Closed-box testing (also called "black box testing") means testing of the AI-system in which the principal information used as the basis for designing and implementing test cases is the external inputs and the AI-system's output. Usually, the tester has no access to internals such as the AI model.
ISO/IEC TR 29119-11:2020 testing (3.131) in which the principal test basis (3.84) is the external inputs and outputs of the test item (3.107), commonly based on a specification, rather than its implementation in source code or executable software[ISO/IEC 29119-1:2022, 3.75]