Fake references are turning up in physics and engineering papers, and the rate more than tripled since 2023
An audit of 500,000 science, engineering and computing papers found references to studies that don’t seem to exist, and they’re becoming more common each year.
What the scientists did
Every scientific paper ends with a list of references: earlier studies the authors relied on. Readers trust that those studies are real. AI chatbots are known to sometimes invent realistic-looking references, so researchers wanted to know how often references that point to nothing are slipping into published science.
The team took 500,000 papers in chemistry, physics, materials science, engineering and computer science, published from January 2023 to March 2026. An automated tool checked 21.6 million references that had a DOI and flagged ones whose cited title couldn’t be found in online databases. Two experts then checked a random sample of 400 flagged references by hand.
Words to know
Reference (citation)
A note in a paper that points to an earlier study the authors used.
- A list at the end of a research paper
- The “Works Cited” page of a school report
Fabricated reference
A reference to a paper that doesn’t actually exist, even though it may look real.
- A made-up title with real-sounding author names
- A real journal name paired with an article it never published
DOI
A permanent ID code for a paper, like a barcode, that links to it online.
- 10.1073/pnas.2623869123 (the ID for this study)
What they found
The tool flagged 1,135 references in 619 papers, and the experts confirmed 85.8% of the sampled ones as fabricated. Affected papers rose from 6.3 per 10,000 in 2023 to 15.4 per 10,000 in 2025 and 23.0 per 10,000 in early 2026, or about 1 in every 435 papers.
Physics had the highest rate (25.6 per 10,000 papers), followed by materials science (18.4), computer science and AI (11.1), engineering (10.3) and chemistry (4.3). Most affected papers had one or two bad references, but 27 papers had five or more. Scaling up to about 7 million papers, the authors estimate around 8,500 published papers contain at least one fabricated reference. Only 2 of the 619 had been retracted.
Why it matters to you
If you write a research paper or report using an AI chatbot, this is a real warning: check every source yourself before you cite it. It also means that even published science needs careful reading, and that journals may soon run automatic reference checks, the same way teachers run plagiarism checks.
Where this could lead
Automatic reference checks
Journals could screen every reference when a paper is submitted, as the authors recommend.
1–3 yearsCleaning up past papers
Publishers could scan already published papers and correct or flag fabricated references.
1–3 yearsThese are possibilities the research points toward, not promises. Most early findings take years of testing before they reach everyday life, and some never do.
Why scientists care
Science builds on earlier work, so references that point to nothing weaken that chain of trust. The study shows the problem isn’t limited to medicine, where the same team found it earlier, and that it is growing fast in the fields closest to AI and engineering.
Keep in mind
The study did not prove where the fake references came from; AI tools are a suspected cause, not a confirmed one. Only references with a DOI were checked, about 14% of sampled flags were not actually fake, and the 8,500 figure is an estimate. We read the press releases, not the full paper.
How far along is the science?
- Peer-reviewedReviewed by independent experts for a journal before publication. Peer review is a quality check, not a grade of evidence strength: confidence still depends on study design, limitations, replication and the wider evidence.
- PreprintA manuscript shared publicly, often before or alongside journal review. Check its separate journal and review status (each story says what we know). Results may change.
- Conference talkPresented to other scientists at a meeting. Usually short and early; a full paper may come later.
- Agency reportPublished by a science agency such as NASA, NOAA or NIH. Reviewed internally, but not by an outside journal.
- Press releaseAn announcement from a university or company. Useful, but always check the research it describes.
How soon could it reach you?
- Could reach you soonCould show up in products, services or advice soon. Our rough editorial guess is about 1 to 3 years, not a forecast from the research.
- A few years outPromising, but needs more testing or engineering first. Our rough editorial guess is several years, with a lot of uncertainty.
- Long-term scienceBasic research that builds knowledge. Real-world uses, if any, are far off.
- Small stepA small update to earlier work. Useful to researchers, but not a change you would notice yet.
Every story links to its original source: the paper itself, or the conference program when no paper is out yet. Explanations are written with AI, usually from the paper’s abstract (each story says what it was written from), then checked against the source. We only use journals, recognized preprint servers, science agencies, scientific conferences and established science news outlets.