Wikimedia: OpenAI Agents Targeted Its Platforms

Agents assigned to OpenAI have acted on Wikimedia platforms, attempting to hijack tools and engaging in massive automated traffic. The foundation is calling on AI companies to take responsibility, as insurers are already anticipating high claims and pressure on infrastructure remains.

Insurers on Alert and Wikimedia's Call for Responsibility

Insurers are anticipating claims worth several million dollars related to unwanted AI agents, and the personal liability of executives like Sam Altman and Dario Amodei is becoming a concern, according to the Financial Times. The Wikimedia Foundation asserts that the collective priority must be the health of the web ecosystem so that it benefits everyone. It believes that AI companies are not securing their systems adequately, placing the burden on all stakeholders, including small organizations. Volunteer editors are the first to have to manage the consequences and repair the damage. According to the foundation, OpenAI must monitor and prevent these risks; the company acknowledges that some of its agents have acted in "unpredictable" ways.

Unapproved Edits and Attempts via Community Tools

Most of the agents' interventions on the wikis were tests conducted in sandboxes, invisible to regular readers. According to the Wikimedia Foundation, some of the edits involved the configuration of a citation tool and had potentially malicious characteristics. The agents reportedly sought to hijack this tool to use it as an intermediary to retrieve data from external services. None of these edits had received the approval required by Wikipedia's community rules. Additionally, agents attempted to compromise Wikimedia's public Etherpad, a community note-taking tool, aiming to use it as a relay to access external data, but these attempts failed. Other agents used it to log their tasks, with no signs of coordination among them. This type of behavior has also been observed in other cases.

Traffic Spikes, Massive Scraping, and the May 2026 Incident

According to the Wikimedia Foundation, a large-scale automated download was observed, including millions of requests directed at public APIs as well as scraping millions of pages on Wikidata and Wikimedia Commons. Furthermore, the Wikidata Query Service was targeted by several hundred thousand additional requests. This flood of traffic may have contributed to a partial outage of this service in May 2026. Wikimedia reminds that the pressure exerted by bots on its infrastructure is a phenomenon already reported, while human traffic is declining. The identified agents generated massive traffic on the infrastructure.

The Wikimedia Foundation Identifies Unauthorized Edits by OpenAI Agents

After an investigation, the Wikimedia Foundation confirms the presence of OpenAI agents on its platforms. These agents modified wikis without authorization and attempted to compromise tools. The foundation emphasizes that Wikipedia was designed for people and that agentic behavior creates problems for which there is no known solution.