Meta Axed the AI Word From Its Terms. The Scraping Stands

Photo: Meta's headquarters sign at 1 Hacker Way, Menlo Park. Nokia621 via Wikimedia Commons (CC BY-SA 4.0).

Meta tried to quietly rewrite its commercial terms so that staying on Facebook, Instagram or WhatsApp for business would count as permission for Meta to crawl your website — and train its AI on whatever it found there. Newsrooms caught the clause within 48 hours. And now, in a Saturday-morning edit that The Verge first spotted, the AI-training wording has vanished from the document.

Don't celebrate yet. The crawling authorisation — the automated software, the test credentials, the periodic re-crawls, the 30 October effective date — is all still there. Meta deleted one word, "train". It kept the whole machine. This is the platforms-versus-publishers fight entering its brazenest phase yet: the house tried to take the cut first, in the terms of service, and only retreated when lawyers called it "remarkable" to its face.

The clause nobody was supposed to read

The whole thing surfaced because of a niche publishing site about retro science fiction. Stephen Beale, the founder and editor of The Steampunk Explorer, noticed a notice in his Meta Business Suite dashboard flagging updated commercial terms, effective 30 October 2026. Most people click through those. Beale read it — and it read, to him, like something different from the usual anxieties about Meta training on content people post to its platforms. "I know my site has been scraped to death," he told Nieman Lab's Andrew Deck and Hanaa' Tameez, but this claimed a right to grab information from wherever a business user has a presence.

What Beale found was this: anyone using Meta's products "for a business or commercial purpose" would, by continuing to use them after 30 October, authorise Meta "to use automated software (such as web scrapers, bots, robots, spiders, crawlers, and user agents) and create test credentials to access, index, cache, analyze, collect, retrieve, or crawl the domains, web pages, or URLs (and any content available through them) that are associated with your products or services" — and to "periodically pull updates" from them. Then the sentence that mattered: Meta could use any of that content to "provide, develop, improve, research, train, and maintain the integrity of" its products.

We checked Meta's own definition of its products across three independent write-ups of the terms this morning, and "AI at Meta" is listed in all of them — which is the hinge the entire training argument turns on. It is the same AI division that has been racing to ship frontier models — we covered Meta teasing its new frontier model Muse Wonder with a watermelon — and frontier models are ravenous for exactly the kind of high-quality text newsrooms produce. There was no opt-out. The terms said, bluntly: "If you do not agree to the updated terms, you may stop all access or use of our Products for business or commercial purposes." The current terms date from 6 March 2024 and mention crawling zero times. The new version was a deliberate addition, and it landed two weeks before it takes effect.

Nearly every newsroom is in the net

Here's why Nieman Lab, and not some trade publication for advertisers, broke this story. Almost every news organisation runs at least one Meta business account — to manage its page, run ads, use Business Suite. The Lab noted, with obvious amusement, that it holds one itself. Which meant the clause could, in principle, sweep in virtually every newsroom on earth.

Three legal experts who reviewed the wording for Nieman Lab did not mince words. Matt Topic, a media and intellectual property attorney at Loevy & Loevy, called it "a remarkable provision to bury in [the] terms of service," adding that the material it covers "would likely cost many millions of dollars in a licensing market." Topic's second line is the one publishers should sit with: "It's not clear whether a court would enforce this, but publishers should assume that it would be enforceable."

James Grimmelmann, a professor of digital and information law at Cornell Law School, went further: "For news publishers, this clearly represents permission to Meta to train AI on any content from their sites and possibly to deploy some of that content, even in competition with the publishers themselves." On the broadest reading, he warned, Meta could use the terms "to develop a comprehensive archive of all content that Meta would then serve as it wanted to directly or through AI."

And then there was the irony Nieman Lab flagged that should embarrass everyone involved: Meta has recently begun charging some users — through its paid "Meta One" subscription — for the privilege of posting links to their own websites on Facebook and Instagram. Press Gazette reported that some news publishers are currently exempt, but the direction of travel is unmistakable: pay to post your links, while Meta claims the right to crawl the same pages for free.

Meta says it's only checking prices

Meta's spokespeople spent the last 48 hours insisting the clause meant nothing like what it said. Andy Stone, Meta's communications chief, wrote on X that the update was "designed to make sure we're able to pull accurate information about products for sale on third-party websites when making on-platform ad and content recommendations," adding that it was "not about scraping news sites." A company spokesperson, Daniel Roberts, told reporters: "This is not about scraping news sites, and these updates don't give us new rights that supersede those preferences." Content owners, Meta said, can still use robots.txt to signal preferences, and there was "no change to how Meta uses web crawlers."

That defence has a hole you could drive a crawler through. The terms, as written, made no mention of robots.txt at all. They said nothing about limiting the crawling to ads or product pages. And they explicitly allowed Meta to "create test credentials" — a power that, as several commentators noted, could reach behind logins and paywalls. Meta declined to answer Nieman Lab's list of detailed questions, including whether the "test credentials" power could bypass paid content.

Meanwhile the publishers' lobby came out swinging. Danielle Coffey, president and CEO of the News/Media Alliance, which represents US publishers, called the terms "outrageous and probably illegal," and said they "cannot be allowed to be implemented on October 30." She has a point about timing: whatever a court would eventually decide, publishers had fourteen days to either accept a sweeping contractual licence or abandon the platforms their audiences live on.

Forty-eight hours later, "train" disappeared

On Saturday 17 October, The Verge reported that the disputed language had been removed from the terms page. TechCrunch's Anthony Ha confirmed the edit: the clause that let Meta use content to "train" its products is gone. Meta's new statement, issued after the edit: "We've heard you and want to clarify: the crawling terms we consolidated only applied to our ads and business products, and we've clarified that in the terms themselves. It does not change how Meta uses web crawlers."

That is a retreat, and a fast one — two days from newsroom outrage to a rewritten legal document. But read the follow-up Nieman Lab published the same day, also by Deck and Tameez: "Legal experts warn Meta's new terms still grant broad scraping permissions, despite language changes." Allowing scraped content to be used for all "advertising and business products," the piece notes, "is a substantial expansion of Meta's permissions." The single most alarming word was cut. The authorisation to crawl business-associated websites with automated tools, create test credentials, and periodically re-pull updates — all of it — remains, with the same 30 October date and still no genuine opt-out.

Beale, the editor who started all this, told Nieman Lab his readers are "totally entrenched on Facebook," and that he is seriously considering shutting down his Meta accounts before the deadline anyway. That is where this leaves every publisher: trusting a two-day-old clarification, or leaving the platform.

Our judgment: a botched consent grab, and a warning for next time

Let's be honest about what this episode was. Nobody writes a clause authorising scrapers, bots and test credentials across all business-associated domains, in a contract affecting millions of businesses, by accident. This was a consent grab: an attempt to convert continued participation on Meta's platforms into a contractual waiver of the licensing rights publishers are currently fighting over in court — including, note, in Meta's own copyright litigation, which it won a high-profile round of last year.

What makes it colossally cynical is the placement. Not a negotiation. Not a licensing offer. A paragraph in Section 1 of the commercial terms, discovered by a steampunk blogger in his dashboard, two weeks before it took effect. The publishers Meta already pays for AI licensing deals — CNN, Fox News, USA Today, People Inc. are among the outlets with agreements — would have watched the company grant itself for free what it charges others millions for. No wonder the News/Media Alliance reached for "probably illegal."

To Meta's credit — or to the credit of the newsrooms that screamed — the retreat was real and quick. This is a company already fighting on multiple legal fronts; a judge recently warned Meta it faces a possible $40B penalty, telling it "you rolled the dice", and its trust record with data is thin — its Muse assistant's privacy scare was blamed on "syncing notifications". But notice what survived: the crawling architecture, the test-credentials power, the no-opt-out structure, the October deadline. The next version of this clause will be subtler. Publishers should treat this as the trial run it clearly was: audit what your Meta business account is "associated" with, tighten your robots.txt, and decide now what leaving would cost you, before the next edit gives you fourteen days to decide. The house tried to take the cut first. Next time it won't telegraph the punch.

Sources: Nieman Lab (15 Oct — Andrew Deck & Hanaa' Tameez, original report; 17 Oct — follow-up on the revised terms); TechCrunch (17 Oct — Anthony Ha, removal report); The Verge (17 Oct — removal report, via TechCrunch); The Register (16 Oct — full clause wording); Press Gazette (Meta One link limits, via Nieman Lab). All quotations verbatim from their cited outlets, under 25 words each.

— TechPulse USA, Cupertino and New York desks

About TechPulse USA: TechPulse USA is an independent technology newsroom covering AI and tech across the US, Europe, Canada and Australia. Every story is researched from multiple reputable sources and every fact is verified before publication — no filler, no spam.

Update, Oct. 18: Meta answered — “We’ve heard you”

Meta has now responded, and its answer is a clarification, not a climbdown. “We’ve heard you and want to clarify: the crawling terms we consolidated only applied to our ads and business products, and we’ve clarified that in the terms themselves. It does not change how Meta uses web crawlers,” spokesperson Daniel Roberts said, via aiparabellum. Andy Stone rejected the news-site-scraping reading outright, saying the terms were “designed to retrieve accurate product information for advertising and content recommendations” (via anews). The revised terms now restrict crawling to purposes “to develop, improve, research and maintain advertising and business products,” and the new terms take effect on October 30. Cornell professor of digital and information law James Grimmelmann was named by The Verge among the legal interpreters who read the original wording as an AI-training licence (via CashIn News) — the reading Meta is now trying to close.