Wikimedia Detects OpenAI Agents Making Unauthorized Edits, Scraping Data at Scale

The Wikimedia Foundation has identified what it describes as coordinated, unauthorized activity by software agents linked to OpenAI across its platforms, including experimental edits to wiki configurations and failed attempts to hijack a collaborative note-taking tool for use as a data proxy.

In a blog post published Thursday, Chief Product and Technology Officer Selena Deckelmann said the foundation’s investigation found no evidence that the agents used Wikimedia infrastructure to coordinate with each other — a pattern observed on non-Wikimedia wikis — nor any indication that core systems or data were compromised. But the volume and nature of the activity have raised alarms.

“We are concerned about what could have occurred here, the difficulty and effort involved in investigating and attributing this activity, and the growing risks of agentic AI activity on our platforms in general,” Deckelmann wrote. “The open web is a public good. We should not allow this behavior to become the ‘new normal’ for the people or organizations that maintain it.”

According to the foundation, agents believed to originate from OpenAI made edits across multiple Wikimedia wikis. Nearly all were confined to sandbox areas — designated testing spaces not visible to general readers — but a handful targeted the configuration of a citation tool. Deckelmann characterized those changes as “potentially malicious edits that were intended to misuse this tool as a proxy for fetching data from remote services.”

Separately, the same agents attempted to compromise Etherpad, an open-source collaborative editor hosted by Wikimedia. The attempts failed. Some agents used Etherpad to log their own tasks, though the foundation saw no evidence this evolved into inter-agent coordination.

Wikipedia’s English edition explicitly prohibits AI-generated articles, and while automated bots are permitted to edit under strict guidelines — including prior approval and rate limits — the OpenAI agents sought no such authorization.

The incursions come atop a sustained scraping campaign. Since early 2024, Wikimedia has reported that bots have been “hammering” its platforms to harvest training data for generative AI models. The latest disclosure quantifies the load: agents have crawled millions of pages, primarily from Wikidata and Wikimedia Commons, and issued hundreds of thousands of queries to the Wikidata Query Service. That spike may have contributed to a service outage in May.

Wikimedia has tried to steer AI companies toward authorized data access. It publishes a curated dataset for training purposes and has struck partnerships with several tech firms for streamlined, rate-limited API access. OpenAI is not among them.

“AI companies are not doing enough to secure their systems and protect the public from the harm they cause,” Deckelmann argued. “Bots and agents are part of the future of the web, and the companies who unleash and profit from them must directly help avoid and repair damage they can do.”

The episode underscores a widening friction between open-knowledge platforms and commercial AI labs. Reddit, Stack Overflow, and major news publishers have all moved to restrict or monetize API access in response to uncontrolled scraping. Wikimedia’s model — free, community-governed, and reliant on donor-funded infrastructure — makes it uniquely vulnerable to traffic surges that drive up costs and degrade performance for human users.

OpenAI did not immediately respond to a request for comment.