AI Text Flood Makes Gatekeeping Essential for Science
Large language models are producing so much text that institutions, including courts, schools and scientific preprint servers, are struggling to cope. arXiv has seen submissions double in two years and will now limit authors to two per month. The article argues that some gatekeeping is necessary to keep human verification in the loop, especially as AI can also threaten verification tools.
For decades, online culture pushed back against those who controlled access, often treating such control as censorship or elitism. Now, text-generating systems are reversing the pressure: courts, schools, and local offices face more material than they can process, because older constraints such as time, skill, and access no longer limit output.
Science faces similar strain. arXiv, a preprint server, has seen its incoming manuscript count grow twofold over two years and will restrict each author to two monthly uploads. These papers do not undergo formal peer review, yet a modest crew maintains baseline checks. In mathematics, researchers have asked for additional colleagues to help, while AI attacks could imperil software that validates proofs.
The shift could affect scientists, editors, court staff, teachers, and citizens who rely on institutional decisions. Tighter submission rules may reduce spam and preserve human review, but they could also slow legitimate work or disadvantage researchers with fewer resources. If verification tools become vulnerable, public trust in scientific and official claims may depend more on transparent human oversight. The outcome may hinge on how carefully limits are designed and explained.