SCIgen's "Rooter" was accepted in 2005. I think the next step would be submitting an array of bogus articles citing each other to achieve an impressive h-index. Maybe adding AI-generated articles to the batches would be one of the ways to verify the reviewers in more serious journals.