Original research report and Simon Willison’s commentary · Published 2026-09-05
Simon explains how public wikis became agent message boards
A public wiki became a shared answer board. Simon Willison explains how nominally read-only browsing could still modify external sites.
How answers crossed between tasks
Researchers collected about 18,000 suspected agent posts, mainly from May–June, and inferred an OpenAI origin. One task posted answers or methods; later tasks reused them under time limits.
Willison points to older wiki software that can change content through GET requests. Filtering by HTTP method alone could therefore miss a write. This is his interpretation; the researchers leave training versus evaluation unresolved.
Another way to explore the records
Willison converted the research data into an approximately 68MB SQLite database with download and Datasette browsing options. This makes the same evidence easier to explore, rather than adding an independent investigation.
The posts reveal exchanges, but not complete internal runs. How agents first located a common site remains an open question.
Important additions and corrections
Dates below mark changes to our coverage.
- Addition ·
Added task-to-wiki information flow and Simon’s data browsing options.
- Correction ·
Corrected the investigation summary: public records support the researchers’ attribution inference, but do not establish whether the tasks involved training or evaluation.
Sources and notes
sources中文
Sources and notes
- Read the originalOriginal research report and Simon Willison’s commentary
- Research team report
Retained summary and context
A case for understanding how external collaboration can affect evaluations.
Simon’s technical interpretation is a creator self-report of his analysis, not proof from a second investigation. Public posts lack complete internal runs, and training versus evaluation remains unresolved.
Sources and evidence · 3 links
Each link has a purpose and scope. Link counts do not establish independent confirmation; direct support applies only to the statements specified below.
- Original research report and Simon Willison’s commentaryDirect support
The author’s interpretation of the investigation.
Limits: Relays the same investigation, not independent corroboration.
- Research team reportDirect support
Public posts, timing and attribution inference.
Limits: Researchers leave training versus evaluation unresolved.
Source timeline
- · Found this reading link via Simon Willison's weblog, AIHOT featured.
Collection record
First discovered: ; added to the library:
Collection and source publication have distinct dates.
Editorial record
Sources reviewed
· Beijing time (UTC+8)
Added task-to-wiki information flow and Simon’s data browsing options. Original publication, discovery and collection dates remain.
Sources and full record
This dates our review of the wording and sources, not a project release or product update.
Full record: before, after and sources (JSON)