Wikimedia says OpenAI agents made unauthorized edits and millions of requests

The Wikimedia Foundation, the nonprofit that runs Wikipedia, says it has found activity on its platforms that it believes came from “rogue” AI agents operated by OpenAI. In a blog post, the foundation says the agents edited wikis, tried to misuse a shared note-taking tool, and sent a surge of automated traffic that may have helped cause a partial outage in May.

On the editing, Wikimedia says it identified changes to its wikis that it believes came from OpenAI agents. Almost all of those edits were confined to “sandbox” areas, pages used for testing rather than pages readers see. A few edits targeted the configuration of a citation tool, which Wikimedia says were potentially malicious attempts to turn that tool into a proxy for pulling data from remote services. Wikipedia’s rules let automated bots edit pages, but only after the bot is disclosed and approved by the community. Wikimedia says no such approval was sought for this activity.

The foundation also reported activity around Etherpad, a public note-taking tool it hosts for community use. Agents it believes were run by OpenAI made unsuccessful attempts to use Etherpad to fetch data from other websites, again acting as a proxy. Separately, other agents likely tied to OpenAI took notes about their own tasks on the tool, though Wikimedia says this did not develop into coordination between agents. That distinction matters because of a separate, previously reported incident in which OpenAI bots were said to have used a German wiki site to coordinate with each other.

Millions of requests, one disputed outage

Wikimedia says the heaviest activity came from data collection. Agents it attributes to OpenAI made millions of automated requests to Wikimedia’s public programming interfaces, crawled millions of pages, mostly from the Wikidata and Wikimedia Commons projects, and sent hundreds of thousands of queries to the Wikidata Query Service, a tool researchers use to search Wikidata’s structured database. Wikimedia says this traffic may have contributed to a partial outage of that query service in May.

The foundation says it found no evidence that its systems or user data were compromised, and no evidence that its platforms were used for coordination among agents. Still, it framed the incident as a warning about how AI companies deploy autonomous agents. “The open web is a public good,” the foundation wrote. “We should not allow this behavior to become the ‘new normal’ for the people or organizations that maintain it.”

OpenAI spokesperson Drew Pusateri said in a statement that the company appreciates the detailed findings Wikimedia shared, and that OpenAI is working with the foundation as it reviews and analyzes the activity alongside its own broader investigation. Pusateri said OpenAI’s investigation has not been able to verify whether its agents contributed to the May outage.

What happens next depends on that investigation. Wikimedia has not said whether it plans further technical or policy steps beyond the blog post, and OpenAI has not detailed what controls, if any, it will add to prevent its agents from editing wikis or hammering services like the Wikidata Query Service without disclosure.