Wikipedia’s outage in May was caused by heavy traffic, including millions of automated data requests, from “rogue” OpenAI agents, the online encyclopaedia’s host, Wikimedia Foundation, said.
The partial outage on 7 May was triggered by AI agents visiting millions of pages and making hundreds of thousands of data queries, Wikimedia said in a statement on Monday.
“Agents we believe to be operated by OpenAI made millions of automated requests to our public APIs to access the knowledge on Wikimedia projects,” the foundation noted, “crawled millions of pages and made hundreds of thousands of data queries to the Wikidata Query Service.”
“We are deeply concerned about the impact of ‘rogue’ AI agents on platforms like ours.”
The OpenAI agents even made unauthorised edits to Wikipedia sites, including "potentially malicious edits" to the citation tool Etherpad, the host said.
Wikimedia said it was “concerned about what could have occurred, the difficulty and effort involved in investigating and attributing this activity, and the growing risks of agentic AI activity on our platforms in general”.
“We should not allow this behaviour to become the ‘new normal’ for the people or organisations that maintain it,” it added.
The AI industry has been hit by multiple serious security incidents in recent months, starting with the Hugging Face breach that saw OpenAI agents hack another AI company on its own in July.
Later that same month, Anthropic said its AI models had hacked three other organisations during testing.
More recently, in September, OpenAI said it had discovered that its agents interacted with several American government websites in unexpected ways.
OpenAI admitted on Tuesday that its actions were "not good enough" after one of its rogue AI agents breached Australian government websites in June.
“We are sorry, and we know we have work to do to rebuild trust with the Australian people. We are committed to doing that work,” the company’s chief strategy officer Jason Kwon said.
These incidents have prompted industry experts to call for further research into AI security vulnerabilities and slow down the fast-growing technology.
In the wake of the breaches, OpenAI delayed the release of its latest AI model, which it claimed had shown "higher levels of deception" than its predecessors.
The company said it was spending more than half a million dollars per day working back through its records month by month, “looking for potential unintended activity beyond the cases we’ve already found”.
OpenAI said it appreciated Wikimedia's "detailed findings" on the 7 May incident and was working with the organisation to analyse the activity of its bots.
"We’ll continue to share relevant information as that work progresses," spokesperson Drew Pusateri said.
Wikimedia said in its statement that “AI companies aren’t doing enough to secure their systems and protect the public from the harm they cause”, warning that the “burden is falling onto everyone else, including smaller organisations”.
“This intense pressure on our infrastructure not only adds costs for servers and humans, but if left unaddressed, can block human visitors by overloading systems and causing outages,” it said.
“Incidents like this one, and the many others that have been, and are still being, uncovered, illustrate how AI agents can drain resources and crash servers, as well as attempt to compromise trustworthy information,” it warned, urging OpenAI to “acknowledge their responsibility to monitor and prevent these risks”.
Read MoreEx-Anthropic whistleblower sounds the alarm about dangerous AI companies at hearing
OpenAI CEO sends chilling warning: ‘World should accept some bad things happening’
ChatGPT is going to show you more ads
Meta’s Muse personal AI creates a detailed file on everyone in your life
