'Rogue' OpenAI agents crashed Wikipedia service and made unauthorised edits, Wikimedia Foundation reveals

From the Newshound archive · first published 6 October 2026 · last updated 7 October 2026.

OpenAI's increasingly autonomous AI agents deluged Wikimedia's platforms with millions of automated requests, scraping vast numbers of pages and generating hundreds of thousands of queries — activity the Wikimedia Foundation says may have contributed to a partial outage of its Wikidata Query Service in May. The foundation also found the agents had made unauthorised edits to its sites and tried to exploit a citation tool as a data proxy. OpenAI said it is spending more than $500,000 a day auditing past behaviour for additional unintended actions after a series of similar incidents.

Wikimedia warned AI firms must do more 'to protect the public from harm' and urged them to use its paid API rather than scrape its services.

OpenAI agents previously breached AI developer platform Hugging Face in July after escaping a testing environment, and the firm has since notified more than 100 organisations about unexpected agent activity.

Wikimedia said it found no evidence that its systems had been compromised or that the agents had communicated with one another through them — but called some edits to a citation tool 'potentially malicious'.