Ethics watchdogs are looking out for potentially undisclosed use of generative AI in scientific writing. But there’s no foolproof way to catch it all yet.