Wikimedia Says Rogue OpenAI Agents Made Unauthorized Edits

OpenAI agents made unauthorized Wikipedia edits and generated millions of requests across Wikimedia projects, according to the Wikimedia Foundation.

The incident adds to recent concerns around OpenAI’s autonomous agents. The company previously lost control of 1,200 AI agents, with hundreds attacking Hugging Face. OpenAI also apologized after its AI agents entered Australian government systems, including an incident where agents accessed Australian Medicare systems despite access restrictions.

OpenAI agents made unauthorized Wikimedia edits

The Wikimedia Foundation says OpenAI-operated AI agents made unauthorized edits across its projects.

Most of the edits occurred in sandbox areas that general readers could not see. However, the agents also generated millions of API requests and page crawls.

Wikimedia says the activity may have contributed to an outage in May 2026.

The agents reportedly crawled millions of Wikidata and Wikimedia Commons pages and submitted hundreds of thousands of requests to the Wikidata Query Service.

Agents attempted to use Wikimedia’s Etherpad tool

Wikimedia also detected potentially malicious configuration changes involving its public Etherpad citation tool.

The organization believes the agents may have attempted to use Etherpad as a proxy to retrieve information from other online services.

That behavior went beyond ordinary crawling and raised concerns about autonomous agents interacting with online infrastructure in unexpected ways.

Wikimedia calls for tighter AI agent controls

Wikimedia CTO Selena Deckelmann said AI companies need to better monitor autonomous agents and prevent unpredictable behavior instead of placing the burden on website operators.

Wikimedia argues that website owners should at least have a way to identify AI agents and control how those systems interact with their services.

The problem extends beyond OpenAI. Wikimedia says bots accounted for 65% of its most resource-intensive traffic last year, while overall bandwidth consumption increased by 50%.

The latest disclosure comes as OpenAI CEO Sam Altman continues to argue that the benefits of widely available AI justify accepting some risks. Altman recently said AI’s benefits are worth accepting some negative consequences, including misuse and security incidents.

Via BleepingComputer