Meta drops the word "train" from its commercial TOS after AI backlash

Meta has revised the commercial terms of service that take effect on October 30, cutting the word "train" and the phrase "any content" from the section describing how it may use material from business users' websites. The change follows a week of criticism after legal experts told Nieman Lab that the original language appeared to authorize broad scraping of sites linked from business pages. A Meta spokesperson confirmed the company did not alert users that the language had changed.
The original terms let Meta use "any content" from websites posted on business pages "to provide, develop, improve, research, train, and maintain the integrity" of Meta products, services or features. The revised version drops "train" and the "any content" phrasing. It now permits Meta to "periodically pull updates" from business users' websites so it can "provide, develop, improve, research, and maintain advertising and business products," and to help maintain the integrity and security of its products. Meta describes the scraping as benefiting users: feeding research that helps them create and deliver ads and other commercial content, give users information about their products and services, and connect them with interested users.
In a statement to Ars, a Meta spokesperson said the company had heard the concerns and wanted to clarify that the crawling terms it consolidated applied only to its ads and business products, and that this is now spelled out in the terms themselves. The statement said the update does not change how Meta uses web crawlers, that content owners can use robots.txt to signal their preferences, that the changes are not about scraping news sites, and that they grant no new rights that supersede those preferences. The spokesperson acknowledged Nieman Lab's first report but not a follow-up that restated the experts' concerns.
The experts who raised the initial alarm were not persuaded. Matt Topic, a media and intellectual property attorney at Loevy & Loevy, told Nieman Lab that publishers should remain concerned, because the revised terms still let Meta use scraped content to develop any "business product" — a term he said is not defined, and one that presumably covers AI business products. In his reading, "develop" is broad enough to include "train." James Grimmelmann, a professor of digital and information law at Cornell Law School, credited Meta for adding more language about permitted uses, but called the "develop…advertising and business products" wording still broad enough to drive a truck through. He agreed with Topic that nothing in the revision restricts the "associated with your products or services" language, which in their view would let Meta scrape entire sites.
Benjamin Sobel, a law professor at the University of Wisconsin, pointed to a fix the critics would accept: explicit terms in which Meta agrees to honor robots.txt instructions. Because complying with robots.txt is voluntary, the experts suggested Meta could lean on enforceable contract terms that say nothing about robots.txt to crawl sites whose owners try to block it. As Sobel put it, a site owner can still express the preference, but unless it is legally enforceable it can be ignored all the same.
This is the crux for developers. The robots.txt protocol is a decades-old convention, not a law: a crawler that ignores it faces social and reputational pressure rather than automatic legal consequences, except where courts or regulators treat the disregard as evidence in a broader dispute. A contract term that binds Meta to respect the file would convert a courtesy into an obligation. The revised language leaves that step untaken, so the practical protection for a site linked from a business page remains the same as before — a signal Meta says it honors but has not promised to.
The pushback extends well beyond the Nieman Lab reporting, with news organizations internationally pressing regulators to block the terms before they start. Theo Bamber, chief executive of the UK News Media Association, called it a "shameless land grab" built on the "worst kind of arrogant tech platform bullying," and urged UK regulators to stand up for British creators and the business community by preventing the precedent.
The European Publishers Council has asked the European Commission to intervene. Its executive director, Angela Mills Wad, said publishers must be able to decide whether and how their journalism is used and to negotiate fair payment, and that burying far-reaching claims in platform terms is no substitute for consent. In the US, Danielle Coffey, president and CEO of the News/Media Alliance — which Nieman Lab described as the largest US news publishing trade association — called the terms outrageous and probably illegal. She argued that withholding access to Meta's audience unless businesses allow scraping is a bad-faith policy and an abuse of market power, and said it cannot be allowed to take effect on October 30.
A likely trade-off is visible in the revision itself. Removing "train" narrows the literal text while leaving a broad purpose clause — developing advertising and business products — that critics say can absorb model training anyway, since AI features inside ad systems are business products. Whether a regulator or court reads the clause that way is unresolved; the source records only the experts' predictions. Meta's position is that the consolidation changed nothing about its crawlers and created no rights beyond existing preferences.
Why it matters: any business that runs a Facebook or Instagram page and links to its own site is affected when the terms take effect October 30, because those links sit inside the scope of the scraping language. For developers and product teams, the immediate question is crawler policy: check what your robots.txt signals, and note that Meta has pointed to that file as the mechanism for opting out rather than writing the promise into the contract. If regulators in the UK or EU act before the deadline, as the publisher groups are demanding, the rollout could shift — an outcome the source reports as a demand, not a decision.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Meta changes AI scraping policy but won’t add language respecting robots.txt.