Scientists Warn AI Slop Is Wreaking Havoc in the Research World

Why It Matters

Bogus citations erode trust in the scholarly record and can misdirect future research, prompting repositories to enforce stricter AI‑content controls.

Key Takeaways

• 146,900 fake citations detected across four major repositories
• AI hallucinations surged after large language model adoption
• Problem spans many papers, not isolated incidents
• arXiv will ban authors with unchecked AI-generated citations
• Trust in scholarly record erodes, affecting peer review

Pulse Analysis

The rapid diffusion of large language models has given researchers a convenient shortcut for drafting reference lists, but it also introduces a new class of error known as AI hallucination. When a model fabricates a plausible‑sounding title, authors may insert it without verification, inflating citation counts with non‑existent works. Recent data shows that after 2023, the frequency of such phantom references jumped dramatically, reflecting both the growing reliance on generative AI and the lack of systematic checks in manuscript preparation.

This hidden contamination threatens the integrity of the entire research ecosystem. Scholars depend on accurate citations to trace prior art, assess methodological soundness, and build upon validated findings. Fake references can skew bibliometric indicators, mislead systematic reviews, and even affect downstream technologies that mine scholarly literature for patents or policy guidance. As the volume of polluted citations climbs, peer reviewers and editors face an uphill battle to distinguish genuine scholarship from AI‑generated noise, risking a gradual erosion of confidence in peer‑reviewed literature.

Publishers and pre‑print servers are beginning to push back. arXiv’s recent policy to ban authors who submit papers with unchecked AI‑generated citations signals a broader move toward accountability. Automated reference‑checking tools, cross‑database verification services, and mandatory disclosure of AI assistance are emerging as practical safeguards. Institutions are also training researchers to treat AI as an assistive tool rather than a source of truth. If these measures gain traction, the community can curb citation fraud, preserve the credibility of scientific communication, and ensure that AI augments rather than undermines scholarly rigor.