I'm seeing a use case here. Cops use the blooper-prone AI to save 6-8 hours/week (per the article) writing up drearily routine reports. And the steady stream of AI bloopers - which might well be shared with the public - keeps everyone aware that the AI's output definitely can't be trusted on its own.