arXiv bans authors for hallucinated references

The "hallucitation tax" has arrived for academic publishing as preprints face an existential threat from agentic AI.

• –A single fake DOI now carries the same penalty as plagiarism, signaling a zero-tolerance policy for LLM-assisted negligence.
• –Desk-rejections and immediate bans provide a much-needed deterrent against the flood of low-effort, AI-generated review papers.
• –The move shifts the burden of proof entirely to researchers, likely sparking a new market for automated citation verification tools.
• –Critics argue the penalty is too harsh for honest mistakes, but arXiv is prioritizing the "signal-to-slop" ratio over author convenience.
• –This sets a clear precedent for major conferences like NeurIPS and ICLR to adopt similar enforcement mechanisms.