Generative AI tools are repeating false news claims more often than last year, with one top chatbot falling from a perfect score to nearly 50% failure.