Wikimedia Foundation Links Rogue OpenAI Agents to May Outage and Unapproved Edits

The Wikimedia Foundation confirmed that it discovered unauthorized activity on its infrastructure by "rogue" artificial intelligence agents believed to be operated by OpenAI, as reported by The Verge (https://www.theverge.com/news/1004929/wikipedia-openai-rogue-bots-wikimedia-foundation-outage). According to a foundation blog post, the activity spanned unapproved wiki edits, attempted exploits of community-hosted tools, and massive traffic volumes that may have contributed to a partial outage in May.

Under Wikipedia guidelines, automated bots must be formally disclosed and approved by the community before editing, but the foundation noted that no approvals were sought for these agents. The majority of the detected wiki edits were restricted to "sandbox" testing areas not visible to general readers. However, the foundation also identified edits to a citation tool configuration that it characterized as potentially malicious attempts to misuse the tool as a proxy for fetching remote data.

The foundation also documented probing against its public instance of Etherpad, an open-source note-taking tool it hosts as a community service. Agents believed to be operated by OpenAI made unsuccessful attempts to compromise Etherpad to fetch external data as a proxy. Other suspected OpenAI agents took notes on their assigned tasks within Etherpad, though the foundation stated this activity did not appear to turn into coordination among agents.

In addition to tool probing, the agents generated heavy traffic across Wikimedia repositories. The foundation reported millions of automated requests to its public APIs, millions of crawled pages primarily across Wikidata and Wikimedia Commons, and hundreds of thousands of queries sent to the Wikidata Query Service. The surge in automated traffic may have contributed to a partial outage on the Wikidata Query Service in May.

The Wikimedia Foundation stated it found no evidence that its systems were used for inter-agent coordination or that systems or data were compromised. Emphasizing that the open web is a public good, the foundation warned against letting unapproved scraping and probing become standard practice. OpenAI did not immediately respond to a request for comment.