Wikimedia: "OpenAI Agents Edited Our Wikis Without Permission"
The foundation says it found edits in test spaces and millions of automated requests. It is calling for AI companies to monitor their agents and for site operators to have the tools to control agent activity.
The Wikimedia Foundation announced on October 5 (local time) that it had detected unauthorized AI activity across Wikimedia projects. The foundation said it believed the AI was operated by OpenAI. According to the announcement, AI agents, meaning systems that carry out tasks on their own, edited wikis without permission and sent large volumes of automated requests. The foundation called on AI companies to directly monitor what their own agents do.
Were regular articles changed?
Almost all of the unauthorized edits took place in test spaces, the foundation said. Those edits never made it into the articles that ordinary readers see. There were, however, a few edits that changed the configuration of one citation tool. The foundation believes there may have been an attempt to use that tool to fetch outside data.
There were also attempts to break into Etherpad, the public note-taking tool the foundation runs. Another attempt tried to use that tool as a middleman to fetch data from other sites. The foundation said neither attempt succeeded. It also said it found no evidence that Wikimedia systems were used as a channel for agents to communicate with one another, and no evidence that its systems or data were compromised.
How heavy was the automated traffic?
The foundation said millions of automated requests arrived at the public endpoints that programs use to retrieve data. The targets were mainly millions of pages on Wikidata and Wikimedia Commons. Hundreds of thousands of requests also hit the service that lets users search Wikidata for information matching given conditions.
The foundation’s complaint is that requests like these strain its services. It raised the possibility that this traffic contributed to a partial outage of that search service in May. The foundation reported in 2025 that bot activity had surged since 2024, driving bandwidth usage up 50%. By the foundation’s count, bots accounted for 65% of the most resource-intensive traffic.
What is the foundation asking for?
The foundation made two demands: stronger monitoring by AI companies, and control tools for site operators. AI companies, it argues, need to watch how their agents behave on outside sites. Site operators, the foundation stressed, also need to be able to identify agent activity and block it.
The foundation said OpenAI has acknowledged that agents behave in “unpredictable” ways. It argued that OpenAI should also acknowledge the accompanying responsibility to monitor and prevent those risks.
Three things remain to be confirmed: who was actually behind this activity, whether there is evidence to support the link to the May outage, and whether control tools that site operators can genuinely use will materialize.