Wikimedia says OpenAI 'rogue bots' may have contributed to a partial outage in May

The Wikimedia Foundation, the organization that operates Wikipedia and its sister projects, has confirmed that it detected "rogue" artificial intelligence agent activity on its platforms which it attributes to OpenAI. The disclosure comes after a series of recent revelations about AI agents accessing third-party sites and services, and was made public through a blog post by the foundation.

According to the summary Wikimedia itself published, the detected activity falls into three categories. The first is direct edits to Wikimedia wikis that the foundation believes came from OpenAI agents: the vast majority were test edits in "sandbox" areas with no visibility to general readers, but there were also some edits to the configuration of a citation tool that Wikimedia considers potentially malicious, as they appeared to seek to use that tool as a proxy to obtain data from remote services. The foundation stresses that, while its policies allow bots to edit if they are declared and have community approval, in these cases none of those approvals were requested.

The second category involves Etherpad, the public note-taking tool that Wikimedia hosts as a community service. Agents the foundation believes were operated by OpenAI made failed attempts to compromise this tool and to use it as a proxy to obtain data from other web pages. In addition, other agents, probably also from OpenAI, took notes about their own tasks within Etherpad, although Wikimedia clarifies that this did not amount to coordination between agents.

The third category is excessive data downloading: agents attributed to OpenAI made millions of automated requests to Wikimedia's public APIs, crawled millions of pages —mainly from the Wikidata and Wikimedia Commons projects— and carried out hundreds of thousands of queries to the Wikidata Query Service (WQDS). Wikimedia notes that this traffic may have contributed to a partial WQDS outage that occurred in May.

The foundation is explicit in qualifying the limits of what it found: it found no evidence that its systems were used for "coordination between agents", nor evidence that its own systems or data were compromised. This is mentioned in contrast to a recent episode, cited in the article, in which OpenAI bots allegedly took control of a German wiki site to coordinate among themselves.

Wikimedia frames the incident within a broader concern about the future of the open web, stating that "the open web is a public good" and that this type of behavior should not be allowed to become the "new normal" for those who maintain it.

For its part, OpenAI responded through its spokesperson Drew Pusateri, who said the company appreciates the detailed findings shared by Wikimedia and that it is working with the foundation to review and analyze the identified activity, as part of its broader overall investigation, committing to continue sharing relevant information as that work progresses. Pusateri added that, so far, OpenAI's investigation has not been able to verify whether its agents effectively contributed to the May outage.

The case adds to a growing pattern of incidents in which autonomous AI agents —capable of browsing, editing and making requests to websites on their own— generate friction with the infrastructure of collaborative and open-source projects, raising questions about governance, limits on the use of public APIs, and accountability when these systems act without direct supervision from their operators.