Meta Deletes AI Training Language From Business Terms After Publisher Backlash Days Before New Rules Take Effect

Meta has quietly rewritten its upcoming commercial terms of service to remove language that suggested the company could claim website content for AI training, reversing course after a wave of criticism from publishers, media industry groups and legal scholars. The change was confirmed on October 17, less than two weeks before the updated terms were scheduled to take effect on October 30.

The disputed clause said Meta could use content "obtained from websites, web pages or URLs" to develop, improve, research and train its products and services. Because millions of news organizations and businesses run Facebook and Instagram commercial accounts, critics warned the wording could be read as a sweeping license: simply by continuing to use Meta's business tools, an organization might be deemed to have consented to Meta scraping the very website linked to its account, without any licensing negotiation or payment.

Cornell University law professor James Grimmelmann, among the experts cited in Nieman Journalism Lab's reporting on the terms, argued the original language was broad enough to cover content on third-party websites associated with business accounts rather than just posts on Meta's own platforms. Digital media industry representatives were blunter, saying that a business relationship with a social platform should not automatically confer rights over a publisher's own website content, including articles, photographs and other commercially valuable material.

Meta's response to the controversy has been consistent in one respect: the company denies that anything about its crawling behavior actually changed. Spokesperson Andy Stone had said claims that Meta changed how its web crawler operates were false, and after the edit, a spokesperson reiterated that the updated terms did not reflect any change in how Meta scrapes the web. That framing treats the whole episode as a drafting misunderstanding rather than a policy retreat, a characterization that former critics may find hard to square with the fact that the language disappeared only after it was publicly challenged.

The reversal is the latest turn in an unusually noisy month for Meta's data practices. The company had introduced the business terms update quietly, but scrutiny intensified as publishers noticed the October 30 effective date and the mechanics of constructive consent: no opt-out, no notice, just continued use of business tools. The concern was particularly acute for news organizations, many of which are negotiating or litigating AI licensing deals and saw a standard-terms backdoor as undermining that leverage.

What remains unchanged is the broader posture. Meta still trains its models on vast quantities of public web data, and its separate commercial terms continue to govern the Facebook and Instagram relationship businesses have with the platform. The October 30 effective date stands for the rest of the updated terms. The company has not said whether the removed clause will reappear in another form.

For publishers, the episode is a reminder that platform contracts have become an AI-training battleground parallel to copyright litigation. For the AI industry, it shows the diplomatic cost of drafting AI data rights into standard terms: the language survived roughly a day of sustained attention before Meta retreated, a signal of how much reputational weight even the largest platforms now attach to being seen as taking content without asking.