How to meet it
OWASP AISVS v1.01
Adversarial Robustness · Model Alignment, Safety, and Robustness Testing and Training
Harden the model against adversarial input
Verify that models are hardened against adversarial inputs.
OWASP AISVS v1.01, requirement 11.1.4
In plain English
Make the model harder to fool with adversarial input, for example through adversarial training, input preprocessing or extra guard models.
What good looks like
- At least one hardening measure is in place for each input type.
- Its effect is measured with the tests from 11.1.3.
- Hardening is repeated when the model changes.
How teams usually do it
- Adversarial training where you train the model.
- Input transformations and guard models in front of provider models.
Evidence an auditor would accept
- The hardening measures and before-and-after test results.
Common mistakes
- Claiming the model is hardened with no measurement.
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.