AI Does Not Have to Invent a Study to Poison the Scientific Record

According to the source material, science has a new problem, and its name is slop. A Lancet study published this week found that fabricated citations—references to non-existent academic papers—have surged sixfold in research literature since 2023, with 1 in 458 papers now containing at least one fake entry. The culprit? Generative AI tools, which researchers suspect are either hallucinating references or being misused to pad bibliographies.

Columbia University’s team analyzed over 2 million papers and 97 million citations, identifying around 4,000 fabricated entries in 2,800 papers. The trend isn’t just a statistical blip; it’s a full-blown contamination of the scientific record.

The implications stretch beyond academia. Fabricated citations risk seeping into systematic reviews and clinical guidelines, where flawed references can distort medical decisions. As one researcher quoted in the study put it, “We have all of these papers coming out trying to track the use of [large language models] in science, but none of them really tell us anything about the quality.

This is one of the first papers that’s telling us something about the quality of what’s being produced with LLMs, and it’s a signal of slop.” The question isn’t whether AI is making science faster—it’s whether it’s making it worse.

The source material also shows that the study’s numbers are stark. In 2023, 1 in 2,828 papers contained fabricated references. By 2025, that figure had ballooned to 1 in 458, and in the first seven weeks of 2026, it climbed to 1 in 277. The acceleration suggests a systemic issue, not just isolated incidents.

While the Lancet study doesn’t name specific AI tools, the pattern aligns with the rise of research assistants like Elicit, Scite, and even general-purpose models like ChatGPT. These tools are designed to streamline literature reviews, but their tendency to hallucinate references—whether due to flawed training data or user misuse—is now leaving a measurable scar on academic publishing.

The problem isn’t just about bad citations. It’s about trust. Peer review relies on the assumption that references are real, verifiable, and relevant. When that assumption breaks down, the entire edifice of scientific validation wobbles. Some journals are already experimenting with AI detection tools to flag suspicious citations, but the cat may already be out of the bag. As one researcher admitted, “I was deeply embarrassed: I checked for that, and it still almost happened to me.” The embarrassment isn’t just personal—it’s institutional.