Meta Deletes “Train” From Its Terms, and a Judge Wants to Know Who Approved the Torrents

Meta removed the word "train" from its Commercial Terms for business accounts, effective October 30, after publishers objected, but kept the crawling permission and limited permitted uses to "advertising and business products" in a revision published October 16. Meta spokesperson Daniel Roberts told The Verge that "the crawling terms we consolidated only applied to our ads and business products, and we've clarified that in the terms themselves," while University of Wisconsin law professor Benjamin Sobel told Nieman Lab that Meta should write a promise to honor robots.txt into the terms because otherwise "you can still express the preference, and unless it's a legally enforceable preference, it can still be ignored all the same." UK News Media Association's Theo Bamber called the terms a "shameless land grab," and Search Engine Journal reported the UK English version of Meta's self-serve ad terms still said "train" on October 19.

Meta spent this week answering two versions of the same question: who said you could have that? One version is about websites and comes from news publishers. The other is about books and comes from a federal magistrate judge. Meta’s answers are not very satisfying in either case, though they fail in different ways. The word that left Start with the websites. Meta has new Commercial Terms for business accounts that take effect on October 30, and as we wrote last week https://rews.cc/a/meta-s-new-fine-print-asks-every-business-website-to-feed-it-03a715 , they let Meta crawl websites associated with those businesses. The original clause, as Nieman Lab printed it, said Meta “may use any content” it collected “to provide, develop, improve, research, train, and maintain the integrity of” its products, services and features. If you run a news site with a Facebook page, you are a business account. And “train” in a 2026 contract written by an AI company means one thing. Publishers objected, and on October 16 Meta published a revision https://www.niemanlab.org/2026/10/legal-experts-warn-metas-new-terms-still-grant-broad-scraping-permissions-despite-language-changes/ . It removed “train” from the list and limited the permitted uses to “advertising and business products” instead of all of Meta’s products. The crawling permission stayed. So did the part covering pages “made available through your or users’ use of Meta Company Products,” which could include links that other people share. And business accounts still can’t opt out before the terms kick in. Meta’s position is that this was all a misunderstanding. Spokesperson Daniel Roberts told The Verge that “the crawling terms we consolidated only applied to our ads and business products, and we’ve clarified that in the terms themselves,” and that “this is not about scraping news sites, and these updates don’t give us new rights that supersede those preferences.” “Those preferences” means robots.txt. A Meta spokesperson told Nieman Lab that news sites, like any other content owners, can still block the crawler that way. Here is the problem with that. Robots.txt is a small text file on a website that tells automated crawlers which pages they may fetch. Nothing makes a crawler obey it. Obeying is good manners. Terms of service are a contract, and contracts can be enforced. So suppose, hypothetically, a publisher’s robots.txt says “Meta, keep out” while the contract the publisher accepted says Meta may crawl. A lawyer could argue the contract wins, especially since the contract never mentions robots.txt at all. That is roughly what Ars Technica’s Ashley Belanger https://arstechnica.com/tech-policy/2026/10/meta-drops-the-word-train-from-its-commercial-tos-after-ai-backlash/ reported experts warning about. University of Wisconsin law professor Benjamin Sobel told Nieman Lab the simple fix would be for Meta to write a promise to honor robots.txt into the terms. Without one, he said, “you can still express the preference, and unless it’s a legally enforceable preference, it can still be ignored all the same.” The other lawyers Nieman Lab spoke to were not impressed by the deletion either. “I think ‘develop’ is broad enough to include ‘train,’” said attorney Matt Topic of Loevy & Loevy. Cornell’s James Grimmelmann said the revision “does nothing to restrict” the language about sites “associated with your products or services.” If you strike one verb from a list of near-synonyms, you are left with a list of near-synonyms. Even the deletion happened unevenly: Search Engine Journal reported https://www.searchenginejournal.com/meta-removes-training-crawl-clause-commercial-terms/593584/ that the UK English version of Meta’s self-serve ad terms still said “train” on October 19, and a check the next day found it fixed. The publishers’ trade groups are now asking regulators to step in. Theo Bamber of the UK’s News Media Association called the terms a “shameless land grab” built on the “worst kind of arrogant tech platform bullying,” and asked British regulators to “stand up for British creators and the business community” by “preventing Meta from setting this dreadful precedent.” The European Publishers Council has already gone to the European Commission. “Publishers must be able to decide whether and how their journalism is used, and negotiate fair payment,” said its executive director, Angela Mills Wad. “Burying such far-reaching claims in platform terms is no substitute for consent.” Danielle Coffey, who runs the News/Media Alliance, the largest news publishing trade group in the US, called the terms “outrageous and probably illegal.” She also named the actual mechanism. Meta’s leverage is its audience. If you want to advertise to Facebook and Instagram users, you accept the terms, and the terms come with a crawler. Coffey called Meta’s “decision to withhold access to its audience unless businesses allow Meta to scrape content from their websites” an “appalling, bad-faith policy, and a clear abuse of the company’s market power.” “It cannot be allowed to be implemented on October 30,” she said. Nobody decided anything Now the books. Meta is defending a set of consolidated copyright cases in the Northern District of California, led by Kadrey v. Meta, over its use of pirated books to train its Llama models. One of the plaintiffs is Entrepreneur Media, the publisher of Entrepreneur magazine, which accuses Meta https://www.courtlistener.com/docket/71883091/entrepreneur-media-llc-v-meta-platforms-inc/ of contributing to piracy by seeding works to other people while it torrented them. That matters because torrenting usually means uploading pieces of a file while you download it. By the plaintiffs’ count, as TorrentFreak reported https://torrentfreak.com/judge-orders-meta-to-name-employees-who-approved-torrenting-annas-archive-and-libgen/ , Meta downloaded 80.6 terabytes from LibGen and 35.7 terabytes through Anna’s Archive. That adds up to 116.3 terabytes of shadow-library material. Entrepreneur Media asked Meta a simple question: who decided to do this? Meta’s answer, as Belanger also reported https://arstechnica.com/tech-policy/2026/10/meta-ordered-to-name-execs-who-approved-torrenting-for-ai-training/ , was that the question doesn’t really have an answer. It called the publisher’s theory “speculation” and said there was “no basis” for thinking “someone at Meta must have ‘approved’ a ‘decision’ to perform a seeding activity that specific mitigations were deployed to avoid.” According to Meta, leadership didn’t usually make “formal” decisions about acquiring datasets back in the Llama 1 and 2 days. That “initial informal approach” only later gave “way to more structured review and approval processes.” Approvals came out of “collaborative” discussions, so there was no “single decision-maker.” Meta offered to provide names only, without roles or dates. Magistrate Judge Thomas S. Hixson Ars spells it Hixon rejected all of this in an order dated October 19 https://www.digitalmusicnews.com/2026/10/20/meta-ai-lawsuit-shadow-library-discovery/ . The request never said formal decisions, he wrote. On the collaborative point, the motion did “not presume that Meta had a single decision maker; it asks about all the decision makers.” Coverage of the order quotes him going further: “If a decision is fairly attributable to a group or a team, everyone in the group or team who participated in making that decision is responsive to this rog.” Meta now has to name everyone who “made, authorized, or approved any decision to obtain or download the shadow libraries,” give their approximate dates of involvement and say which decision each one made. It also has to explain what happened to the pirated works that weren’t used for training. This is a discovery fight and not a ruling on liability. Meta already won on fair use for training in the oldest of these cases. But the piracy questions are separate, and US statutory damages run from $750 to $30,000 per work, or up to $150,000 if the infringement was willful. When willfulness is on the table, the names of the people who decided start to matter. That explains both why the plaintiffs want them and why Meta would prefer that nobody had decided anything. So Meta’s terms of service make no promise about robots.txt, and Meta says its early data decisions were never formally made by anyone. Those terms still take effect on October 30, and presumably someone approved them.