Research brief
The most obvious AI audit risk is hallucination, but the more operationally dangerous failure mode may be premature closure: fluent, plausible audit language that makes a conclusion feel resolved before the reviewer has tested the underlying evidence.
Key pages