Meta pulls the AI-training line from its business terms

The revised Commercial Terms, which take effect October 30, 2026, told businesses that Meta could use “automated software (such as web scrapers, bots, robots, spiders, crawlers, and user agents) and create test credentials” to access, index, cache, analyze, collect, retrieve, or crawl the domains, web pages, or URLs associated with their products or services — and then “use any content from such domains, web pages, or URLs to provide, develop, improve, research, train, and maintain the integrity of” Meta products and services, a product list that includes its AI services.

The reach was the problem. The terms cover anyone using Meta's products for commercial purposes — running ads, managing a Page or business Group, listing products, using Ads Manager or Business Suite. As Nieman Lab noted, almost every news publisher keeps at least one Meta business account. And acceptance was passive: keep using the tools after October 30 and you agree, with no partial opt-out. The only way to refuse was to stop using Meta's business tools.

Nieman Lab had legal experts read the clause on October 15, and the verdict was blunt. Matt Topic of Loevy & Loevy, James Grimmelmann of Cornell Law School, and Benjamin Sobel of the University of Wisconsin said the language could support scraping publisher websites to train AI models — Nieman Lab called it a “remarkable provision.” Digital Content Next CEO Jason Kint said the terms reached “outside its walled gardens,” and Grimmelmann called them “absurdly broad.” For newsrooms that spent years negotiating licensing payments for their archives, a contract clause granting the same data away for nothing was the worst possible deal.

Meta pushed back on the reading. Communications spokesperson Andy Stone said on October 16 that the update was designed to let Meta pull accurate product information from third-party sites for on-platform ad and content recommendations: “this is not about scraping news sites.” But the clause as written named training without limiting its scope to product pages — and that gap between stated intent and contract text was exactly what critics seized on.

The Verge reported on October 17 that the training sentence is gone from the terms page. Meta told The Verge: “We’ve heard you and want to clarify: the crawling terms we consolidated only applied to our ads and business products, and we’ve clarified that in the terms themselves.” Meta's statement added that this “does not change how Meta uses web crawlers,” that the updates are “not about scraping news sites,” and that they “don’t give us new rights that supersede” site owners' preferences — adding that owners can still use robots.txt to signal their choices.

What survived: Meta still authorizes itself to send bots and crawlers across sites tied to a business's products or services, create test credentials, and pull updates periodically. The stated purposes are now providing, developing, improving, researching, and maintaining advertising and business products, plus product integrity and security — training is out. The October 30 effective date still stands.

This is a meaningful win for anyone whose content is worth licensing: a few words in a click-through contract nearly became standing permission to train models on publisher archives. Nothing stops Meta from revisiting the language in the next terms update, so the durable defense is the boring one — read the terms when they change, keep robots.txt current, and watch server logs for unfamiliar crawlers.

The bigger lesson travels beyond Meta. AI training data is now won or lost in commercial contracts, not just in scraping lawsuits — and this episode shows how much ground a single sentence can cover before anyone reads it. If you run a business with a Meta account, read the final terms before October 30; if you publish anything of value, treat every platform's fine print as a data negotiation until proven otherwise.