I think the worst response is pretending that the system simply made a small technical error. When AI affects users, a mistake can become a trust problem very quickly. Companies should have an incident process ready before something happens: who pauses the system, who investigates, who talks to customers, and who approves the fix. A serious AI mistake also needs a post-incident review, because the goal is not only to repair one failure but to make the next one less likely.