Implementation guide
OWASP AISVS v1.01
Monitoring, Logging & Anomaly Detection · Model, Data, and Performance Drift Detection
Track hallucination rate over time
Verify that hallucination rates are tracked as continuous time-series metrics to enable trend analysis and detection of sustained model degradation.
OWASP AISVS v1.01, requirement 12.3.3
In plain English
Record hallucination rates continuously and look at the trend. A single figure says little. A rate that has crept up for three weeks says that something changed: the model, the documents or the questions.
What good looks like
- Hallucination rates are recorded as a time series.
- Trends are visible on a dashboard, broken down by feature or source where useful.
- A sustained increase triggers a review.
- Releases and data changes are marked on the chart.
How teams usually do it
- Feed detector results (12.3.2) into your metrics system.
- Alert on trend as well as level, for example the weekly average rising above its usual range.
- Keep the detection method stable, or mark the chart when it changes.
Suggested evidence
- The dashboard.
- Alert rules.
- A record of a trend that was investigated.
Common mistakes
- Looking at single figures with no trend.
- Changing the detector and reading the jump as a change in the model.
Not sure which requirements apply to your system? Take the assessment, or browse every guide in this chapter.
Requirement text from the OWASP AI Security Verification Standard, licensed under CC BY-SA 4.0. This guidance is shared under the same licence.