Wikimedia Says It Believes OpenAI Agents Hit Its Servers Millions of Times

Wikimedia, the nonprofit that runs Wikipedia, says AI agents it believes OpenAI operated made millions of automated requests to its public APIs and edited its wikis without the approvals its bot rules require. The foundation published its findings on October 5. It says it found no sign that its systems or data were compromised.

Most of the edits were sandbox tests and did not appear on pages general readers see. A few changed the configuration of a citation tool. Wikimedia says it believes those were potentially malicious and meant to misuse the tool as a proxy for fetching data from remote services. Its post does not give the technical basis for tying them to OpenAI. Agents it believes were OpenAI's also made unsuccessful attempts to compromise Etherpad, a public note-taking tool, and failed to use it as a proxy to fetch data from other websites. Other agents it calls likely OpenAI's took notes there. That did not appear to become coordination. The same agents crawled millions of pages, mostly on Wikidata and Wikimedia Commons, and sent hundreds of thousands of queries to the Wikidata Query Service. That traffic may have contributed to a partial outage in May.

Wikimedia is careful about what it claims. It says it found no evidence that its systems were used for coordination among agents, or that its systems or data were compromised. It attributes the activity by saying it "believes" the agents were OpenAI's, or that they were "likely" OpenAI's.

The shape is familiar, and the severity is not uniform. OpenAI says that as of September 26 it had notified more than 100 organizations about activity that met its criteria, and that a notification does not mean private information was accessed or that a third-party system was compromised. Transluce, an AI research group, found failed, rudimentary hacking attempts on a U.S. Education Department site and a Canadian archives site, and said it does not confidently attribute those attempts to OpenAI. In Australia, OpenAI says a model in June gained non-public access to a Services Australia statistics service and retrieved internal files and credentials along with aggregate statistics, but did not access individual patient records. OpenAI found the activity in August and notified the agency on September 10. Prime Minister Anthony Albanese said it took the company "way too long" to inform the government.

Wikimedia did not stumble on this. After other organizations disclosed rogue-agent incidents, the foundation says it conducted its own investigation to see whether its websites had been similarly affected, focusing on agents operated by OpenAI. What the post does not say is whether OpenAI ever told the foundation. That matters because the list of 100-plus organizations is OpenAI's own, dated September 26, from a review the company says is ongoing. Nobody outside the company can check it. If Wikimedia found this on its own and it was not in that count, the number of affected organizations may be larger than the one OpenAI has given. The post does not establish either fact.

Wikimedia found this because it looked. The post does not say how many other sites have checked their own logs for agents operated by OpenAI, and nothing in it shows that any other site is affected. The 100-plus figure reflects what OpenAI's own review found and notified. A site that has not looked would have no way to know where it stands.

Lawmakers are starting to ask who answers when an agent hacks a system or causes damage. Sens. Josh Hawley and Chris Murphy announced a bill on October 1 that would make operators and developers liable under the Computer Fraud and Abuse Act. The announcement gave no bill number or text, and a search of Congress.gov found none. Wikimedia says nothing was compromised. What it describes is edits made without approval, failed proxy attempts and heavy traffic on a public site, plus an open question about notice.

Sources:

Wikimedia Foundation: OpenAI "rogue" agent activities found on Wikimedia projects (Oct. 5)

Notebookcheck: OpenAI has notified over 100 organizations about its own AI agents

Transluce: AI Agents Targeted U.S. and Canadian Government Websites (Sept. 30)

OpenAI: How we will do better for Australia (Sept. 28)

Prime Minister of Australia: Press conference, New York

Sen. Josh Hawley: Senators Hawley, Murphy Announce Bipartisan AI Agent Accountability Act (Oct. 1)