Implementation guide
OWASP AISVS v1.01
Monitoring, Logging & Anomaly Detection · Detection and Alerting
Spot unusual or probing behaviour
Verify that behavioral anomaly detection identifies unusual conversation patterns, excessive retry attempts, or probing behaviors.
OWASP AISVS v1.01, requirement 12.2.2
In plain English
Look for unusual behaviour across a conversation or an account: many retries with small changes, systematic probing, or patterns far from normal use. One blocked prompt is noise. Forty variations on it in ten minutes is someone working on a bypass.
What good looks like
- Normal usage patterns are measured, per user type where it matters.
- Deviations raise alerts: high retry or regeneration rates, rapid rephrasing after blocks, unusual session length or timing.
- Detection works at session and account level as well as per message.
- Thresholds are tuned to keep false alarms manageable.
How teams usually do it
- Session-level metrics computed from the AI logs: blocks per session, similarity between consecutive prompts, retry counts.
- Behavioural analytics in your monitoring platform, with baselines per user group.
- Automatic slow-downs or step-up verification for sessions that trip the detector.
Suggested evidence
- Detection rules and baselines.
- Sample alerts with the session behind them.
Common mistakes
- Treating every blocked prompt as a separate event, so repeated probing is missed.
- Baselines taken from a period when an attack was already running.
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.