ACL 2026 desk-rejects 100+ papers over hallucinated citations

A top venue in natural language processing announcing it caught over 100 accepted papers citing literature that does not exist is the kind of headline that lands somewhere between embarrassing and structurally important. ACL 2026's program chairs said the papers were flagged during final consistency checks on camera-ready versions, after peer review had already accepted them.

The detection loop is worth pausing on. Per the statement, an automated system surfaced potential citation issues, and then human experts, the program chairs and senior area chairs, reviewed each flagged reference to confirm it did not exist. So the venue is not just running a spell-checker over bibliographies, it is manually verifying suspected hallucinations before pulling papers.

The policy language is the part with teeth. The chairs write that 'it is the fundamental responsibility of the authors to ensure the accuracy of all content within their manuscript,' and go on to say that 'using an automated system as a proxy to generate such citations is equally unacceptable.' In other words, whether a human hallucinated the citation or a language model did, the paper is out.

The chairs, David Jurgens, Maria Liakata, Viviane Moreira and Jiajun Zhang, initially desk-rejected the papers outright but updated the decision after meeting with the ACL executive: affected papers 'must go through the normal resubmission process' rather than being permanently barred. The honest caveat is that the statement does not break out how many of the 100-plus papers were LLM-assisted versus hand-edited sloppily, does not name the papers or authors, and does not say what share of ACL 2026 acceptances that number represents.

What this really is, though, is a template. If ACL can run automated citation checks at camera-ready time and enforce a rejection policy, the other major venues can too, and the return goes to authors who verify their own references before hitting submit.