What is Performance Drift?
Performance drift occurs when an AI system's behavior degrades over time, leading to increased errors,
incidents, or unexpected outputs. This can happen due to:
- Changes in input data distribution
- Model degradation or concept drift
- Environmental changes affecting system behavior
- Integration issues with other systems
EU AI Act Article 15: Accuracy, Robustness, and Cybersecurity
Article 15 of the EU AI Act requires that high-risk AI systems be designed and developed to achieve
an appropriate level of accuracy, robustness, and cybersecurity throughout their lifecycle.
Monitoring Requirement: Providers must continuously monitor their AI systems for
performance degradation and take corrective action when drift is detected.
How the Dashboard Tracks Drift
The System Robustness widget compares:
- Current Period: High-impact incidents in the last 30 days
- Previous Period: High-impact incidents in the previous 30 days
- Alert Threshold: If error rate increases by more than 10%, an alert is triggered
What to Do When Drift is Detected
- Investigate: Review incident logs to identify patterns or root causes
- Document: Record the drift detection and investigation findings
- Remediate: Take corrective action (retraining, system updates, or deactivation)
- Report: If required, notify relevant authorities per Article 73 (serious incidents)
Related EU AI Act Articles
- Article 15: Accuracy, robustness, and cybersecurity requirements
- Article 12: Record-keeping of incidents and malfunctions
- Article 73: Reporting obligations for serious incidents