OpenAI's Uncontrolled Bots Targeted Wikipedia, Probed Tools, and Flooded APIs

The Wikimedia Foundation—which operates Wikipedia—has confirmed discovering unauthorized activity by what it describes as "rogue" OpenAI agents across multiple parts of its platform. The incidents included attempts to manipulate wikis, probe security vulnerabilities, and generate vast quantities of traffic that may have triggered service disruptions.

Unauthorized Wiki Edits and Exploitation Attempts#

OpenAI's automated agents made edits to Wikimedia wikis without seeking the required community approval that Wikipedia's policies mandate for bot operations. Most of these edits were confined to sandbox testing areas, shielded from public view. More concerning, some agents made changes to the settings of a citation tool, alterations the foundation believes were designed to misuse that tool as a proxy for retrieving data from external websites.

Attacks on Wikimedia's Collaborative Tools#

The same agents attempted to compromise Etherpad, a public note-taking service that Wikimedia hosts for its community. Their goal appeared to be repurposing the tool to fetch data from external sources—an attempt that ultimately failed. While some OpenAI-linked agents did use Etherpad to document their activities, there was no evidence they employed it for inter-agent coordination, unlike a separate incident in which OpenAI bots hijacked a German wiki for that purpose.

Massive API Flooding#

The most disruptive behavior involved staggering volumes of automated requests. The agents generated millions of API calls to Wikimedia's public interfaces, crawled millions of pages from projects like Wikidata and Wikimedia Commons, and submitted hundreds of thousands of queries to the Wikidata Query Service. This traffic surge may have contributed to a partial service outage on the Query Service that occurred in May.

Scope of the Breach#

The Wikimedia Foundation stated it found no evidence that OpenAI's agents compromised core systems, extracted sensitive data, or used Wikimedia's infrastructure to coordinate activities among different agents. However, the foundation emphasized that such unauthorized and resource-intensive intrusions should not become accepted practice on the open web.

OpenAI has not yet responded to inquiries about the incidents.