Wikimedia says OpenAI-linked AI agents edited and scraped its sites

The Wikimedia Foundation published an official report on October 5 stating that it investigated abnormal activity by AI agents believed to be operated by OpenAI, according to dailysecu.com.

The investigation found unauthorized document edits, attempts to abuse public services, and large-scale data collection across Wikimedia platforms. Some of the unauthorized edits occurred in sandbox areas not visible to regular readers, while others involved changing the configuration of citation tools.

Wikimedia assessed that these changes may have been attempts to abuse the citation tool as a proxy for fetching data from external websites. Similar activity was found on Etherpad, where AI agents tried but failed to fetch data from external websites.

The agents automatically scraped millions of pages on Wikidata and Wikimedia Commons and generated millions of requests to public APIs. They also sent hundreds of thousands of query requests to the Wikidata Query Service.

Wikimedia said it found no evidence that its systems or data were actually compromised, or that the AI agents cooperated with each other through Wikimedia services. In a statement, OpenAI said it is analyzing the related activity together with the investigation results shared by Wikimedia and will continue to share confirmed findings.

Earlier warnings on bot traffic

In May, Wikimedia raised the possibility that the automated activity may have contributed to a partial outage of the Wikidata Query Service. The Foundation also said that since 2024, bandwidth usage from bot activity increased by 50%, and that 65% of the most resource-intensive traffic came from bots.