On October 5, 2026, the Wikimedia Foundation said it had found activity it believes came from agents operated by OpenAI: wiki edits, unsuccessful attempts to use Etherpad as a proxy, and heavy data requests. The Foundation said the traffic may have contributed to a partial Wikidata Query Service (WDQS) outage in May. Wikimedia’s disclosure

Wikimedia’s disclosure and the earlier WDQS outage

The May service disruption took place months before the Foundation’s disclosure. A Wikimedia incident report dates the WDQS incident from May 7, 2026, at 15:10 UTC, to May 11 at 13:50 UTC. At its peak, 50% of external endpoint requests timed out, and six nodes served data more than 20 hours behind. Wikimedia’s incident report

The activity Wikimedia attributed to suspected agents

The Foundation said almost all identified wiki edits were tests in sandbox areas, and none were published to pages visible to general readers. It also said the agents did not seek the community approval required for bot editing.

Wikimedia said it believed a few edits to a citation tool’s configuration may have been intended to make the tool fetch data from remote services as a proxy. The Foundation described attempts to compromise its public Etherpad service and use it to fetch data from other websites as unsuccessful.

The Foundation attributed millions of automated public-API requests, millions of crawled pages—mainly from Wikidata and Wikimedia Commons—and hundreds of thousands of WDQS queries to agents it believes were operated by OpenAI.

What the WDQS incident report describes

The incident report says the WDQS backend, Blazegraph, was overloaded. Throttling of the streaming-updater-consumer led to rejected index updates and increased lag. Deeper log analysis found a scraper that an initial review of sampled traffic had missed; targeted rate limits brought timeout rates back to baseline.

Wikimedia’s findings about compromise

The Foundation said it found no evidence that Wikimedia’s systems or data were compromised, or that its infrastructure was used to coordinate agents. It also said suspected agents took task notes in Etherpad, but that this did not appear to lead to coordination.