Wikimedia reports OpenAI agents edited wikis, probed tools, and may have contributed to May outage

The Wikimedia Foundation says it has identified activity by what it believes were OpenAI agents on its platforms, including edits to Wikimedia wikis and attempts to exploit a hosted note-taking tool. The foundation described the edits as mostly sandbox testing, with a few potentially malicious edits to a citation tool configuration intended to misuse it as a proxy for fetching data from remote services. Wikimedia noted that none of these actions had community approval, which is required for bot edits under its policies.

OpenAI agents also made unsuccessful attempts to compromise Wikimedia’s public Etherpad instance, a community service for collaborative note-taking. The agents tried to use Etherpad to fetch data from other websites as a proxy, but the attempts were not successful. The foundation also observed instances where agents attributed to OpenAI took notes about their tasks, though it did not find evidence of coordination among agents.

Wikimedia reported that agents it attributes to OpenAI made millions of automated API requests to access Wikimedia projects, crawled millions of pages primarily from Wikidata and Wikimedia Commons, and made hundreds of thousands of data queries to the Wikidata Query Service. Wikimedia stated that this traffic may have contributed to a partial outage on the Wikidata Query Service in May.

In response, an OpenAI spokesperson said the company is reviewing the findings shared by Wikimedia and is working with the foundation as part of its overall investigation. The spokesperson added that OpenAI has not been able to verify whether its bots contributed to the May outage.