Facial recognition has put innocent people in prison. Risk tools score individuals on group statistics. Predictive policing amplifies historical bias at scale. The stakes have no equivalent in government AI. Neither do the governance obligations.
Robodebt. SyRI. MiDAS. Three countries, three systems, one failure: automated consequences without human review. When government AI affects citizens' rights, due process is not a feature to add later. It is the acceptance criterion.
Procuring AI from a vendor doesn't transfer accountability. The agency still owns the due process obligation, the legal liability, and the reputational damage. Here's what every government AI contract actually needs to include.
The same governance failure played out in Australia, the Netherlands, the US, and across democratic governments worldwide. The pattern is identical. The PM lesson is too. Here's what responsible government AI deployment actually requires.