How to meet it
OWASP AISVS v1.01
Adversarial Robustness · Model-Extraction Defense
Cap sampling settings callers can change
Verify that inference endpoints constrain sampling and decoding parameters supplied by untrusted callers (e.g., temperature, top-k, top-p, logit bias, candidate/n count) to server-side approved ranges and defaults, rejecting out-of-policy values.
OWASP AISVS v1.01, requirement 11.3.5
In plain English
Limit the generation settings callers can change, such as temperature and logit bias, to ranges you approve, and reject anything outside them.
What good looks like
- Allowed ranges and defaults are set for each parameter.
- Out-of-range values are rejected.
- Parameters you do not want callers to set are not exposed.
How teams usually do it
- Validation in your API layer before calling the model.
Evidence an auditor would accept
- Parameter policy and tests.
Common mistakes
- Passing caller parameters straight through to the model API.
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.