Personally, I think confabulations would be a better term. To the best of my understanding, these AI rely on a model similar to the reconstructive theory of memory in humans. The connotation of the word confabulation indicates no maliciousness while highlighting the erroneous nature of the action.