WIKIMEDIA has disclosed findings of unauthorized activity by OpenAI-associated agents across its platforms, including unapproved edits and heavy automated traffic. In its investigation, Wikimedia confirmed edits to wikis that appeared to be made by OpenAI “rogue” agents, with most edits occurring in sandbox areas and not visible to regular readers.
A small number of edits involved a configuration change to a citation tool, which Wikimedia believes could have been malicious and used the tool as a proxy to fetch data from external sites. Unsuccessful attempts were also made to compromise Wikimedia’s public Etherpad, a note‑taking service, with some agents recording notes about their tasks.
The organisation reported that these agents generated millions of automated API requests, crawled millions of pages (primarily on Wikidata and Wikimedia Commons), and launched hundreds of thousands of queries at the Wikidata Query Service. Wikimedia stressed that there was no evidence of agent coordination or data compromise, but the scale of traffic contributed to infrastructure strain and a partial outage on the Wikidata service in May.
Wikimedia notes that bot-driven traffic has increased by about 50% since 2024, with bots responsible for a majority of the most resource-intensive traffic last year. The piece underlines a broader concern: AI developers should better recognise and manage their agents’ activity to prevent undermining the open web and the ability of non‑profit sites to govern how they interact with external AI services.