Wikipedia slams the brakes on ai-generated edits, forcing editors back to human hands
Wikipedia’s English-language editors just drew a hard line: no ChatGPT, Gemini, Claude or any other large-language model may be used to write or rewrite articles, effective immediately. The ruling, ratified this week by the volunteer-run community, brands generative AI a policy violator that too often warps facts, invents citations and subtly twists neutral prose into promotional mush.
The ban is sweeping. AI can no longer draft paragraphs, expand stubs or rewrite leads. The only exceptions are two tightly-controlled safety valves: error correction and human-supervised translation. Even then, editors must personally verify every AI-suggested comma, because experience shows the models can “fix” a typo and accidentally slip in an unsupported claim.
Why the crackdown happened now
Frustration had been mounting since early 2023, when AI-generated drafts flooded the volunteer queue. Reviewers found themselves spending more time scrubbing hallucinated sources than improving actual knowledge. One editor likened it to “letting a smiling stranger walk into your library with a flamethrower.” The final straw came when a biographical entry quietly inserted by ChatGPT claimed a living chemist had won a Nobel Prize she never received; the falsehood survived three weeks and 67,000 page views before detection.
Administrator Chaotic Enby, who proposed the ban, says the goal is not anti-tech posturing but source integrity. “We already require everything to be traceable to a reliable source. LLMs can’t sign that contract.”
The narrow escape routes
Editors may still run their own text through a grammar bot, yet must eyeball every change. Translation from non-English wikis is allowed, but only if the editor becomes the human guarantor of every rendered sentence, a process the policy chillingly calls “post-editing with liability.”
Violators risk the standard wiki death penalty: reverts, warnings and, for repeat offenders, the permanent loss of editing rights. No algorithm gets an appeal.
What happens next
Other language editions are watching. German, French and Japanese deliberations already cite the English ruling as precedent. Meanwhile, AI companies scramble to release “Wikipedia-safe” fine-prints that promise stricter citation grounding—pledges volunteers greet with cold skepticism. The message is blunt: if you want to shape the world’s largest open encyclopedia, bring a human brain or stay home.