How to meet it
OWASP AISVS v1.01
Model Lifecycle Management & Change Control · Model Validation & Testing
Re-test models after quantisation
Verify that models subjected to post-training quantization are re-evaluated against the same safety and alignment test suite on the compressed artifact before deployment.
OWASP AISVS v1.01, requirement 3.2.2
In plain English
If you shrink a model through quantisation, run the safety tests again on the shrunk version. Compression can weaken safety behaviour even when task accuracy looks fine.
What good looks like
- Quantised models go through the same safety and alignment suite as the original.
- Results are compared with the original model.
- Large drops block deployment.
How teams usually do it
- Adding quantised variants as separate entries in the evaluation pipeline.
- Recording both sets of results in the model registry.
Evidence an auditor would accept
- Test results for the original and quantised models side by side.
Common mistakes
- Assuming the quantised model behaves the same because it scores the same on the main task.
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.