Journalists have uncovered a handful of preprint academic studies with hidden prompts instructing A.I. reviewers to give positive responses