The Wikimedia Foundation confirmed that artificial intelligence agents that it believes were operated by OpenAI made unauthorized edits, attempted to exploit tools hosted by the organization, and generated millions of automated requests on its services. The investigation was released on Monday (5).
According to the organization responsible for Wikipedia, nearly all the identified edits occurred in test areas, known as sandboxes, and were not visible on pages accessed by the public. The bots, however, had not received the required approval under community policies to perform automated edits.
The foundation also found some changes to the configuration of a citation tool that it classified as potentially malicious. The suspicion is that the agents were trying to use the feature as a proxy to fetch data from external services.
Another front involved Etherpad, a public note-taking tool hosted by Wikimedia. Agents attributed to OpenAI made unsuccessful attempts to compromise the service and to use it to fetch content from other sites. Other agents recorded notes about their tasks, but the investigation found no evidence that the system was used for coordination among them.
Traffic may have contributed to Wikidata outage
The biggest impact at scale came from automated traffic. The agents made millions of requests to public APIs, crawled millions of pages, mainly on Wikidata and Wikimedia Commons, and ran hundreds of thousands of queries on the Wikidata Query Service, a service used to search Wikimedia's structured database.
The foundation says that this volume may have contributed to a partial outage of the Wikidata Query Service in May. The causal relationship, however, has not yet been confirmed. OpenAI said that its investigation could not verify whether its agents contributed to the outage.
Despite the identified activities, Wikimedia said that it found no evidence of compromise of its systems or data. It also found no signs that its platforms were used by the agents to coordinate actions among themselves.
OpenAI said it is working with Wikimedia to review the incidents and analyze the activities identified within its broader investigation into agents.
The incident comes amid increasing pressure from bots on Wikimedia's infrastructure. In 2025, the organization said that its bandwidth use had grown 50% due to the increase in automated activity since 2024 and that bots accounted for 65% of the traffic that consumed the most resources in its projects.
