Testing and security teams break the new models before release

Developer and security teams intensively use and "break" new artificial intelligence models in advance to uncover their shortcomings before release. Fixes are made based on the problems found, so the final model becomes more reliable and robust.