All AI-generated content — including items, rationales, and references — should be carefully fact-checked by human subject matter experts. AI-generated inaccuracies can be more difficult to detect than human errors because they often lack the typographical mistakes and uncertainty cues that are typically associated with human errors.