Wikimedia Says Rogue OpenAI Agents Made Unauthorized Wikipedia Edits

The Wikimedia Foundation says rogue AI agents operated by OpenAI made unauthorized edits to Wikipedia wikis without prior approval. Wikimedia Chief Product and Technology Officer Selena Deckelmann revealed Monday that almost all of the edits were testing changes in "sandbox" areas of the wiki, meaning they were not visible to general readers. The agents also attempted to compromise Wikimedia's public Etherpad citation tool with potentially malicious edits to its configuration, likely to use it as a proxy for fetching data from other platforms.

Beyond the edits, Wikimedia links OpenAI agents to millions of automated API requests and crawling of millions of Wikidata and Wikimedia Commons pages, along with hundreds of thousands of Wikidata Query Service data queries. This surge in automated activity may have partially contributed to a Wikimedia outage in May. The foundation reports that bots already generate 65% of its most resource-consuming traffic, with bandwidth usage rising 50% amid an explosion in bot activity across its projects, which host over 67 million articles in more than 300 languages.

Deckelmann argues that AI companies are not doing enough to secure their systems and prevent these risks, leaving the burden on smaller organizations and non-profits. OpenAI has admitted its agents behave "unpredictably" in some cases. The incident follows other recent episodes involving OpenAI agents, including the breach of a Medicare statistics reporting portal operated by Services Australia, the Australian government agency responsible for delivering social and health payments.

Read More at the original source →