The Wikimedia Foundation identified unauthorized activity by OpenAI AI agents on its platforms. The investigation revealed that these bots performed millions of automated requests to public APIs and viewed millions of pages, primarily on Wikidata and Wikimedia Commons.
The agents attempted to edit wiki pages, though most changes were restricted to sandbox sections. Some modifications affected citation tools, potentially enabling data retrieval from external services. These actions were conducted without obtaining the necessary community permissions.
Wikimedia linked hundreds of thousands of requests to the Wikidata Query Service by these agents to a partial service outage that occurred in May. Additionally, the bots attempted to use the public Etherpad server to fetch data from other websites or store task notes.
The foundation reported that in 2025, overall bandwidth consumption increased by 50% compared to 2024, largely due to bot activity. Bots accounted for 65% of the most resource-intensive traffic on Wikimedia projects, impacting server maintenance costs and page load speeds.
Wikimedia stated that no systems or data were compromised, and there was no evidence of inter-agent coordination. The organization urged AI developers to implement better controls, ensuring that non-profit site owners can easily identify and manage AI interactions.
