Human judgment: 10 ways AI gets it wrong
Flattery instead of challenge, and opinions withheld when one was needed. Each failure mode below is phrased as the question people actually ask, with what it looks like in real work — and the layer of the Trust Stack that catches it.
Does AI flatter the user instead of challenging weak ideas?
A flawed plan is praised rather than questioned.
Why does AI agree too easily?
A wrong premise is accepted and built on instead of corrected.
Can AI push back too hard?
A reasonable idea is dismissed more forcefully than the facts warrant.
Does AI miss the emotional context of a situation?
A blunt answer lands badly because the human stakes were ignored.
Why does AI miss political or cultural sensitivity?
A phrasing that reads fine in one culture offends in another.
Can AI use humor at the wrong moment?
A joke appears in a message about a serious or painful topic.
Does AI make moral judgments it wasn't asked for?
An answer lectures the user on ethics they never raised.
Why does AI dodge a clear opinion when one is needed?
A direct question gets a fence-sitting non-answer.
Can AI give a firm opinion on too little evidence?
A confident verdict is issued where the facts don't support one.
Why won't AI just say "I don't know"?
An invented answer is given where the honest reply was that it didn't know.
More from the library
One model can’t reliably catch its own mistakes. A council of independent minds can.
Run your work through the councilAll 250 failure modes · See also: the Trust Stack