Who’s Responsible When Government AI Gives the Wrong Answer?

If human reviewers fade away and government work product is AI-produced, who will be held responsible for when things go wrong?