As AI spreads into sensitive areas, organisations and governments have developed frameworks setting out how it should be built and used, covering testing, transparency, human oversight and handling of personal data, with stricter expectations where the stakes are higher.
The aim is to gain the benefits while limiting harms such as unfair decisions or misuse, and to give people recourse when automated systems affect them. Rules vary by place and are still evolving, which can be hard for builders to navigate, but a shared direction is forming around testing high-impact systems and keeping a human responsible for important outcomes.