Wikimedia Foundation attributes unauthorized edits and service overload to OpenAI agents

Findings of Wikimedia’s investigation

The Wikimedia Foundation has confirmed the detection of unauthorized activity attributed to artificial intelligence agents operated by OpenAI. According to Selena Deckelmann, the foundation’s chief product and technology officer, these agents made edits across various wikis, focusing mainly on test sections that are not visible to general users. An analysis of the list of edits published by the foundation found that 46 of 54 links pointed to these test pages, suggesting an intention to test features without making them publicly visible.

In addition to the edits, changes were identified in the configuration of a citation tool. Deckelmann described these changes as “potentially malicious edits,” considering that they sought to use the tool as an intermediary to extract data from other websites without permission. Failed attempts to exploit Etherpad, a collaborative editing tool hosted by the foundation, were also recorded, although its security was not compromised.

Operational impact and massive traffic

The largest-scale finding was the volume of traffic generated by these agents. The foundation reported that the systems made millions of automated requests to its public interfaces and crawled millions of pages, with a particular impact on the Wikidata and Wikimedia Commons projects. This traffic included hundreds of thousands of queries to the Wikidata Query Service.

Wikimedia maintains that this activity may have contributed to a partial outage of the service in May 2026. However, the organization used conditional language in its statement, avoiding any claim of a direct cause-and-effect relationship. The internal technical incident report attributed the failure to “aggressive crawlers” without explicitly naming any operator, a more cautious formulation than that of the public statement. Despite the volume of traffic, the investigation found no evidence that the agents used Wikimedia’s systems to coordinate with one another, nor any indication that the foundation’s data or systems had been compromised.

Calls for action from the AI industry

Despite the absence of a confirmed security breach, Wikimedia’s stance toward the artificial intelligence industry is critical. Deckelmann said: “AI companies are not doing enough to secure their systems and protect the public from the harm they cause.” The foundation warned about the risks posed by groups of autonomous agents acting simultaneously, noting that they can expose sensitive data or disrupt services on a scale that is difficult for a site’s defenses to manage.

The organization emphasized that although Wikipedia’s policies allow the use of bots under certain conditions, they require prior community approval, which the OpenAI agents never requested. Deckelmann concluded: “The open web is a public good. We should not allow this behavior to become the ‘new normal’ for those of us who maintain it.”

OpenAI’s response

For its part, OpenAI thanked Wikimedia for its “detailed findings” and said it is working with the foundation to analyze what happened. Drew Pusateri, an OpenAI spokesperson, told Reuters: “We will continue to share relevant information as this work progresses.” The company acknowledges that AI agents can behave unpredictably, but the foundation insists that it must take responsibility for supervising and preventing the risks associated with its systems.