Meta buried a clause letting it train AI on scraped business websites inside new terms taking effect October 30, then deleted the word after Nieman Lab exposed it and publishers pushed back.
Nieman Lab reported that Meta's updated commercial terms, set to take effect October 30, 2026, would have business accounts on Facebook, Instagram or WhatsApp automatically consent to Meta's web scrapers crawling and collecting content from websites "associated with" those accounts, simply by continuing to use the platforms. The original language went further than ordinary site crawling: it said Meta could use that scraped content to "provide, develop, improve, research, train, and maintain the integrity of" its products. Three legal experts who reviewed the terms for Nieman Lab said that wording could let Meta argue it has permission to train its AI systems, including Llama, on news publishers' own websites, no license required.
Matt Topic, a media and intellectual property attorney at Loevy & Loevy, didn't mince words about what that meant. "This is a remarkable provision to bury in terms of service, given that this would give Meta access to material that would likely cost many millions of dollars in a licensing market," he told Nieman Lab. He added that while it's unclear whether a court would actually enforce the clause, publishers should assume it would be.
Then, sometime around October 16 or 17, the word "train" disappeared. According to reporting from TechBuzz and Android Headlines, Meta quietly revised the commercial terms page so the scraping permission now covers advertising and business products, with the AI training language stripped out. Meta didn't announce the change. It didn't need to: publishers and reporters who'd been tracking the original clause noticed the edit almost immediately.
Here's the thing: Meta says nothing actually changed about how its crawlers behave. The company maintains it already respects robots.txt and other standard web protocols, and a spokesperson framed the edit as a clarification rather than a retreat. But removing a word from a public terms page is also a tell. You don't rewrite a clause that's drawing attorney commentary and publisher alarm unless you're worried about where that trail leads, legally or in the court of public opinion.
Meta glasses lawsuit says intimate footage trained AI via Kenyan contractors A newly amended class action accuses Meta of routing intimate footage from its Ray-Ban smart glasses to Kenya-based contractors at Sama, who allegedly viewed and labeled the recordings to train Meta's AI. The August 31 filing expands the case to cover bystanders who never bought the glasses at all, arriving weeks after Meta's $18 billion... - Meta Ray-Ban glasses privacy lawsuit Kenya contractors - intimate footage labeled training AI without consent
The News/Media Alliance isn't buying the clarification as a full fix. The trade group pointed out that training was removed from the list of stated uses but never explicitly banned. Meta hasn't actually promised it won't train on scraped publisher content. It's just stopped saying so in writing. That's a meaningful distinction for any publisher weighing whether to block Meta's crawlers outright.
This didn't happen in a vacuum. Meta is already a defendant in a separate, and much bigger, copyright fight: publishers including Elsevier, Cengage, Hachette, Macmillan and McGraw Hill, along with author Scott Turow, sued the company in Manhattan federal court alleging it pirated millions of books and journal articles to train Llama. That case is still working its way through discovery. Against that backdrop, a terms-of-service clause that reads like a forward-looking consent mechanism for the exact same kind of scraping looks less like an accident and more like Meta trying to paper over, in advance, the thing it's currently being sued for doing without permission.
The timing also lines up with a broader reckoning across the AI industry over how training data gets sourced. The New York Times' lawsuit against OpenAI and Microsoft, still active after more than two years, forced OpenAI to hand over a 20-million-conversation sample under a January 2026 order from Judge Sidney Stein. Nearly 400 local newspapers have separately sued OpenAI and Microsoft over scraped articles. Every one of these fights turns on the same question: did the company get permission, or did it just take the content and sort out the legal exposure later. Meta's edited clause suggests the company would rather not find out in court whether "it's in our terms of service" counts as permission.
For publishers, the practical lesson isn't really about Meta specifically. It's that business-platform terms of service have become a quiet front in the AI training data war, and the fix for an unwanted clause isn't to assume good faith, it's to read the terms. Meta pulled the word "train" only after Nieman Lab put a spotlight on it and three attorneys said the quiet part out loud. Nobody should bet that's the last clause like it. Also read: ForgeCode claims it beats Claude Code and Codex, but the fine print tells another story • Nvidia Bets Local AI Agents Can Dodge the Cloud Trust Problem • Barclays downgrade wipes out billions in Dell, HPE and Oracle shares
This article is posted in Technology News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.