In the past few months, AI models have gone from producing hallucinations to becoming effective at finding security flaws in software, according to developers who maintain widely used cyber ...