A young economist’s bold claim about artificial intelligence revolutionizing scientific discovery captured the imagination of policymakers and scholars alike. But what seemed like a breakthrough in understanding AI’s role in the workplace turned out to be built on sand.
Aidan Toner-Rodgers, a 27-year-old once hailed as a rising star in MIT’s Ph.D. program, authored a paper that promised real-world evidence of AI boosting innovation—only for the entire study to unravel as a complete fabrication.
The paper, titled “Artificial Intelligence, Scientific Discovery, and Product Innovation,” landed like a revelation in early 2025. It described an experiment at a major materials science lab where an AI tool reportedly accelerated the discovery of new compounds and spurred patent filings. According to Toner-Rodgers, the system slashed research time while increasing output, offering a glimpse of how AI could transform high-stakes industries without gutting jobs.
The work drew quick attention: citations in congressional hearings, features in outlets like The Wall Street Journal, and nods from top economists eager for data in the AI boom. It painted a picture of technology as a net positive, one that enhanced human ingenuity rather than replacing it.
That narrative crumbled under scrutiny. Charles Elkan, a computer science professor at the University of California, San Diego, first raised red flags in a detailed blog post. He pointed to glaring inconsistencies: the AI described was years ahead of commercially available tech, and the setup—hundreds of R&D teams running identical randomized trials—sounded more like a textbook ideal than a corporate reality.
“Why would a large company like this take such pains to run a randomized trial on its own employees… only to anonymously give this data to a single researcher from MIT?” Elkan wrote, questioning the “academic integrity” of the claims.
MIT’s response was swift and damning. By spring 2025, an internal review by the Committee on Discipline concluded the data was unreliable. In a stark press release, the economics department declared it had “no confidence in the provenance, reliability or validity of the data and has no confidence in the veracity of the research contained in the paper.” The paper was pulled from arXiv and withdrawn from submission at The Quarterly Journal of Economics. Toner-Rodgers, no longer enrolled in the program, faced expulsion.
Digging deeper reveals a web of deceit that goes beyond sloppy science. Toner-Rodgers allegedly named two industry giants, 3M and Corning, as partners in the study—though he anonymized them in the paper. Both companies issued firm denials: neither had rolled out the described experiment nor shared any data with him.
When questions mounted, he reportedly registered a fake domain, corningresearch.com, to fabricate a data-use agreement after the firm balked at his requests. Data across drafts shifted suspiciously, with “neat” results that screamed invention rather than observation. As one analyst noted in a Substack breakdown, real corporate labs don’t operate with such uniformity; they chase profits, not perfect econometric models.
Toner-Rodgers has since downplayed the fallout to peers, insisting it stemmed from “data rights” disputes—not outright fraud. He claims he accessed legitimate data from a materials firm but forged the agreement when they pulled back. Yet evidence piles against him: changing figures, nonexistent tools, and a pattern that echoes broader worries in academia.
This isn’t an isolated slip. Recent scandals, from manipulated images in high-profile journals to retracted COVID studies, show how the rush for prestige can erode standards. In AI research especially, where peer review lags behind the hype machine, one fabricated paper can ripple through policy debates, misleading lawmakers on everything from job protections to R&D funding.
The MIT economics department, shaken by the breach, is now overhauling its safeguards. Faculty are pushing for mandatory raw data reviews on grad papers and more rigorous vetting of external collaborations. Students, too, are adapting—proactively sharing audit trails to prove their work’s legitimacy. It’s a reminder that in fields driven by trust, one bad actor can poison the well for everyone.
What makes this case sting is the opportunity lost. Genuine AI applications in materials science could unlock real advances, from stronger batteries to cleaner manufacturing—innovations that bolster American competitiveness without the ethical shortcuts. Instead, Toner-Rodgers’s stunt fed the frenzy around unproven tech, potentially diverting resources from honest inquiry.
One can’t help but wonder: in an era where elite institutions churn out “groundbreaking” findings at breakneck speed, how many more will slip through until the system demands ironclad proof over polished narratives? For now, the lesson stands clear—extraordinary claims demand extraordinary evidence, not just a clever draft.
