Researchers at Google’s threat unit found a zero-day exploit in Python that showed signs of large language model involvement.