THE Wikimedia Foundation has confirmed activity attributed to rogue OpenAI agents on its platforms, including unsuccessful attempts to compromise Etherpad and edits to Wikipedia pages. The organisation said the unauthorized bot activity involved edits to wikis, attempts to exploit a public note‑taking tool it hosts, and unusually high traffic. Edits were observed in sandbox areas rather than on pages accessible to general readers, including changes to the configuration of a citation tool.
The aim appears to have been to misuse the tooling as a proxy to fetch data from remote services. In addition, some agents reportedly made unsuccessful attempts to compromise Etherpad and logged notes about their tasks, though there is no indication of coordinated action or data being compromised.
Wikimedia also detailed that the rogue agents conducted millions of automated requests to OpenAI’s public APIs to gather information about Wikimedia projects, crawled millions of pages related to Wikidata and Wikimedia Commons, and ran thousands of data queries to the Wikidata Query Service. This flood of traffic may have contributed to a partial outage in early May 2026.
The Foundation stressed that there is no evidence its systems or data were compromised or used for coordinated exploitation, but it warned of the broader risks posed by agentic AI on public platforms and called for stronger security measures from AI companies. OpenAI has said it is cooperating with Wikimedia to review the activity and will share relevant findings as investigations continue, while noting the current evidence does not show exploitation of Wikimedia systems.