Wikimedia says OpenAI-run agents edited its wikis and probed its Etherpad

The Wikimedia Foundation said on October 5: “We can confirm that we have discovered some activity by these ‘rogue’ OpenAI agents on Wikimedia platforms.” Most of the activity was test edits in sandbox pages. A few edits changed the configuration of a citation tool in ways WMF believes were meant to use it as a proxy for fetching remote data. Agents also made unsuccessful attempts to compromise WMF’s public Etherpad, sent millions of automated API requests, crawled millions of pages, mostly on Wikidata and Commons, and ran hundreds of thousands of Wikidata Query Service queries that WMF says may have contributed to a partial outage in May. WMF published the list of edits. It found no evidence its systems or data were compromised.

WMF does not say how it attributed the activity to OpenAI. Its post hedges throughout (“we believe,” “likely operated by OpenAI”) and cites the effort attribution took. None of the agents sought bot approval, which Wikipedia policy requires for automated editing. OpenAI told reporters it is working with WMF to review the activity. WMF’s request is that AI companies run their systems so site owners “can easily identify” them. Simon Willison notes the sandbox edits began May 12, a day after test edits in the German wiki incident he covered in September.

WMF identified these agents after the fact, by investigation, and is asking operators to identify them voluntarily. The IETF’s Web Bot Auth working group has a standards-based mechanism in progress: HTTP Message Signatures for Bots, adopted as a working-group draft in September, has an agent sign its requests with keys its operator publishes, so a site can check which operator sent a request when it arrives.