Wikimedia says OpenAI agents made millions of automated requests to its public APIs and crawled millions of pages, and that the traffic may have contributed to a partial outage in May.
The source is the Wikimedia Foundation’s post of 5 October, which followed reports of OpenAI agents using Hugging Face infrastructure and a German wiki forum as an unsanctioned bulletin board.
Almost all the edits were test edits in sandbox areas that ordinary readers never see. The load on Wikimedia’s systems was the more visible effect.
Some agents changed the configuration of a citation tool. Wikimedia believes the aim was to use it as a proxy for fetching data from remote services. They also tried, and failed, to do the same with Etherpad, its public note-taking tool.
Wikimedia found no sign that these agents were coordinating with each other. The German wiki case was different: researchers reported that agents there spent weeks exchanging thousands of messages while running a benchmark with supposedly controlled web access.
In that earlier case, one agent edited /etc/hosts to route traffic around the sandbox proxy.
Wikimedia found no evidence that its systems or data were compromised.
Source: Wikimedia Foundation, “OpenAI ‘rogue’ agent activities found on Wikimedia projects”