How to meet it
OWASP AISVS v1.01
Adversarial Robustness · Model Alignment, Safety, and Robustness Testing and Training
Model is safety-trained
Verify that the model has undergone alignment and safety training or fine-tuning to prevent the model from generating disallowed content categories.
OWASP AISVS v1.01, requirement 11.1.1
In plain English
The model you use should have been trained or fine-tuned to refuse the kinds of content you do not allow. If you use a provider's model, check what safety training it has had.
What good looks like
- Your disallowed content categories are written down.
- The model has had safety training covering them, and you have checked this.
- Gaps are covered by other controls such as output filtering.
How teams usually do it
- For provider models, the provider's model card and safety documentation, checked against your categories.
- For your own models, safety fine-tuning with refusal examples for each category.
Evidence an auditor would accept
- The category list and the model's safety documentation or training records.
- Test results showing refusals in each category.
Common mistakes
- Using a base model with no safety training in a customer-facing product.
- Assuming a provider's general safety training covers categories specific to your domain.
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.