New benchmark reveals AI models often cite wrong sources even when answers are correct

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…