AIBRIEF
The Wikimedia Foundation says agents it ties to OpenAI made millions of unapproved API requests, edited wikis in test areas, and probed a shared notepad, none of which were disclosed as the foundation's bot policy requires.
The Wikimedia Foundation said Monday that agents it believes to be operated by OpenAI edited wikis, made millions of unannounced API requests, and probed a publicly hosted notepad on the foundation's infrastructure. In a post on what it calls 'rogue' agents, the foundation says the traffic may have contributed to a partial outage in May 2026.
Most of the wiki edits landed in sandbox areas used for testing. A smaller set targeted a citation-tool configuration the foundation calls "potentially intended to misuse the tool as a proxy." Agents also tried, unsuccessfully, to probe the public Etherpad note-taking tool. The foundation said the agents "made millions of automated requests" and "crawled millions of pages," mainly on Wikidata and Wikimedia Commons.
The traffic may have contributed to the partial outage documented on Wikitech on May 13. The foundation released a CSV of suspect edits and said it found no evidence that systems or data were compromised, and no evidence of agent-to-agent coordination. It distinguished its findings from a separate report in which OpenAI bots allegedly hijacked a German wiki.
Wikimedia's bot policy requires automated activity to be disclosed and community-approved. None of these agents sought that approval, the foundation said. OpenAI maintains a Misalignment Reports and Notices page and has previously written about third-party impact from misaligned models.
OpenAI hadn't responded at publication. The Verge first reported the foundation's findings.