- Retracted papers reveal how science corrects its own mistakes.
- Wakefield's vaccine-autism fraud still fuels misinformation decades later.
- AI-generated images create new verification challenges for journals.
Andrew Wakefield had twelve children, a hypothesis, and a paycheck from lawyers who needed medical cover for a lawsuit.
His is one of the most notorious scientific misconduct examples in modern history. The 1998 paper he produced in The Lancet claimed the MMR vaccine triggered autism through an unspecified gut disorder. It was built on selection bias, adult reference values applied to children, and financial conflicts that would take journalist Brian Deer thirteen years to fully document.
Wakefield's paper lasted twelve years before retraction. The damage has lasted decades longer.
This entertaining video from the SciShow Youtube channel lists more lies - we dig into some of them here.
A Rat, a Protein, and a Blog Post
Wakefield's fraud was deliberate. Other failures arrive through newer, stranger doors.
In February 2024, a review paper on the JAK/STAT signalling pathway in rat sperm stem cells appeared in Frontiers in Cell and Developmental Biology. The text was unremarkable. The figures were not.
Key figure
3 days
Time between publication and retraction of the AI-generated rat anatomy paper
The authors had used generative AI to create their illustrations, producing anatomical images so grotesquely wrong that the paper became an instant cautionary meme. It survived three days.
The journal's policy permitted AI-generated figures if disclosed. The deeper worry, as SciShow host Reid Reimers notes in his survey of notorious retractions, is what happens when AI-generated errors are subtler.
Not every fake image announces itself with impossible anatomy.
The Fraud That Misdirected Alzheimer's Research
Some fabrications hide for years. In 2006, a team published a paper in Nature identifying a specific amyloid protein called Aβ*56, which they claimed caused Alzheimer's symptoms in mice. The finding offered a concrete target for drug development.
The field seized on it.
Amyloid plaques
Clumps of misfolded proteins that accumulate around nerve cells in the brains of Alzheimer's patients. Whether these plaques cause the disease or merely accompany it remains one of neuroscience's most contested questions.
Nobody could replicate the finding. For sixteen years, Aβ*56 remained elusive, while the original paper accumulated nearly 2,500 citations.
Key figure
2,500
Citations accumulated by the fraudulent Aβ*56 paper before its 2024 retraction
In 2022, neuroscientist Matthew Schrag at Vanderbilt University found evidence that gel electrophoresis images in the paper had been manipulated. Protein bands appeared to have been copy-pasted to fabricate results.
The paper was retracted in June 2024 with the agreement of all co-authors except one. The amyloid hypothesis itself remains contested, not debunked, but the field lost years chasing a protein that may never have existed in the form described.
P-Hacking and Superconductor Dreams
Cornell food psychologist Brian Wansink built a career on catchy findings about eating behaviour. Shop hungry, buy junk food. Use smaller plates, eat less. In 2016, he wrote a blog post advising graduate students to keep reanalysing data until something statistically significant emerged.
Statisticians recognized the technique immediately. It has a name: p-hacking. Run enough tests on the same dataset and random noise will eventually clear the significance threshold. The more tests, the more false positives.
Scrutiny of Wansink's published work revealed inconsistencies across multiple datasets. He resigned from Cornell in 2019 after eighteen papers were retracted.
These papers were retracted, which means the system is working. The bad science gets called out and cleared away.
Reid Reimers, SciShow
Fabrication crosses disciplines, too.
In 2023, a team claimed to have created a superconducting material that worked at room temperature and relatively low pressure. The paper was retracted months later after a University of Rochester investigation found fabricated data, falsified results, and plagiarism.
The response was instructive: in 2024, physicists convened an entire conference dedicated to preventing fraudulent superconductor research.
The Study That Was Faked, Then Done Properly
Political scientist Michael LaCour published a 2014 paper in Science with his advisor Donald Green, claiming that gay canvassers could change minds about marriage equality through brief doorstep conversations.
Graduate students at other universities tried to build on the work. They discovered the survey company LaCour cited did not recognise him. When Green asked for the original data, LaCour said he had accidentally deleted it. Green requested a retraction.
But those same sceptical graduate students, David Broockman and Joshua Kalla, designed their own version. They examined conversations about transgender rights in South Florida with genuine follow-up surveys over three months.
Their results showed that both transgender and nontransgender canvassers could reduce transphobia through direct conversation, and the effect persisted. The fabricated study was gone. The replacement confirmed something more interesting: honest conversation changes minds, regardless of who initiates it.
The Mechanism Works, on Its Own Schedule
Six cases across six disciplines. An anti-vaccine hoax, AI-generated nonsense, manipulated Alzheimer's images, statistical fraud in food psychology, fabricated superconductor data, and invented survey results. Each paper passed peer review. Each was eventually caught.
The timeline varies wildly. Some retractions take days, others take decades. More than 10,000 papers were retracted in 2023 alone, a record that suggests the system is working harder than ever, though whether that reflects better detection or more fraud remains genuinely unclear.
As we have previously reported, scientific fraud has evolved into an industrial enterprise that threatens the foundations of peer review.
More On Lies
AI Hallucinations Are Inevitable, OpenAI Researchers Prove
OpenAI researchers prove AI hallucinations are mathematically inevitable. Nine out of ten benchmarks reward guessing over admitting uncertainty.
→By 2026, AI-generated content has become what researchers at Frontiers in Research Metrics call a "dominant vector of misconduct." The rat anatomy paper was obvious. The next generation of AI-fabricated figures may not be.
Schrag, the Vanderbilt neuroscientist who unravelled the Alzheimer's fraud, keeps investigating. He has since flagged concerns about additional papers from the same laboratory. The self-correcting mechanism depends on people like him, scientists willing to spend years verifying what their colleagues claimed to have found.
The system works. It just requires someone stubborn enough to check.
Sources
- Primary Source: These Are The Worst Research Papers Of All Time (SciShow, 2024)
- Additional Context:
- Potential fabrication in research images threatens key theory of Alzheimer's disease (Science)
- More than 10,000 research papers were retracted in 2023 (Nature)
- Cornell nutrition scientist resigns after retractions and research misconduct finding (Science)
- Durably reducing transphobia: A field experiment on door-to-door canvassing (Broockman & Kalla, Science, 2016)
