Kinda the whole point of Asimov's three laws were that even something so simple and obviously correct has subtle flaws.

Also the reason we're talking about this again is that machines are significantly less 'mere' than they were a few years ago, and we need to figure out how to approach this.

Agree that 'the computer effect' (if it doesn't already have a pithier name) results in humans first discounting anything that comes out of a machine, and then (once a few outputs have been validated and people start trusting the output) doing a full 180 and refusing to believe the machine could ever be wrong. However, to err is human and we have trained them in our image.