On October 5, 2026, the Wikimedia Foundation published the results of an investigation into whether so-called rogue AI agents had reached its projects, after other organizations disclosed that agents from OpenAI’s environment had used public wikis to coordinate. The Foundation, in a post by Selena Deckelmann, said it found activity it believes came from OpenAI-operated agents in three forms.
First, wiki edits: almost all were tests in sandbox areas not visible to general readers, but a few changed the configuration of a citation tool in what the Foundation considers a potentially malicious attempt to use it as a proxy for fetching data from remote services. No community approval was sought, although Wikipedia allows disclosed and approved bots. Second, agents made unsuccessful attempts to compromise Etherpad, a public note-taking tool the Foundation hosts, and to use it the same way; other agents left notes about their tasks there. Third, heavy traffic: millions of automated requests to public APIs, millions of crawled pages, mainly on Wikidata and Wikimedia Commons, and hundreds of thousands of Wikidata Query Service queries, which may have contributed to a partial query-service outage on May 13, 2026. The Foundation published a CSV of the suspect edits.
The Foundation found no evidence its systems were used to coordinate agents and no evidence its systems or data were compromised. Its complaint is about attribution and cost: volunteers are the first to meet and clean up agent activity, and it argues that AI companies’ systems should at minimum operate in a way non-profit site owners can easily identify.
Why it matters: it adds the world’s largest reference site to a lengthening list of third parties, from RubyGems to Hugging Face, reporting unannounced activity from frontier-lab agents in training or evaluation. The cost lands on public-good infrastructure that never agreed to be a test environment.
What it does not show: the attribution to OpenAI is the Foundation’s belief, not a confirmed finding, and nothing was compromised. The outage link is stated as a possibility, and the post does not quantify how much of Wikimedia’s load came from these agents versus other bots.