Forrester published research arguing that AI applications need continuous testing rather than pre-launch checks. It says AI systems fail differently from ordinary software. They can give inconsistent answers, retrieve wrong information, misuse tools, drift over time, or produce biased output while sounding confident. Forrester calls for structured behaviour checks, adversarial testing, human review, governance controls and live monitoring. It says these checks should be versioned and built into delivery pipelines like test suites. High-risk uses need stronger release gates; low-risk uses still need defined success criteria. Forrester also expects classical software testing vendors and AI validation specialists to merge, through partnerships and acquisitions. It recommends inventorying AI applications by business risk and assigning shared ownership beyond the model team. The research was written by Diego Lo Giudice and Rowan Curran.