Attention Is All You Need Has Five Fake Twins on Crossref

Query Crossref for "Attention Is All You Need," the 2017 Vaswani et al. paper that introduced the Transformer and has been cited more than 140,000 times, and five separate records come back.

Same title. Same eight authors. Byte for byte identical. All five list the publication year as 2025. All five carry a DOI under the prefix 10.65215, registered not to a computer science venue but to an entity calling itself the Shenzhen Medical Academy of Research and Translation, via a Chinese preprint host.

10.65215/r5bs2d54
10.65215/ysbyhc05
10.65215/mdcm8z23
10.65215/nxvz2v36
10.65215/2q58a426

All five titled "Attention Is All You Need." All five dated 2025. All five resolving to live, indexed pages at the time of writing.

I checked each one directly rather than trusting the search result at face value. The DOIs resolve. The Crossref records are real and indexed, not an artifact of my own tooling. The actual 2017 NeurIPS paper has no Crossref DOI at all. NeurIPS doesn't systematically register one for its papers, unlike venues indexed through the ACL Anthology. A title search against Crossref alone has nothing genuine to return. The five fabrications are the only candidates on offer.

Why this matters beyond one paper

Anything that verifies citations against registry data is exposed to this mechanism. Mine, or anyone else's. If the correct source doesn't answer, and every registry has its own coverage gaps and occasional downtime, a single spam duplicate can be the only voice in the room. Agreement across independent sources is the meaningful signal, not any one source's word in isolation. A claim resting on exactly one unfamiliar registry deserves real skepticism before it gets treated as settled.

The pattern isn't rare

I picked ten of the most-read AI papers on Hugging Face's Daily Papers listing that week. Real preprints, real citation lists, not curated for this piece. I ran their actual .bib / .bbl files, taken directly from each paper's own arXiv source, through the same thirteen-registry checks StrictCite uses. 831 references total. More than 300 got flagged for closer review.

A meaningful share of that is noise, not findings, and it's worth saying plainly rather than folding it into a bigger number. Semantic Scholar's keyless tier shares one rate-limited pool with every anonymous caller on the internet, and it throttled heavily partway through the run. A reference that would normally clear with two independent sources agreeing can read as unconfirmed for reasons that have nothing to do with the paper citing it.

Real, specific, independently verifiable problems showed up anyway:

• One paper's citation year read 1609. Not a year. A fragment of an AAAI DOI (10.1609/...) that some bibliography tool had mistaken for one. The same pattern, digits lifted from an identifier instead of an actual year, showed up independently in three separate papers.
• Another paper cited a diffusion-model paper by DOI. That DOI resolves to an unrelated paper on human-factors modeling for autonomous driving, credited to four different authors with no connection to diffusion models.

OpenAI's and Meta's own citations are broken too

This isn't confined to fast-moving preprints from smaller teams. One paper in the sample cites GPT-4o and Llama 3 as baseline models. I pulled the official arXiv BibTeX export both companies publish for their own papers, straight from arXiv's own /bibtex/ endpoint, not a third-party citation manager.

OpenAI's own GPT-4o System Card export reads, in full:

author={OpenAI and : and Aaron Hurst and Adam Lerer and ...}

A bare colon, listed as a co-author, in the citation OpenAI itself publishes for its own paper.

Meta's Llama 3 Herd of Models export is broken a different way. Real people's names split in half:

...and Grigory Sizov and Guangyi and Zhang and Guna Lakshminarayanan...

One author, Guangyi Zhang, becomes two nonexistent people: "Guangyi" and "Zhang." The same fault recurs later in the same list for Yu Wang. Somewhere in the tooling that generated these official citations, the literal string "and" landed inside a name instead of only between names. It shipped, from two of the best-resourced labs in the industry, into every paper that copies the citation as published.

The corruption doesn't stop at the manuscript. DataCite, the registry arXiv reports to, has ingested the identical broken author lists as the record of standing. A tool that treats a registry's word as ground truth would confirm both entries as correct, because the fault originates at the source, not somewhere downstream of it.

What this doesn't mean

It doesn't mean citation verification is a lost cause, or that these registries are unreliable on the whole. For most of what they cover, most of the time, they aren't. It means "the registry says so" and "it is true" are different claims, worth treating as different claims regardless of which tool is doing the asking.

I run StrictCite, a citation checker that runs a manuscript's real references against thirteen registries in the browser, no server upload, and reports disagreements instead of assuming any one source is right. If you want to run your own bibliography's actual reference file through the same checks: strictcite.com. The free tier needs no card.

Originally published on StrictCite's Field Notes, where I write up what turns up when real bibliographies get checked against real registries.