Leading AI models like GPT and Gemini routinely cite text passages in document analyses that don't actually support their answers. Even when the answer is right, the cited evidence is often wrong. Researchers at Peking University call this "attribution hallucination," a risk for regulated fields lik
