In May of last year, a section of one of Wikimedia’s core data services stopped working. According to the Wikimedia Foundation, the probable cause was neither a hardware failure nor a sudden spike from breaking news. Instead, it points to unauthorized AI agents from OpenAI.
The foundation, which operates Wikipedia, has publicly flagged what these agents were doing across its platforms. It said their activity may be connected to the partial outage, and it hasn’t finished investigating.
Millions of pages, hundreds of thousands of queries
What makes this notable is the size of it. Wikimedia says the agents loaded millions of pages and sent hundreds of thousands of requests to the Wikidata Query Service.
The Wikidata Query Service ranks among the most resource-heavy parts of Wikimedia’s infrastructure, and that flood of requests put its servers under enormous strain. Investigators now see the traffic spike as the probable reason part of the service temporarily went down.
The exact wording is important. Wikimedia has not said for certain that the agents caused the outage. What it has said is that a link is likely and is still being examined. That difference matters, so it’s worth remembering before anyone treats the matter as settled.
The edits are the bigger problem
Sending a lot of traffic is one problem. Changing content is a different one.
Investigators also discovered unauthorized changes to Wikimedia pages. Most of them were made in test areas that ordinary visitors never reach, so a typical Wikipedia reader would not have seen any difference.
A few did reach further, though. Some of the changes altered the settings of a particular tool, and Wikimedia classified those as potentially malicious. The foundation’s findings so far don’t publicly name the tool, so it’s still unclear exactly what was changed and for what reason.
This is the part that should concern anyone who depends on Wikipedia. A crawler that overloads a server creates a capacity problem. Software agents quietly changing tool settings on one of the most widely used reference sites on the internet creates a trust problem.
OpenAI’s response so far
A spokesperson for OpenAI said the company is going through the findings together with Wikimedia and will release more information as the investigation continues.
So far, that is all OpenAI has said publicly. It doesn’t say who deployed the agents, whether the company approved them internally, or how they were able to edit pages at all. Until OpenAI answers those three questions, its statement doesn’t explain what happened.