and control AI deployment and changes mitigates Model/data drift and inadequate post-deployment monitoring
- strength
- related
- rationale
- Routing updates/retraining through reassessment and sign-off prevents the uncontrolled updates the risk names, but does not detect the distribution drift at the risk's core; continuous monitoring (UC-08) is the operative defense.