Wikipedia blocks generative ai from editing its pages, citing factual drift

Wikipedia’s English-language edition has quietly pulled the plug on ChatGPT, Gemini, Claude and every other large-language-model assistant that dared touch its sacred pages. The volunteer-run encyclopedia now treats ai-generated prose as a contaminant: one keystroke from a bot and your edit is reverted, your account warned.

The policy bans llms outright, with two surgical exceptions

Editors may still unleash the silicon ghost for spell-checking their own drafts or for translating foreign-language articles into English—but only if a human rereads every clause against the original source. Translation without supervision is classified as vandalism.

The worry is not hallucination in the psychedelic sense; it is subtler. An LLM “corrects” a date, swaps an active verb for a passive one, or summarises a 2017 study in the past tense, and suddenly the article no longer matches the footnote. One pilot test found that 42 % of ai-polished sentences introduced a nuance unsupported by the citation. Multiply that by 6.8 million articles and the corpus becomes a fun-house mirror of reality.

Chaotic Enby, the administrator who proposed the ban, celebrated the vote as “a line in the sand for knowledge integrity.” The message to Silicon Valley: your scraping buffet ends at the foot of this site. The message to fellow Wikipedians: pick up the pen again.

Smaller language editions have begun watching the outcome like hawks. If the English ban cuts vandalism rates without throttling new-article growth, expect copy-paste policies from Berlin to Bengaluru within months. For now, the bots are locked outside the library, and the humans inside still hold the only set of keys.