arXiv petition seeks author control over AI bot scraping

An unsigned petition hosted on an MIT math faculty page is asking arXiv to add a submission-time toggle that would let authors decide whether AI bots can scrape their full-text PDFs and LaTeX source, or only the title, abstract and citation metadata.

The grievance, in the petition's own words: "authors posting to arXiv have no control over whether bots scrape their full-text PDFs," and "open access for humans has become a free pass for commercial AI models to harvest our research." The core ask is that arXiv "implement a simple permission toggle during submission, allowing authors to decide what automated scrapers can access."

Two tiers are proposed. Full Access lets bots scrape complete text and source files. Metadata Only restricts them to title, abstract and citation information.

Signatures are verified only through institutional email, meaning university, research institute or laboratory domains, and the page notes that "Personal webmail (Gmail, Outlook, Yahoo and similar) and disposable addresses are not accepted." No organiser is named, and the page lists no signature count. "Open science for human readers should not mean mandatory, unconsented data harvesting," the petition says. A couple of the researchers we follow were already passing the link around when we picked it up.