Meta drops AI training clause from its commercial terms after criticism from publishers and legal experts

Meta has deleted a passage from its commercial terms that critics had interpreted as permitting the company to use material from websites linked to its business customers for training its AI systems, The Verge reported. The removal followed criticism of a revision to those terms which, according to The Register, had been scheduled to take effect on October 30, 2026.

As described by The Register, the revised terms had business customers authorize Meta, or anyone acting for it, to deploy automated tools such as crawlers, bots and scrapers, and to create test credentials, so it could access, index, collect and regularly refresh content from domains and URLs tied to their products or services. A further sentence said Meta could use any such content to develop, improve, research and train its products and features. Meta's definition of a business customer spans activities including running ads, selling or listing products, building apps, managing a Page and using tools such as Ads Manager or Meta Business Suite. The Register pointed to Nieman Lab's observation that nearly all news publishers hold at least one Meta business account, and said at least a million businesses use Meta's AI agents in WhatsApp and Messenger.

Critics reacted sharply. Danielle Coffey, president and CEO of the News/Media Alliance, said in a statement that the terms were "outrageous and probably illegal", and described tying access to Meta's audience to permission for scraping as an abuse of the company's market power. Eric Goldman, a law professor at Santa Clara University, told The Register the change raised procedural questions, such as whether existing customers had a genuine way to reject it and whether prepaid fees would be refunded for those who walk away. He also said the wording did not explicitly commit Meta to honoring robots.txt instructions. According to The Verge, Nieman Lab and Cornell law professor James Grimmelmann read the language as an attempt to let Meta train AI on scraped content without paying for it and to override blocking rules in robots.txt files.

Meta initially rejected that reading. The Verge reported that Meta communications vice president Andy Stone and spokesperson Daniel Roberts said: "There is no change to how Meta uses web crawlers; any suggestion to the contrary is false". The Register said Meta had not responded to its request for comment. The company then updated the terms again, and The Verge reported that the sentence allowing Meta to use content from such domains for training and other purposes no longer appears on the commercial terms page.

In a follow-up statement to The Verge, Roberts said the consolidated crawling terms only covered Meta's ads and business products, that the terms now make this clear, and that content owners can still express their preferences through robots.txt. He added that the change was not about scraping news sites and that the updates "don't give us new rights that supersede those preferences". The reports do not state whether the separate authorization for automated access remains in the terms or whether the October 30 effective date still applies. Goldman had compared the episode to earlier terms-of-service disputes at GeoCities in 1999 and Instagram in 2012, which he said did not end in disaster.