Meta Drops Terms That Could Let It Scrape News Sites for AI

Sometimes the biggest AI stories aren't new models or flashy demos. They sit in the terms of service that nobody reads. This week's example involves Meta, news publishers, and a paragraph of legal text that has now been pulled.

The sequence is short but worth walking through. It shows how AI training data gets claimed, contested and sometimes given back.

What Meta's new commercial terms said

On October 15, Nieman Lab reported that Meta's updated commercial terms of use, set to take effect on October 30, let Meta use web scrapers to crawl and collect content from websites "associated with" business accounts. A business account is the tool that companies, public figures and publishers use to manage Facebook pages, Messenger, Instagram profiles and WhatsApp channels from one place. According to the report, Meta alerted users to the update in early October, in a blog post and in the Meta Business Suite.

The text went further than crawling. Meta could use any content from those domains, web pages or URLs to "provide, develop, improve, research, train, and maintain the integrity of" its products and services. Elsewhere in its terms, Meta defines "AI at Meta" as one of those products. Read plainly, that is a training permission. Three legal experts who reviewed the terms at Nieman Lab's request said that, as written, they could be used to argue that Meta may scrape news sites and train its AI on the content.

The language also let Meta create "test credentials" to access sites. Cornell Law professor James Grimmelmann told Nieman Lab this would technically allow Meta to "sign up as a fake user" and get past registration walls. Meta declined to say whether those credentials would bypass paywalls or other paid content.

Why publishers were alarmed

Nieman Lab noted that virtually every news organization has at least one Meta business account, itself included. So this wasn't a niche clause. It reached almost the whole news industry, and other industries too.

There was also no opt-out ahead of October 30. The notice said that anyone who didn't agree should stop all access or use of Meta's products for business or commercial purposes. Grimmelmann summed up the choice: publishers either trust that Meta doesn't mean the full extent of what the terms say, or they stop using business accounts.

He also described the broadest possible reading. In his words, it could be used to build a comprehensive archive of all content that Meta would then serve as it wanted, directly or through AI. That is his interpretation, not a statement of what Meta planned. But it explains why the clause worried people.

The choice is hard for a small outlet. Nieman Lab first heard about the terms from Stephen Beale, who runs The Steampunk Explorer and saw the notice in his dashboard. His community is, in his words, "totally entrenched on Facebook." He said he was seriously considering shutting his accounts before October 30, even though he knows his site has already been scraped heavily. His anger was about the principle: it struck him as an arrogant move on Meta's part.

The licensing money at stake

Matt Topic, a media and intellectual property attorney at Loevy & Loevy, called it a "remarkable provision to bury" in terms of service. He said it would give Meta access to material that would likely cost many millions of dollars in a licensing market. He added that it isn't clear a court would enforce it, but publishers should assume it would be.

That is the economic heart of the story. If AI companies pay to license publishers' content, a blanket clause that grants the same access through a platform contract could undercut that market, at least for the company writing the terms.

The antitrust angle: tying and Google's precedent

Benjamin Sobel, a law professor at the University of Wisconsin, told Nieman Lab that tethering business products to scraping permissions echoes "tying" in antitrust law. In short, if you want one product you must accept another condition. He compared it to Google's AI search policies. Publishers once had to consent to appearing in AI Overviews and AI Mode to remain in traditional search results. As of August 2026, all sites can opt out.

Nieman Lab also pointed to timing. Meta recently launched a paid subscription, Meta One, which limits how many external links some Facebook and Instagram accounts can post without subscribing. Put another way, Meta has begun charging users to post links to content it could crawl for free. Meta declined to answer Nieman Lab's question about the service, and Press Gazette reported that some news publishers are currently exempt from the link limit.

Meta's first defense, then the removal

Meta declined to answer Nieman Lab's list of questions. A public affairs spokesperson said by email that crawling language had been scattered across several terms, including those on generative AI in ads, and that the update consolidated it in one place. He also said the terms wouldn't change how Meta crawls publishers in the future and that content owners can still manage access with robots.txt.

Legal experts weren't fully reassured. They said the new wording gave Meta an opening to argue that its commercial terms override other blocking instructions. Robots.txt is a plain file that tells bots which parts of a site they may visit, and it has worked mostly on trust.

Then the language was withdrawn. TechCrunch reported on October 17, citing The Verge, that the provision had been removed. Meta's statement said it had "heard you" and that the consolidated crawling terms applied only to its ads and business products, and that it had clarified this in the terms themselves. It added that the change does not alter how Meta uses web crawlers. The sources I read don't say exactly when the text came out.

What the removal does and doesn't settle

It's good that the provision is gone. It's also worth being careful about what that proves. Meta says its practices were never meant to change, and the removal fits that story. But the original text didn't read like a narrow ads clause to three legal experts, and Meta's explanation came after outside scrutiny. My view: it's a lesson in how easily a consolidation exercise can double as a rights grab.

I haven't seen the revised terms myself, so I can't judge how tightly the new wording is drawn. Publishers will probably want to read it closely before October 30. What replaces the deleted provision matters at least as much as the deletion.

Nothing here resolves the underlying legal question either. Nieman Lab noted that any attempt to train AI on content collected this way would still turn on unsettled fair use questions. TechCrunch notes that Meta defeated a high-profile copyright lawsuit last year, while other cases against AI companies, Meta included, continue. Nieman Lab also mentions a class action filed in May by five large book publishers alleging Meta pirated millions of books to train its Llama models.

The bigger pattern: consent by checkbox

The pattern goes beyond one company. AI developers need huge amounts of text, and one legal route is a contract. If a platform you depend on can rewrite its terms, and the only alternative is leaving, consent becomes a very thin idea. This is why the debates about ethical AI development keep returning to data: where it comes from and who agreed to what.

For publishers there's a second bind. Social platforms drive traffic, and AI products can compete for the same audience. Handing one company both the traffic and the training rights is uncomfortable. For small creators like Beale, the platform is the community, so leaving costs far more than a line item.

What to take from this

If you run a website, a newsletter or a small business page, the practical advice is dull but real. Read the update notices from the platforms you use, especially phrases about "associated" sites, "train" and "improve." Check your robots.txt, but don't treat it as a legal shield, since legal experts warned that contract terms could be argued to override it.

As a reader and a writer, I'm glad the provision was pulled and uneasy that it appeared in the first place. Every online profile, every page we publish, is potentially someone's training set. How much of that should be our decision, and how much should depend on whether we can afford to leave? That question will outlast this particular clause, and it touches anyone who posts anything online.