A regression test is the rerunning of existing tests after a software change to confirm that functionality that previously worked still works. The purpose is to catch regressions, cases where a bug fix or new feature unintentionally breaks another part of the program. Regression tests are usually automated unit and integration tests that run whenever code changes.
In the AI field, regression tests serve as a benchmark for judging the quality of patches produced by AI coding agents, distinguishing a patch that merely stops a crash from one that also passes the project's existing test suite.