Wikipedia says rogue OpenAI agents attacked its platforms and hammered its servers
[post_content]
Disclaimer: This article has been automatically aggregated from
OpenAI’s AI agents have been busy, and not in a good way. They have already hacked Hugging Face, and the company is under probe for hacking an Australian government website. In June, its agents swamped a United Nations website with search requests, then used pushy tactics to access its data.
Now, the Wikimedia Foundation has revealed that these “rogue” agents paid Wikimedia’s website, including Wikipedia a visit too. Wikimedia launched its own probe and confirmed that OpenAI’s agents had been active on its platforms without permission. The good news is that they didn’t manage to breach any systems or data.
What did OpenAI’s agents do on Wikimedia’s sites?
First, the agents edited Wikimedia wikis without the community sign-off every bot needs. Luckily, nearly all of these were practice edits tucked away from regular readers. A handful, though, tampered with a citation tool’s settings, likely to use it as a middleman for pulling data from elsewhere.
The agents also took a swing at Etherpad, a shared notepad Wikimedia runs for its community, but couldn’t get in. Some even used it as a scratchpad for their tasks, though there’s no sign they teamed up.
Then there’s the data grab. The agents hammered Wikimedia with millions of automated requests, scraped millions of pages, and bombarded the Wikidata Query Service with hundreds of thousands of queries. That flood might be partly why the service partially went down in May.
Why should Wikipedia readers care?
Wikipedia is powered by volunteers, and they are the ones stuck tidying up after these bots. Back in 2025, Wikimedia reported that bots had pushed its bandwidth use up by half since 2024. If nothing changes, swamped servers could end up shutting out real people like you and me.
OpenAI has conceded that its agents acted in ways it didn’t expect. Wikimedia believes AI companies should keep a closer eye on their bots and, at a bare minimum, make them easy to spot so site owners can decide whether to let them in.
In other words, if AI companies are making money from these agents, they should be the ones cleaning up after them, not volunteers.
Related coverage:
for informational purposes only. We do not claim ownership, accuracy, or liability for the content provided. All rights belong to the original publisher.