Where have you caught AI being confidently wrong?

Ford just hit number one in JD Power quality for the first time since 2010. The twist nobody saw coming: it got there by rehiring more than 350 veteran engineers to fix what its automated systems got wrong. Ford's own VP admitted they thought plugging in AI would produce a quality product on its own. It did not. The systems were trained on fragmented data, and the people who could have caught the errors had already left, taking decades of gut level judgment with them. What gets me is the failure mode. AI rarely breaks loudly. It breaks fluently. The output looks polished and sure, and the mistake hides until someone who actually knows the work looks twice. That is the Oversight Gap, and it is not just a Ford problem. It shows up in the classroom, the client deliverable, the homework answer nobody double checked. So I will put it to you: where in your own work have you caught an AI output that looked perfect and was quietly wrong, and what tipped you off? Full breakdown and a four question framework here: https://www.harvestkernel.com/ford-automation-oversight-gap-ai-literacy/