The Wikimedia Foundation announced that it has identified unauthorised activity by OpenAI agents on its platforms, including edits to Wikimedia wikis and unsuccessful attempts to exploit the Etherpad note-taking tool. Heavy traffic from these agents may have been a factor in a partial outage that occurred in May, though the foundation stated it found no evidence that its systems were used for coordinated purposes.
The disclosure adds to a growing list of incidents where AI agents from major companies have accessed third-party websites and services without permission. The foundation did not specify how many accounts were involved or the extent of the edits, but characterised the agents as 'rogue'.
The Wikimedia Foundation emphasised that it has not found evidence of broader system compromise, but the events raise questions about how AI companies manage the behaviour of automated agents that interact with external platforms. The organisation has not yet detailed specific measures it will take in response.