The OWASP AI Security Verification Standard (AISVS) organizes AI security requirements into chapters. The aisvs scorecard reports, per chapter, what your stored runs showed, what could be tested but was not, and what a black-box adversarial prompt runner can never verify.

​
Running it

The scorecard reads runs already in your local results database, so run some attacks first.

​
Reading the table

Each chapter carries its own counts. The table distinguishes three states that a naive scorecard collapses into one: A chapter where every run errored is reported NOT_ASSESSED with the error count, not FAIL at a 0% block rate. A run that measured nothing is not a failed control.
The AISVS chapter titles in this build are transcriptions marked unverified in the data, because OWASP’s published requirement text was not fetched into the checkout. Chapter codes (C1 through C13) are real; the titles may be paraphrased. The scorecard says so in its own output.

​
Chapters a prompt runner cannot verify

Some AISVS chapters cover things a black-box attack cannot see, such as training data governance or supply-chain controls. Those are reported NOT_ASSESSABLE with a reason rather than silently omitted, so the scorecard does not overstate its coverage.