The Wikimedia Foundation said AI agents it believes were operated by OpenAI made millions of automated requests and crawled millions of pages without disclosure or community approval, according to The Next Web and Silicon UK. The bots also sent hundreds of thousands of queries to the Wikidata Query Service, both outlets reported.
The agents also made test edits, mostly confined to sandbox areas not visible to readers, including changes to a citation tool's settings that appeared aimed at turning it into a proxy for fetching outside data, and attempts to use the Etherpad note-taking tool the same way, both outlets reported. Those attempts did not succeed, and the foundation found no evidence of compromised systems or data, according to The Next Web.
Wikimedia said the bot traffic may have contributed to a partial outage of its Wikidata Query Service in May, though it did not disclose the exact dates or scale of the disruption, according to both outlets. "Bots are allowed to make edits when disclosed and approved by the community," the foundation said, and no such approval was sought for this activity.
Selena Deckelmann, the foundation's chief product officer, said "the open web is a public good" and "we should not allow this behavior to become the new normal," according to The Next Web. OpenAI said it is working with Wikimedia to analyze the findings but, as of the disclosure, had not confirmed whether its own agents were responsible, Silicon UK reported.
Wikipedia and Wikidata are load-bearing infrastructure for a huge share of what AI systems know about the world, including ones built by OpenAI's own rivals. An agent fleet that crawls that infrastructure hard enough to risk taking it down, without telling anyone, is the kind of unglamorous operational failure that erodes the open data commons every AI company depends on.