ArXiv Imposes Strict Penalties for AI-Generated Falsified Preprints

The Growing Problem of AI-Generated Slop in Research

Across the scientific publishing landscape, a troubling trend has emerged: artificial intelligence tools are being used to produce content that appears scholarly but is fundamentally flawed. This includes fabricated citations, unedited outputs from AI prompts, and diagrams that make no logical sense. These issues have repeatedly slipped past the scrutiny of editors and peer reviewers, raising questions about accountability. Until now, consequences for those who submit such material have been largely unclear.

Fake Citations and Nonsensical Diagrams

Examples of AI-generated slop include references to nonexistent papers, incorrect mathematical relationships presented as if they were verified, and image-based data that are entirely invented. These products not only waste the time of reviewers but also pollute the scientific record, potentially misleading future researchers. The phenomenon has become so widespread that some fields are now taking preemptive action to address the challenge before the formal peer-review process even begins.

ArXiv’s New Enforcement Policy

One of the most notable responses comes from the preprint server arXiv, a central repository for physics, astronomy, and related disciplines. According to a senior member of arXiv’s editorial advisory council, the organization is implementing a two-tier penalty for those who submit AI-generated hallucinations in their preprints.

Following detection of such submissions, arXiv intends to impose a one-year ban on the responsible submitter. During this period, they will be prohibited from uploading any new preprints. Furthermore, after the ban ends, any future submissions by the same individual will be required to undergo a formal peer-review process before arXiv will host them. This is a permanent condition, meaning that previously unrestricted posting privileges are revoked indefinitely.

One-Year Ban and Permanent Peer Review Requirement

The policy was outlined in a social media thread by Thomas Dietterich, an emeritus professor at Oregon State University. Dietterich serves on both arXiv’s editorial advisory council and its moderation team, giving him direct insight into the organization’s procedures. While arXiv leadership has not yet officially confirmed the policy — the original article’s author is awaiting a response — Dietterich’s statements are considered authoritative given his roles.

Who Announced the Policy?

Thomas Dietterich is a well-known figure in the machine learning community, having contributed extensively to the field. His involvement with arXiv places him in a position to understand and communicate its enforcement strategies. The announcement signals that arXiv is taking a proactive stance against the misuse of AI in scholarly communication, rather than waiting for problems to be corrected after publication.

It is important to note that the policy targets only inappropriate AI-generated content — that which contains fabricated information or obviously unsupported claims. Legitimate uses of AI for assistance, such as language polishing or figure generation under human oversight, are likely not affected.

Implications for the Scientific Community

This move by arXiv could set a precedent for other preprint servers and even peer-reviewed journals. By imposing consequences before the formal review stage, arXiv aims to deter researchers from submitting AI-generated slop in the first place. The permanent requirement of peer review for former offenders adds a significant deterrent: individuals who rely on the rapid dissemination of preprints will face a substantial loss of privilege.

Return to top: The growing problem. The policy also raises questions about detection. How will arXiv identify AI-generated hallucinations? Likely through a combination of human moderation, community reports, and automated screening tools. Given the large volume of submissions, consistent enforcement will be challenging, but the announcement sends a clear signal: the free-for-all of AI-generated slop is no longer acceptable.

For researchers, this development reinforces the need for responsible use of AI tools. Automated language models can be powerful aids, but they are not a substitute for genuine scholarship. Scientists must ensure that every claim in a preprint is verifiable, every citation is real, and every figure accurately represents data. The arXiv policy, if applied consistently, will help preserve the integrity of the preprint archive and the broader scientific discourse.