A contribution to practical testing
Roei Arpaly contributed adversarial input-leakage testing material to the OWASP AI Testing Guide. The work appears in AITG-APP-04, Testing for Input Leakage, under Adversarial Attacks for Input Leakage.
What the guide tests
The section moves beyond direct repetition of sensitive inputs. It includes changes to a sensitive value's representation, changes to the words that identify it, affix injection and combinations of these techniques. These examples help testers examine input and output filters together.
Connection to the research
The contribution translates ideas investigated in AdvPIIBench into examples that AI application testers can use in the community guide.