Meta Pulls AI-Training Scraping Clause From Business Terms

Meta has removed the provision in its commercial terms of use that legal experts said could let the company scrape any website linked to a business account and use that content to train its AI models.

The clause, due to take effect October 30, authorized Meta to deploy "web scrapers, bots, robots, spiders, crawlers, and user agents" against "the domains, web pages, or URLs... that are associated with your products or services," and to use any content found there to "provide, develop, improve, research, train, and maintain the integrity of" its products. As of Saturday, that language is gone, The Verge reported.

What Meta changed

Meta initially framed the update as housekeeping, saying it was consolidating crawling language scattered across separate terms. Spokesperson Daniel Roberts told TechCrunch the provision "only applied to our ads and business products." This morning's follow-up went further: "We've heard you... the crawling terms we consolidated only applied to our ads and business products, and we've clarified that in the terms themselves."

The retreat followed a week of pressure. Nieman Lab flagged the clause first, noting that virtually every news organization holds a Meta business account and that the terms offered no opt-out beyond abandoning Meta entirely.

Why publishers treated it as a licensing end-run

The stakes were financial. News organizations are already litigating whether training AI on their articles without a license is fair use, and the clause appeared to grant Meta the same permission regardless of how those cases resolve. Cornell law professor James Grimmelmann warned it could let Meta train on publisher content and then "deploy some of that content, even in competition with the publishers themselves." Attorney Matt Topic called it "a remarkable provision to bury in terms of service," given the material would likely cost millions in a licensing market, Ars Technica reported.

The timing compounded it. Meta had just started charging users to post links, creating a path where publishers pay to share links that Meta could then treat as licensed training data. And because the text never used the word "AI," the permission rested on one word: "train."

What builders should take from it

For anyone running agents, crawlers, or content pipelines, the episode is a reminder that robots.txt is the only machine-readable signal that reliably survives a contract rewrite. Meta now says content owners "can use robots.txt to signal their preferences" and that the update grants no rights superseding them. The next consolidated terms update will arrive with the same boilerplate and no changelog.

The practical move is to audit what your own sites expose. If your marketing site, docs, or client portal is tied to a business account on any large platform, assume the terms can expand to cover the linked domains, and treat machine-readable opt-outs as the enforcement layer you actually control.