Wikipedia at 25 Confronts AI Scraping and Musk's Grokipedia

Wikipedia turned twenty five in January 2026, and the anniversary story worth watching, as reported by The New York Times, isn't the milestone. It's what the site is now signing to stay alive. The Wikimedia Foundation has lined up paid deals with Microsoft, Meta, Amazon, Perplexity and Mistral AI, all routed through its Wikimedia Enterprise product, and is asking the rest of the AI industry to stop hammering the open URLs and license the corpus properly, as TechCrunch summarized. For a project that has always defined itself as free knowledge, that is a real pivot.

The pressure is coming from two directions at once. On the traffic side, human page views are down roughly 8% year over year, which the foundation attributes to AI answer summaries and to bots built to evade detection. On the content side, xAI launched Grokipedia on October 27, 2025, and Musk was clear that the Grok team instructed the model to compile Wikipedia's top 1 million articles and make content changes to them, positioning the result as an alternative to what he calls a 'woke' and propaganda-tainted source. Since launch, Grok itself is reportedly generating and approving more than three quarters of the suggested edits, so the alternative is largely a model editing a corpus it lifted from Wikipedia.

Why this matters if you don't work on Wikipedia: the site is the invisible ground truth under most consumer AI answers, and it holds 65 million articles across 300 languages that models have gotten for close to nothing. If the human editor base thins out because the traffic that recruits new editors is being routed to chat summaries, the ground truth degrades on a lag, and every downstream model degrades with it. The Wikimedia Enterprise deals are the first serious attempt to price that dependency.

The honest caveat is that the reporting doesn't give you the numbers that would settle it. We don't know what share of Wikimedia's budget the Enterprise contracts actually cover, whether OpenAI, Anthropic and Google are on similar terms or still scraping, or how quickly Grokipedia-style derivatives could leak back into training sets. Grokipedia's own direct traffic has reportedly dwindled since its peak at launch, but scraped copies don't need direct traffic to do damage.

The forward-looking read is that other high-quality open corpora, from Stack Exchange to public news archives, now have a working template for charging AI labs for structured, licensed access. Wikipedia's next twenty five years may end up being less about the encyclopedia anyone can edit and more about being the licensed spine that keeps machine answers connected to human sources.