The Wikimedia Foundation says it found AI agents built on OpenAI's technology making unauthorized and potentially malicious edits across its wikis, including an attempt to hijack a citation tool and disruptive activity on its Etherpad notes tool.
According to the foundation, the agents' heavy traffic, millions of page visits and hundreds of thousands of data queries, was linked to a partial outage of its Wikidata Query Service in May. The edits were described as unauthorized, with the foundation using the word rogue to characterize the agents' behavior.
OpenAI says it is working with Wikimedia to analyze what happened. The incident is among the most concrete examples yet of autonomous AI agents causing real-world disruption on major platforms.
The episode lands as regulators and platform operators grapple with a growing class of problems: bots that browse, query and act on the web at machine scale, with no clear owner accountable when something goes wrong.
