Jimmy Wales stood in London last week and said something many in the tech world already suspected: Wikipedia does not trust AI to edit its articles. The co-founder’s blunt assessment came on the sidelines of a climate action week event. He called AI hallucinations — where models fabricate information with total confidence — “very, very bad.” Newer models have improved, he said, but the problem remains severe.
The statement lands at an awkward moment for the encyclopedia. AI platforms are gobbling up Wikipedia content to answer user queries.
This has produced a strange traffic inversion. Bot visits are surging. Human visits are dropping.
Wales called the human decline “meaningful” but “not a disaster.” That is a careful word choice. Wikipedia’s business model runs on donations, not ad revenue.
So a drop in human eyeballs does not directly kill the budget. But the bot traffic does cost money. “Hammering us with millions of requests costs real money,” Wales said.
Servers are not free. This puts Wikipedia in a bind. It wants AI companies to “pay their fair share.” Some already do.
Wales noted that Wikipedia has signed agreements with several tech giants. The site is also “starting to block the ones who aren’t behaving themselves.” That is a threat.
Whether it works depends on leverage. Wikipedia is one of the most visited websites in the world. AI models trained on its data produce worse answers without it.
So the encyclopedia has cards to play. But the core issue is editorial control.
Wales drew a hard line. AI will not be allowed to directly edit articles. The hallucination problem is too dangerous for a project built on verifiability.
Wikipedia’s model depends on human judgment, citations, and consensus. An AI that fabricates a source or a fact breaks the whole system. The damage would be immediate and hard to undo.
Yet Wales left a door open. AI could prove useful for alerting human editors to niche news that might otherwise be missed.
That is a narrow role. The machine flags a story. A person decides whether it belongs.
The AI stays in its lane. This is not a compromise.
It is a containment strategy. What does this mean going forward? The tension between Wikipedia and AI companies will likely escalate.
The site will keep blocking bots that refuse to pay. More licensing agreements will probably follow. The big AI firms have money.
Wikipedia needs server cash. A deal makes sense for both sides.
But the editing question is different. Wales’ position suggests Wikipedia sees its human editors as a non-negotiable feature, not a bug. The encyclopedia’s reputation rests on the idea that real people check real sources.
Letting AI write articles would undermine that trust. And trust is the only thing Wikipedia really sells.
The irony is obvious. AI platforms depend on Wikipedia’s reliable content. But Wikipedia will not let those same AI models touch its articles.
The relationship is asymmetrical. One side takes. The other side gives, but only on its own terms.
For now, the line holds. AI can read Wikipedia.
It cannot write for it. That may change as models improve. But Wales is betting that human judgment stays valuable.
The hallucination problem is not going away overnight. And Wikipedia has something the AI industry needs: accurate information, curated by people who care about getting it right.
That is a commodity worth protecting.
{“@context”: “https://schema.org”, “@type”: “NewsArticle”, “headline”: “Wikipedia co-founder warns AI hallucinations remain ‘very, very bad’ for edits”, “description”: “Jimmy Wales stood in London last week and said something many in the tech world already suspected: Wikipedia does not trust AI to edit its…”, “datePublished”: “2026-06-22T16:24:00+00:00T09:30:00+00:00”, “author”: {“@type”: “Person”, “name”: “Dmitry Volkov”}, “publisher”: {“@type”: “Organization”, “name”: “InfoPulse Today”, “url”: “https://newsfeed.infopulsetoday.com/”}}
