If you can tune the specifics-sensitivity curve, you should be able to handle both cases. You have to be willing to refuse to provide an answer when you have low confidence.
But some projects as presented don't have a clear refusal option.
ML would go a long way if the Hippocratic Oath (or derivative thereof) were taken & adhered to by the industry.
It seems like a problem when classifying user-provided images (eg. identifying obscene images on a social network) but not so relevant when you own the sensors.