Wikipedia bars AI bots from direct editing over accuracy and trust concerns

The free encyclopedia’s founder says hallucinations persist, and AI bots are already driving 88bn views yearly, prompting stricter AI access controls.

Wikipedia will not allow AI systems to directly edit its entries, according to its co-founder Jimmy Wales at a climate-focused event on 23 June 2026 in London.

Wales also warned that the technology still cannot be trusted to meet the platform’s accuracy standards. The risk of AI-generated errors remains too high for a resource that millions rely on for factual information, he announced.

While newer AI models have improved, he noted that so-called “hallucinations”, where systems produce confident but incorrect claims, are still a serious problem. Also, Wales acknowledged that the growing use of AI tools is beginning to affect Wikipedia’s audience.

The Wikimedia Foundation reported an 8% drop in human traffic, a decline first disclosed in October 2025. The shift is largely attributed to generative AI platforms and social media services that summarize Wikipedia content without directing users back to the original site.

Despite the drop, Wales characterized the decline as notable but not catastrophic. At the same time, automated traffic has surged, with AI bots accounting for 88bn views on Wikipedia in 2025 alone, placing additional strain on the platform’s infrastructure. He argued that AI developers should contribute financially, noting that the volume of automated requests creates significant operational costs.

To address this concern, the Wikimedia Foundation has expanded its Wikimedia Enterprise program, securing paid licensing agreements with big tech and AI firms, providing them with structured access to its database of roughly 65m articles. The foundation has already started restricting access for entities that fail to comply with its usage policies, signaling a more assertive stance on resource management.

Still, Wales acknowledged that AI could play a constructive role by helping identify obscure or underdeveloped topics for human editors to review.

The situation underscores a broader challenge facing online knowledge platforms: the same AI systems trained on their content are increasingly diverting users away, raising questions about sustainability, attribution, and the long-term economics of open information.