How to meet it
OWASP AISVS v1.01
Training Data Integrity & Traceability · Training Data Quality and Security Assurance
Detect poisoned training data
Verify that training and fine-tuning pipelines implement poisoning detection techniques to identify potential data poisoning or unintentional corruption in training data.
OWASP AISVS v1.01, requirement 1.3.1
In plain English
Check training data for signs that someone has slipped in records designed to change how the model behaves, and for accidental corruption.
What good looks like
- Each training and fine-tuning run includes a poisoning check on new data.
- Outliers, duplicates and sudden shifts in label distribution are flagged for review.
- Flagged records are held back until someone has looked at them.
How teams usually do it
- Statistical checks for outliers and label distribution changes between data versions.
- Data validation tools in the pipeline, plus spot checks of flagged items by a person.
Evidence an auditor would accept
- Pipeline configuration showing the checks.
- Reports from recent runs, with what was flagged and what happened to it.
Common mistakes
- Checking only the original dataset and not the feedback or fine-tuning data added later.
- Checks that log warnings but never stop a run.
Not sure which requirements apply to your system? Take the assessment.
Requirement text from the OWASP AI Security Verification Standard, licensed under CC BY-SA 4.0. This guidance is shared under the same licence.