ArXiv Implements Strict One-Strike Policy Against AI-Generated ‘Slop’
- arXiv is implementing a one-year ban for authors who submit research papers containing clear evidence of unverified AI-generated content.
- The policy specifically targets "hallucinated" citations and remnants of LLM-generated text, shifting the burden of accuracy entirely onto the authors.
- While the use of AI tools remains permitted, authors must now guarantee the integrity of their work or risk mandatory peer-review verification for all future submissions.
Maintaining Scientific Integrity in the Era of Generative AI
As large language models (LLMs) continue to permeate every facet of academic writing, arXiv—the preeminent open-access repository for preprint research—is drawing a hard line. The platform has officially announced a new enforcement policy aimed at curbing the proliferation of low-quality, AI-generated scientific papers that threaten the credibility of research dissemination.
The Policy: Accountability Over Prohibition
Thomas Dietterich, the chair of arXiv’s computer science section, clarified that the platform is not banning the use of AI tools entirely. Instead, the initiative focuses on the necessity of authorial accountability. The core mandate is clear: authors must take full responsibility for their submissions, regardless of the tools used to produce them. The shift addresses a growing trend of researchers submitting work containing hallucinations, fabricated citations, and AI-specific artifacts without performing basic due diligence.
Defining ‘Incontrovertible Evidence’ of Negligence
The new enforcement protocols focus on identifying submissions that clearly demonstrate a lack of human oversight. According to arXiv’s leadership, evidence of negligence that could trigger a ban includes:
- Hallucinated References: Citations or academic sources generated by an LLM that do not exist.
- LLM Artifacts: The presence of prompts, disclaimers, or metadata left behind by LLMs (e.g., “As an AI language model…”).
- Unfiltered Inaccuracies: The inclusion of plagiarized, biased, or demonstrably incorrect technical data that a human reviewer should have easily identified.
Consequences and Enforcement
The policy operates under a strict “one-strike” rule. If moderators flag a submission for containing the aforementioned evidence, and section chairs confirm the findings, the authors face a significant penalty: a one-year suspension from the platform. Following the suspension, any subsequent submissions will require proof of acceptance from a reputable, peer-reviewed venue before arXiv will host the preprint.
While this approach is stringent, arXiv has built in a safety mechanism allowing authors to appeal these decisions. This ensures that in the case of a false positive, researchers have a path to contest the findings.
Why This Matters for the Scientific Community
The rise of “AI slop” in scientific literature has become a major concern, particularly in fields like biomedicine, where fabricated citations can lead to the spread of medical misinformation. By forcing authors to manually verify their outputs, arXiv is attempting to preserve its status as a trusted source for cutting-edge research. As arXiv transitions into an independent nonprofit, these stricter governance policies represent a broader shift toward greater professional responsibility in the age of automated writing tools.
