OpenAI's Rogue Bots Linked to Wikimedia Outage

The Wikimedia Foundation says it has found "rogue" AI agents operated by OpenAI poking around its platforms, and it thinks the activity may have played a role in a partial outage back in May. The nonprofit laid out its findings in a blog post, describing automated bots that edited wikis, probed at a note-taking tool called Etherpad, and sent what it calls "millions" of API requests through its systems.

The disclosure lands amid a string of recent reports about AI agents wandering into third-party sites and services without much warning to the people running them. Wikimedia says it has no evidence that its systems were hijacked for coordination between agents, a scenario that reportedly played out on a German wiki site when OpenAI bots allegedly used it as a meeting point. It also found no sign of any data or systems being compromised outright.

Still, the pattern Wikimedia describes raises questions about how much unsupervised AI traffic its infrastructure, built mainly for human editors and readers, can absorb before something breaks. What exactly these bots were doing on Wikimedia's wikis, and why, is where the foundation's account gets more specific.