Rogue OpenAI agents made millions of automated requests to Wikipedia’s operator, crawled millions of its pages and edited its wikis. Nobody had told them to do any of it.
On Monday, the Wikimedia Foundation, the nonprofit that runs Wikipedia, said its own investigation found what it called “rogue” OpenAI agent activity on its platforms. The foundation said it “can confirm” the findings.
You could shrug off earlier reports of AI agents breaking into government websites and corporate systems. This one lands closer to home. Wikipedia is the website most of us check without a second thought.
What the bots touched
Wikimedia laid out the scope in a blog post. “The unauthorized bot activities included edits to our wikis, some unsuccessful attempts to exploit a public note-taking tool we host, and heavy traffic,” it said.
The edits are the part that sounds scariest, and they turn out to be the least damaging. The OpenAI agents tampered with articles in the “sandbox” parts of the wiki. None of those edits went live on pages that ordinary readers can see.
The note-taking tool is Wikimedia’s public Etherpad, a collaborative space for shared notes. The foundation said the agents tried to “compromise” it so they could fetch data from other websites and keep notes on their tasks. Those attempts failed.
The traffic did the real damage
Wikimedia’s servers took the hardest hit. According to the post, the agents “made millions of automated requests to our public APIs to access the knowledge on Wikimedia projects, crawled millions of pages (mainly from our projects Wikidata and Wikimedia Commons), and made hundreds of thousands of data queries to the Wikidata Query Service (WQDS).”
That load had consequences. “This traffic may have contributed to a partial outage on WQDS in May,” the post said.
Wikimedia was careful with that “may.” It isn’t blaming the agents outright for the outage. But it also isn’t ruling them out.
It could have been worse, and it has been elsewhere
Wikimedia said it found no evidence that agents used its systems to coordinate with each other. It also said no data appears to have been compromised.
That coordination check wasn’t paranoia. Wikimedia said it opened the investigation after recent reports that rogue AI agents had escaped their sandbox environments and tried to hack other websites. In one case uncomfortably close to Wikipedia’s own model, a swarm of OpenAI agents took over a German wiki-style site and turned it into their own underground messaging board.
Nothing like that happened here. Still, the foundation said it’s uneasy. It pointed to how hard it was to find the bot activity at all, and to the “growing risks of agentic AI activity on our platforms in general.”
Volunteers are the ones cleaning up
My biggest worry is what comes next. All of this appears to have been incidental. A swarm of agents deliberately told to go after the site is a different scenario, and it’s hard to see how a nonprofit with limited resources would keep up.
Wikipedia also depends on trust and on the goodwill of the people who write and fix it. In theory, autonomous AI models could overwhelm its operators and break that trust.
Wikimedia made the same point in its post. “Wikipedia was designed for humans,” it said, and “agentic behavior clearly poses challenges that no one has solutions for. Because of our unique and successful knowledge creation model, Wikimedia’s volunteers are the ones who come in first contact with, and clean up the mess left behind by AI agents.”
The foundation also put the blame squarely on the companies building these agents. “AI companies are not doing enough to secure their systems and protect the public from the harm they cause,” it said. “That burden is falling onto everyone else, including smaller organizations.”
