“This is more complex than it may at first appear, as references can be detailed by authors in a variety of different ways, often do not include DOIs, and simple tools to identify hallucinated references can produce false positives,” Graf told us by email.

It’s not complex when you have a fucking style manual which specifies exactly how you detail references so you don’t have this kind of problem.

I was gonna say, surely there’s a database of published works you could literally query to check for validity. I mean literally traditional algorithmic verification. Why the hell would you need an AI for that. AIs are for black box problems but this has been solved way back. This tells me whoever is in charge is more concerned with optics than with things going well or they are highly incompetent.

For medical articles, there’s PubMed, the medical research database mandated by the U.S. and E.U. For other fields, YMMV.