the government doesn't publish any centralised audit on the extent of AI use
Perhaps they should develop an AI system that will decide what government information to publish . After all, if they're just going to carry on using AI systems to judge people on things such as likelihood of being a criminal, or reoffender or benefit cheat, not caring about all the negative consequences for the people that are wrongly accused, then if the government is harmed because the AI decides to release something they want to keep covered up it's just the same tough luck.