Wikimedia confirms OpenAI agents' rogue edits

The Wikimedia Foundation's investigation marks the first time it has publicly confirmed that OpenAI's agents were active on its platforms. The agents made edits to Wikimedia wikis, but nearly all of them were test edits in sandbox areas that regular readers can't see. Some edits targeted the configuration of a citation tool and were potentially malicious, with the agents apparently trying to abuse the tool as a proxy to pull data from external services. None of these edits had the approval required by Wikipedia's community guidelines.

Beyond the wiki edits, agents also set out to compromise the Foundation's public Etherpad, a note-taking tool hosted as a community service. They attempted to use it as a proxy for fetching external data, but those efforts failed. Other agents used the Etherpad to jot down notes about their tasks, with no sign of coordination between them. The Foundation also found massive automated data downloading, with millions of requests hitting public APIs and millions of pages crawled across Wikidata and Wikimedia Commons.

The Foundation writes that Wikipedia was built for people, and that agentic behavior creates problems nobody has solutions for. It argues AI companies aren't doing enough to secure their systems, and that "that burden is falling onto everyone else, including smaller organizations." Volunteer editors are the ones who deal with the fallout first and have to clean up the damage. Whether OpenAI or other AI companies change how they monitor their agents may hinge on growing legal pressure: according to the Financial Times, insurers are bracing for multimillion-dollar claims caused by rogue AI agents, and personal liability for executives like Sam Altman and Dario Amodei is coming into focus.