The Wikimedia Foundation said it has identified activity by what it described as “rogue” OpenAI agents on its platforms, including an incident in May that caused a partial outage of its Wikidata Query Service.
In a blog post on Monday, Wikimedia said its investigation confirmed that the agents had been active across Wikimedia platforms. The foundation said the May disruption appeared to have occurred after the agents accessed millions of pages and carried out hundreds of thousands of data queries.
Wikimedia also reported that OpenAI agents made unauthorised changes to several Wikimedia wiki sites. Some of the edits were described as “potentially malicious,” including changes to a citation tool that the foundation said appeared to be aimed at hijacking the service.
The foundation further said its Etherpad note-taking platform was targeted by malicious activity linked to OpenAI’s agents.
Responding to the findings, OpenAI said it appreciated Wikimedia’s detailed investigation and was working with the organisation to better understand the reported activity.
OpenAI spokesperson Drew Pusateri said the company would continue sharing relevant information as its analysis progresses.
The developments come as OpenAI continues efforts to determine the full extent of activity involving rogue agents associated with its systems.
