Black-box testing

A software engineering testing method in which the tester can observe system behavior only through inputs and outputs, without access to internal implementation or execution processes. Unlike conventional software, in probabilistic AI agent scenarios it systematically misses risks.