Every record, in the open

What the dataset contains

2,026 documented incidents at 539 U.S. colleges and universities, spanning 1969–2026, containing 4,392 alert messages, of which 3,803 (87%) are confirmed word-for-word from a cited source. Alongside the cases sit 339 published alert-and-warning policy records. Each case documents one incident: the institution, the timeline, every alert message in sequence, casualties where reported, sources for every claim, and an explicit confidence rating.

The distinction the archive never blurs: messages marked verbatim were confirmed against the cited source, exactly as sent, typos preserved. This edition never publishes paraphrased or reconstructed alert text: where a message's exact wording isn't confirmed, its text field is simply empty. The full collection and verification pipeline is documented on the methodology page.

Scope and limits

This is an AI-assembled research collection, not a census and not a representative sample of US campus alerting. Cases enter the archive because web sources documented them well enough to verify, which over-weights recent, serious, and heavily covered incidents. Nothing in the dataset estimates how colleges and universities perform in general, and aggregate figures derived from it describe this collection only: please do not cite them as sector statistics.

The collection records what was preserved, not everything that was sent. A message, follow-up, or all-clear that left no public trace is invisible here, so any count of “no documented X” is a documentation floor: it cannot distinguish “never sent” from “sent, but not preserved anywhere we can cite.”

Two record types should never be pooled: emergency notifications (34 CFR 668.46(g)) and timely warnings (668.46(e)) answer different legal obligations with different timing expectations, and every case carries its category label so analyses can keep them apart.

Version

v2026.07 · released July 2026 · 2,026 cases at last build

Versions name monthly snapshots (the archive grows continuously, so the case count can rise within a version). Downloads always carry the version and a generation timestamp in their wrapper, so any copy can be traced to its snapshot.

How to cite

To cite the whole archive, use the citation below (add your access date). To cite a single incident, every case page carries its own “Cite this case” card with a per-case citation, BibTeX, and the canonical JSON record.

@misc{campusalertarchive-2026.07,
  title        = {{Campus Alert Archive: a living archive of verbatim U.S. campus emergency alerts}},
  author       = {{Campus Alert Archive}},
  year         = {2026},
  howpublished = {Campus Alert Archive},
  note         = {Version 2026.07, 2,026 cases; verbatim alert texts remain the work of their issuing institutions},
  url          = {https://campusalertarchive.com/}
}

Downloads

Direct, static files: no account, no API, no rate limits. For filtered subsets, chosen columns, or Excel/PDF, use the export menu on the search page (it builds files in your browser from these same artifacts).

License

Compilation, metadata, summaries, and annotations © Campus Alert Archive, licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). Verbatim alert and policy texts remain the work of their issuing institutions, reproduced from publicly available sources, and are not covered by the CC BY license.

In short: share and adapt the compilation freely with credit (CC BY 4.0). The alert messages themselves are federally mandated public-safety communications, reproduced for research and education; see terms for the full attribution and fair-use statement.

Integrity notes

• Alert texts are preserved exactly as sent, typos and all. Nothing in the confirmed texts is cleaned up or sanitized. This public edition includes only cases with at least one confirmed-verbatim alert; messages whose exact wording is not preserved remain in each case's timeline with an empty text field; no unconfirmed wording is published.
• The per-case files at /data/case/<slug>.json are the published edition of each case record; the flat CSV/spreadsheet exports are projections of them.
• Every record carries per-claim sources and an explicit confidence rating (high: verbatim from an official source; medium: reliable secondary sourcing).
• CSV artifacts only: cells beginning with =, +, @, or a tab are prefixed with a single quote to prevent spreadsheet formula injection. JSON is never altered.
• Case URLs are permanent. A case is corrected in place when better sourcing emerges; its lastUpdated stamp moves, its URL never does.
• How cases enter the archive, the validation gates, and who does what (including the AI systems involved) are documented end to end on the methodology page.