| When: | Between September 1 and September 27 (2026) |
|---|---|
| What: | We observe repeated searches for historical newspaper coverage across five archives in 201 September urlquery reports. |
| Who: | Attribution to agents is based on use of urlquery.net to request data, which is known agent-infrastructure, as well as intermediary services to convert webpages into a format suitable for AI. However, the activity could also be human-authored automation. The model provider and operator are unknown. |
| Assessment: | A possible automated research workflow with no identified benchmark match. The activity is consistent with a historical data retrieval evaluation. |
Timeline
September 1, 2026 – September 27, 2026
Observations
As recently as September 27, we observed potential agent activity on the public website-scanning service urlquery. The task centers around retrieving data from historical news publications. We observe 201 urlquery reports involving requests to: Newspapers.com, the British Newspaper Archive, Trove (the National Library of Australia), Findmypast, and Chronicling America (the Library of Congress).
The requests include one test probe and searches by person, publication, as well as attempts to retrieve newspaper pages, images and OCR-extracted text. The largest search groups are:
- Annie Besant: 40 reports. Searches include the terms Southport Visiter as well as June 12, 1906, alongside broader date searches.
- Billy Carter: 35 reports. A September 18 report contains three Jina requests (a service converting webpages into text suitable for AI tools) for a Newspapers.com search with terms Carter and October 1, 1980.
- Brooklyn Eagle searches: 30 reports. Searches include terms July 28, 1889 and several dates in 1900.
This preserved Carter report, for example, records HTTP 403 responses from Jina.
Activity by archive
Urlquery reports by date and archive.
Activity by request category
Labels group requests by topic or endpoint. Page/image/OCR requests are attempts to retrieve specific pages, thumbnails or OCR text without a named search term in the URL. OCR term searches include “weather,” “Canada,” “temperature” and “154.”
Attribution
We are moderately confident that this represents agent activity, based on the volume, regularity, and common tactics across different data targets. The workflow uses urlquery, which has previously been observed in agent activity, and some requests use Jina to convert webpages into text suitable for AI tools. However, this is also consistent with human-authored automation, and we cannot with certainty confirm AI involvement or identify a model provider or operator.
Assessment
This activity may represent a previously undocumented evaluation family and is not known to be in public benchmarks. We searched for this task in public bechmarks, including DeepSearchQA and Humanity's Last Exam, but cannot be certain that the task is absent from all benchmarks or that it belongs to a private evaluation.
Links
Selected public urlquery reports (report dates are in September 2026):
- September 18: Billy Carter search through Jina, with HTTP 403 responses.
- September 22: Besant search in Findmypast, restricted to the Southport Visiter on June 12, 1906.
- September 22: Brooklyn Eagle browse request for July 28, 1889.
- September 22: Brooklyn Eagle phrase search for “Raw Sponges,” dated July 28, 1889.
- September 22: Request for OCR text from a newspaper page.
- September 27: Annie Besant search on Newspapers.com, restricted to 1906.