Chatbots Misvalidate Mathematical Proofs

Recent tests reported in Econ Journal Watch show four large chat models (Gemini, Refine, Claude, ChatGPT) sometimes endorse incorrect mathematical proofs and produce false positives, with performance varying by model and depending on whether corrected literature was already available. That reveals a concrete failure mode where AI 'validation' can amplify errors rather than catch them, even in tightly formal domains. — This undermines claims that chatbots can be trusted as automatic validators in research, legal, or regulatory settings and argues for stricter provenance, human oversight, and rules around AI use in scholarly and technical verification.