Meta drops contract language that could have let it scrape news sites for AI training

Meta Platforms this week quietly rewrote, and then just as quietly withdrew, a clause buried in the commercial contract that governs every business account on Facebook, Instagram and WhatsApp. For a few days, the new wording appeared to give the company broad latitude to pull content from the websites linked to those accounts and feed it into the systems that train its artificial intelligence models. After lawyers and publishers raised alarms, Meta deleted the disputed passage, saying it had never intended the scope that critics described.

The episode centers on Meta's commercial terms of use, the contract that advertisers, Page owners and other business users must accept to keep operating on Meta's platforms. The media-industry publication Nieman Lab first reported on the updated clause last week; after the backlash it set off, Meta's rollback was confirmed in reporting published by TechCrunch on Saturday. An update to those terms, set to take effect October 30, told business account holders that Meta could use content from their "domains, web pages, or URLs" to "provide, develop, improve, research, train, and maintain" its products and services. Because Meta counts huge numbers of news organizations among its business account holders, legal experts said the language read as a potential license to scrape entire publisher websites for AI training, not merely the ads and Page posts the terms were ostensibly written to cover.

The numbers behind the alarm

The scale of exposure is what turned a contract tweak into a news story. Virtually any outlet that runs a Facebook Page, buys a single Instagram ad or verifies its domain through Meta's business tools counts as a "business account" bound by the terms, and Nieman Lab noted that this includes nearly every news organization, including itself. Three legal specialists who reviewed the clause for that publication concluded it could be read to cover a publisher's entire website, not just the handful of posts or ad creatives a business actually uploads to Meta's platforms, and one called it a "remarkable provision." Another said it remained unclear whether a court would enforce such terms against a publisher that never meaningfully negotiated them, but advised publishers to assume they could be enforceable unless a court ruled otherwise.

Meta disputes that reading, and says the clause was never meant to reach that far. The company also points publishers toward its longstanding crawler documentation, which explains how site owners can block its bots through robots.txt, and to its separate public notice on how generative AI features draw on user data, which describes training on "publicly available" information as a distinct category from the content at issue in the commercial terms.

How this fits a longer fight over AI training data

The dispute lands in the middle of a broader legal reckoning over how AI companies obtain the material that trains their models. Meta already won one major round: a federal judge ruled last year that the company's use of 13 authors' books to train its Llama models counted as fair use, though the judge was explicit that the decision was narrow and did not establish that Meta's AI training is lawful more broadly. That case has not ended the legal pressure. In May, five major publishers sued Meta, including Elsevier, Hachette and McGraw Hill, along with novelist Scott Turow, accusing the company and chief executive Mark Zuckerberg personally of training its models on pirated books obtained through online shadow libraries, a claim Meta has said it will "fight aggressively."

Against that backdrop, any new contract language that could be read to authorize scraping publisher content carries more weight than it might have a few years ago. News organizations have spent the past two years trying to negotiate licensing deals or block AI crawlers outright, even as research firms report that only a small fraction of websites have configured their robots.txt files to stop Meta's bots. A clause that effectively asked publishers to consent to scraping simply by keeping a Facebook Page open was, to media lawyers, the kind of one-sided bargain regulators and courts have increasingly scrutinized.

Who is affected

The immediate audience for the change is the enormous population of businesses, including most newsrooms, that maintain a presence on Meta's advertising or Page tools. For publishers specifically, the stakes went beyond routine terms-of-service boilerplate: many derive meaningful referral traffic and ad revenue from Facebook and Instagram, making it impractical to simply close their business accounts in protest. Smaller outlets without in-house counsel were seen as especially vulnerable to accepting sweeping terms without realizing their implications, while larger publishers with existing AI licensing agreements with Meta, reached separately over the past year, were comparatively insulated.

• The disputed clause applied to any business account's associated domains, web pages or URLs.
• Meta's commercial terms update is scheduled to take effect October 30, 2026.
• A federal judge's 2025 ruling for Meta on AI training was limited to 13 named plaintiffs and explicitly called "narrow."
• Five publishers' separate copyright suit against Meta, filed in May, remains pending in New York federal court.

What people are saying

Meta's public response has focused on narrowing, rather than defending, the original wording. A company spokesperson said in a statement:

"We've heard you and want to clarify: the crawling terms we consolidated only applied to our ads and business products. It does not change how Meta uses web crawlers."

Media attorneys remained cautious even after the rollback. Even with the specific "train" language deleted, lawyers were still reviewing the rest of the commercial terms this weekend to check what crawling and data-use permissions remain intact ahead of the October 30 effective date.

What happens next

Meta has not said whether it will revise the commercial terms further before they take effect at the end of the month, or whether the removal announced this week is the final word. Publishers and digital-rights lawyers are expected to keep combing through the remaining text, including provisions that still permit Meta to deploy "web scrapers, bots, robots, spiders, crawlers, and user agents" against domains tied to business accounts for advertising and business purposes. For newsrooms, the practical options remain the same ones available before this week: negotiate a direct licensing arrangement with Meta, rely on robots.txt to try to block its crawlers, or accept the terms as written and hope Meta's narrower reading holds. With AI copyright litigation against Meta still working through the courts in two separate cases, the company's contract language is likely to face continued scrutiny the next time it is updated.