ArXiv enforces policy against papers generated with unchecked AI, implementing year-long ban
The preprint platform will ban researchers for one year if papers contain evidence of unverified LLM output, requiring future submissions to clear peer review before resubmission.
1 source · cross-referenced
- ArXiv announced a formal enforcement mechanism for papers containing unchecked AI-generated content, specifically hallucinated references and unvetted LLM text, resulting in a one-year ban followed by mandatory peer-review acceptance before future submissions.
- Thomas Dietterich, chair of ArXiv's computer science section, clarified that authors bear full responsibility for all paper contents regardless of generation method, and that incontrovertible evidence of author negligence triggers the penalty.
- Examples of actionable violations include hallucinated citations, leftover meta-comments from LLMs, and placeholder text intended for manual completion that was never filled in.
- Authors can appeal ban decisions, and ArXiv will require moderator documentation and section chair confirmation before imposing penalties.
- This follows a 2024 policy restricting computer science review articles and position papers to peer-reviewed venues, citing a flood of low-effort LLM-generated content.
ArXiv, the widely-used platform for scientific preprints, has formalized enforcement of a policy against papers submitted with unverified AI-generated content. According to Thomas Dietterich, the section chair overseeing ArXiv's computer science division, papers containing what he termed "incontrovertible evidence that the authors did not check the results of LLM generation" will trigger a one-year ban, after which authors must have any subsequent submissions accepted at a peer-reviewed conference or journal before ArXiv will host them.
Dietterich outlined specific criteria for what constitutes unverified AI use: hallucinated references, left-over LLM system prompts, and placeholder instructions embedded in the final text—such as "fill this table with real experimental numbers" or "would you like me to revise this?"—that were never replaced. The policy pins responsibility squarely on authors, regardless of how the content originated.
The enforcement mechanism involves a two-step review process: a moderator documents suspected violations, and the section chair must independently confirm the finding before any ban is imposed. Dietterich indicated that researchers may appeal decisions, and enforcement applies only to cases meeting the threshold of incontrovertible evidence.
This action extends policies ArXiv introduced last year, which barred unvetted computer science review articles and position papers from the platform. The platform justified that change by noting the ease with which large language models enable rapid production of low-substance papers that amount to little more than padded bibliographies without original research insights.
- May 19, 2026 · arXiv cs.AI
AgentWall introduces runtime safety layer to intercept and control local AI agent actions
Trust79 - May 15, 2026 · Ars Technica
Zero-day BitLocker bypass lets attackers with physical access decrypt Windows 11 drives instantly
Trust65 - May 11, 2026 · arXiv
Large language models demonstrate capability for text-in-text steganography
Trust69