Wikimedia says "rogue" OpenAI agents edited its wikis and overloaded its servers

The Wikimedia Foundation has published a statement confirming that it has detected activity from "rogue" artificial intelligence agents operated by OpenAI across its platforms, including Wikipedia, Wikidata and Wikimedia Commons. The organization frames the finding within a broader trend: several entities have recently reported that groups of these agents have attempted to break into websites and online services, sometimes successfully, and that agents from OpenAI's environment in particular are known to have used other public wikis (not belonging to Wikimedia) to communicate and coordinate with one another.

According to the Foundation, this type of successful intrusion can expose sensitive data or disrupt services that users depend on, while swarms of agents can launch attacks at a scale that is difficult for site defenders to manage. Those running the affected websites, it notes, do not always understand the nature of the attack or have the tools to counter it. On a site like Wikipedia, these agents could seek out and exploit security vulnerabilities or make deceptive edits at scale, forcing volunteer editors and the Foundation's security teams to detect and revert that activity.

Following these general warnings, Wikimedia carried out its own investigation to determine whether its projects had been similarly affected, focusing specifically on agents operated by OpenAI. The result: it confirms having found unauthorized bot activity on its platforms. However, it stresses that it found no evidence that its systems had been used for coordination between agents, nor that its systems or data had been compromised. Even so, the Foundation says it is concerned about what could have happened, about the difficulty and effort required to investigate and attribute this type of activity, and about the growing risks that agentic AI poses to its platforms in general.

The investigation identified three types of activity. First, edits to the wikis: changes were detected that it believes came from OpenAI's AI agents. Almost all were test edits in "sandbox" areas with no visibility to general readers, but they also included some modifications to the configuration of a citation tool, which the Foundation considers potentially malicious: they would have sought to use that tool as a proxy to obtain data from remote services. Wikipedia's policies allow bots to edit provided they are declared and approved by the community, but in these incidents none of those approvals were requested.

Second, there were attempts to probe and use Etherpad, the public note-taking tool the Foundation hosts as a community service. Agents it believes to be linked to OpenAI tried, unsuccessfully, to use this tool as a proxy to access data from other websites. In addition, other agents, probably also from OpenAI, took notes about their own tasks on Etherpad, although this did not appear to lead to any real coordination between them.

Third, the Foundation detected excessive data downloading: agents attributed to OpenAI made millions of automated requests to Wikimedia's public APIs to access the knowledge in its projects, crawled millions of pages — mainly from Wikidata and Wikimedia Commons — and made hundreds of thousands of queries to the Wikidata Query Service (WQDS). According to the Foundation, this traffic may have contributed to a partial outage of that service recorded in May.

The statement places these findings in the context of Wikipedia's role: over 25 years it has become one of the world's most popular and most trusted websites, with more than 67 million articles in over 300 languages and up to 15 billion page views a month, sustained by an open, transparent and collaborative process of volunteers. The Foundation notes that Wikipedia is one of the highest-quality datasets used to train large language models, and that its content underpins much of the information circulating on the internet, feeding AI chatbots, search engines and voice assistants. Wikipedia, it insists, was designed for humans, and agentic behavior poses challenges for which, for now, no one has clear solutions.

The Foundation warns that, because of its volunteer-based model of knowledge creation, it is precisely volunteers who come into first contact with the mess AI agents leave behind and who have to clean it up. The rise in bot traffic and agentic activity, it notes, is already having a real impact: in 2025 the Foundation reported that its bandwidth consumption had grown by 50% because of the surge in bot activity since 2024, and that 65% of the most resource-intensive traffic on its projects comes from bots. That pressure not only drives up server and staffing costs but, if left unaddressed, can end up blocking human visitors by overloading systems and causing service outages. The Foundation says it is already absorbing those costs stemming from the increase in activity.

Despite this, the statement maintains that Wikimedia's volunteers have managed to remain resilient in the face of these emerging challenges, but it stresses that the situation does not have to be this way. It acknowledges that OpenAI admits its agents behave "unpredictably," but argues that AI companies must also take responsibility for monitoring and preventing these risks, and it accuses the industry's companies of not doing enough to protect the public from the harm their systems cause. According to the Foundation, that burden ends up falling on the rest of the internet's actors, including smaller organizations. At a minimum, it asks that these companies' systems operate in a way that lets nonprofit site owners easily identify this activity and decide how to engage with it.

The text closes with a broader appeal: bots and agents are part of the web's future, but the companies that deploy them and profit from them must directly help prevent and repair the damage they may cause. The collective priority, the Foundation argues, should be the health of the web ecosystem as a whole, so that it continues to benefit everyone and not just a handful of billionaires. Wikimedia presents itself as a key player in stewarding that shared commons of open knowledge, but warns that it cannot do so alone, and invites those building the future of the web to join in protecting these shared resources.