“Intersectional hallucinations”: The AI flaw that could lead to dangerous misinformation