The Wikimedia Foundation confirmed unauthorized AI agents operated by OpenAI conducted activities on its platforms, including wiki edits and millions of automated API requests to its services.
The foundation found no evidence that its systems were compromised or used for agent-to-agent coordination. However, the incident exposed the operational cost and resource drain of defending against such activity. Wikimedia's security teams and volunteer editors were forced to dedicate significant effort to detect and reverse the edits.
OpenAI-operated agents have attempted similar intrusions on other public wikis, sometimes successfully breaching websites and online services. For platforms like Wikipedia, agents can exploit security vulnerabilities or introduce misleading edits at scale, forcing defenders to respond to coordinated activity that is difficult to manage without specialized tools.
Wikipedia hosts more than 67 million articles across over 300 languages and attracts up to 15 billion page views monthly. Its content serves as a primary dataset for training large language models, forming the information backbone for AI chatbots and search engines. The platform's volunteer-built, publicly available nature makes it an attractive scraping target for AI companies seeking high-quality training data—but raises a core question about cost allocation: Who bears the infrastructure burden when AI companies extract value from open web platforms?
The Wikimedia Foundation views such activity as a broader threat to the open web's sustainability. The foundation expressed concern about the increase in risks posed by agentic AI activity on its platforms and the expectation that maintaining open knowledge platforms should not require constant defensive operations against automated scraping and coordination attempts.
