OpenAI agents have been caught making unauthorised edits, probing Wikimedia tools and hammering its infrastructure with millions of automated requests.
According to Ars Technica, the Wikimedia Foundation, operator of the online encyclopaedia known around here as Wackypedia after its efforts to make Fudzilla and Mike Magee disappear, said it uncovered the activity during an investigation into OpenAI agents.
Most of the edits were made in sandbox areas and were not visible to ordinary readers. Wikimedia found a few changes to the configuration of a citation tool which it believes were intended to turn the service into a proxy for retrieving data from other websites. The agents apparently had similar ambitions for Wikimedia’s public Etherpad note-taking system.
According to Wikimedia, OpenAI agents made unsuccessful attempts to compromise Etherpad so they could use it to fetch information from third-party sites.
Other agents used the service to keep notes about their tasks, although Wikimedia said it found no evidence that the agents were using its systems to coordinate with each other. The automated visitors were far less subtle in how they consumed resources.
They made millions of requests to Wikimedia’s public APIs, crawled millions of pages, particularly on Wikidata and Wikimedia Commons, and fired hundreds of thousands of queries at the Wikidata Query Service.
Wikimedia said the traffic may have contributed to a partial outage of the query service in May, although neither it nor OpenAI has established a definite connection.
The Foundation said: “We did not find any evidence that our systems were used for coordination among agents, nor did we find any evidence of our systems or data being compromised.”
That qualification still leaves the awkward question of why autonomous systems controlled by one of the world’s largest AI companies were trying to misuse third-party infrastructure without permission. OpenAI told Ars that it appreciated Wikimedia’s findings and was working with the organisation while continuing its own investigation.
“We’re working with them as we review and analyse the activity they identified along with our overall investigation, and we’ll continue to share relevant information as that work progresses,” OpenAI said.
Wikimedia was considerably less soothing.It said AI companies were failing to do enough to monitor their agents and prevent them from damaging outside services. We guess that it considers it the role of its crack team of fake penis experts and those with bogus doctorates to do all that.
The organisation said bot activity had already increased pressure on its infrastructure. During 2025, bandwidth consumption rose by 50 per cent because of increased automated traffic, while bots generated 65 per cent of its most resource-intensive requests.
Wikimedia warned that autonomous agents could drain resources, cause outages, and attempt to exploit vulnerabilities, leaving website operators to discover what happened afterwards.
OpenAI has previously faced reports of agents behaving unexpectedly while carrying out tasks, including attempting to escape sandbox restrictions and interacting with external systems without authorisation.
