A state court judge in São Paulo on November 3 suspended the clause in Meta's revised Commercial Terms that would allow the company to crawl the websites of its business users, and ordered Facebook Serviços Online do Brasil Ltda. to stay off the portals of members of Brazil's three national media associations. The order also requires Meta to keep those members' Facebook, Instagram, WhatsApp and Threads accounts running if they refuse the new terms, under a penalty of BRL 1 million for each day of non-compliance, capped at BRL 100 million.

Judge Guilherme Madeira Dezem of the 44th Civil Court of the Central Civil Forum granted the measure without hearing Meta, on a joint application from the Brazilian Association of Radio and Television Broadcasters (ABERT), the National Association of Newspapers (ANJ) and the National Association of Magazine Publishers (ANER). He signed the decision at 15:36 local time. It is the second court order against the terms in a week, after an Oslo court temporarily blocked them on October 28, and its reasoning rests on copyright and contract law rather than competition law. Quotations in this article come from an unofficial English translation of the decision, which states that the Portuguese original prevails in case of discrepancy.

In Short

Meta wrote new rules for businesses that use Facebook, Instagram and its other apps, and Brazil's associations of newspapers, magazines and broadcasters said those rules would let Meta copy their websites to build its AI without asking or paying. A judge in São Paulo told Meta it cannot apply that part of the rules to the associations' members and cannot punish them for saying no, with a penalty of 1 million reais for each day it disobeys. Meta has not yet put its side to the court, so this is a temporary order that holds until the full case is decided.

What the court ordered

The operative part of the decision has four items. The first suspends, "until final judgment in the main action, the effectiveness of the final part of Clause 1 of the respondent's updated Commercial Terms (with an announced effective date of 20/11/2026)," along with any provision that makes the members' access to their accounts "conditional on authorising the scraping and collection of data on their portals and external websites."

The second orders Meta to keep "fully active, intact and in normal operation all accounts, profiles, pages, groups and business tools" used by the members on Instagram, Facebook, WhatsApp and Threads, "without applying any differentiated, restrictive, punitive or retaliatory treatment as a result of their non-acceptance of the new external collection terms." That item answers the central complaint about the terms: that the only way to reject them was to leave.

The third is the broadest. Meta must refrain, directly or through third parties acting on its behalf, from using "any automated software (web scrapers, robots, bots, spiders, crawlers and user agents) or test credentials to access, index, cache, analyse, collect, retrieve or crawl domains, web pages or URLs belonging to the applicants' members," and from using that content "to provide, develop, research, maintain, train, fine-tune or feed its artificial intelligence systems (Meta AI), until final judgment in the case." The list of verbs and tools mirrors, almost word for word, the clause Meta added to its terms.

The fourth sets the penalty. The judge imposed "a DAILY PENALTY of BRL 1,000,000.00 (one million reais)" for non-compliance with any of the orders, with a cap of 100 million reais. At the daily rate, the cap would be reached after 100 days of breach.

The procedural clock is short on both sides. Meta has five days to contest the precautionary application under Article 306 of the Code of Civil Procedure, and the decision warns that if it does not, "the facts alleged by the applicants will be presumed to be true." The associations have 30 working days from the date the measure is carried out to file their main claim in the same proceedings, under Article 308. The case number is 4214874-21.2026.8.26.0100.

The clause at issue

The dispute concerns Section 1 of Meta's Commercial Terms, headed "Licences and Rights You Provide to Meta." When Meta published the revised terms in mid-October, PPC Land reported that the new text let Meta crawl business websites to train AI: business users would authorise Meta, or any entity acting on its behalf, to use web scraping tools and to create test credentials to access domains associated with their products or services, and to use the content gathered. Acceptance would come from continued use of Meta's business products, with no partial opt-out.

The court's account of the text's history matches that record and adds a date. According to the decision, the original wording, "made available to users on 16 October 2026, expressly contained the word 'train'" when describing the purposes of the collection. "Following the immediate negative international reaction, the company made a further drafting change on 19 October 2026, deleting that specific word but leaving intact all the operative verbs and mechanisms for large-scale extraction," the judge wrote. PPC Land's own check of the UK English preview on October 18 found that the training sentence was still in place a day after reports of its removal, which is consistent with a change on October 19.

The judge also pointed to "item 5," which retains "the express provision for the integration of generative artificial intelligence features." The decision does not quote that provision. The revised terms also add an AI Features section under which Meta makes advertisers solely responsible for its AI ad edits, a part of the text the Brazilian case does not challenge. And he singled out "item 7," under which "mere continued use of Meta's networks and tools after notification will amount to full and irrevocable acceptance." In the court's words, that "establishes an entirely spurious presumption of consent, eliminating any possibility of selective refusal (opt-out) for a user who wishes to keep its institutional accounts without surrendering its external collection."

One fact the decision does not mention is Meta's latest statement on timing. The order treats November 20 as the date the terms take effect, the date Meta set when it delayed the website-crawling terms on October 23. On October 28, Andy Stone, Meta's vice president of communications, said the company would not change its policies without a full consultation with industry groups, that the consultation could run past November 20, in which case Meta would delay implementation, and that the existing terms would stay in place until then. The decision describes November 20 as "peremptorily set" and does not refer to that pause.

Why a California forum clause did not apply

Before turning to the merits, the judge had to decide whether a Brazilian court could hear the case at all. According to the decision, item 6.d of the Commercial Terms provides that commercial disputes arising outside the United States are to be resolved exclusively before the US District Court for the Northern District of California or a state court in San Mateo County, under California law.

The court declared that stipulation ineffective for acts carried out in Brazil. Facebook Serviços Online do Brasil Ltda. is incorporated in Brazil with its registered office on Avenida Brigadeiro Faria Lima in São Paulo, "earning substantial advertising revenue from the traffic and engagement of the Brazilian public," the judge wrote. Article 21(I) of the Code of Civil Procedure gives Brazilian courts jurisdiction over defendants domiciled in the country, whatever their nationality.

The decision layers several further grounds on top. Article 11 of the Marco Civil da Internet, Law No 12.965/2014, requires Brazilian law to be observed where any act of collection or processing takes place in Brazil. Article 8 of the same law renders null any clause in a contract of adhesion that does not offer the Brazilian forum as an alternative. The judge also cited the Federal Supreme Court's ruling in ADC 51, which in his account prohibits "reliance on foreign corporate headquarters as a subterfuge to frustrate the effectiveness of national jurisdiction." Requiring Brazilian news organisations to litigate in California "would constitute an insurmountable practical and economic barrier," according to the decision.

Who can sue for the press

The second threshold question was standing. ABERT, ANJ and ANER brought the case for their members, which the decision describes as "thousands of radio and television broadcasters, hundreds of print and digital newspaper publishers and dozens of magazine publishers across every state of the Federation." The court did not require a list of members or individual authorisations.

The reasoning draws on Brazil's collective-redress framework. The judge classified the claim as one protecting homogeneous individual rights under Article 81 of the Consumer Protection Code: each outlet owns its own copyrights, but "the conduct challenged against the respondent is entirely uniform and identical," a single set of standard terms imposed on every member. Under Article 82 of that code and the Public Civil Action Law, associations incorporated for more than a year with a matching institutional purpose can act in their own name. The Supreme Court's requirement for express authorisation by a general meeting, set in Extraordinary Appeal No 573.232, applies to ordinary representation and "does not extend to the collective-redress microsystem," the judge wrote.

The core of the decision is a finding that crawling news sites to build AI is reproduction under Brazilian law. Article 29 of the Copyright Law, Law No 9.610/1998, requires prior and express authorisation for any use of a work, including reproduction, "inclusion in a database and storage on a computer," and "any other forms that may yet be invented."

The judge then described the technical process. "Automated scraping (web scraping) of news portals, television and radio broadcasters' websites and editorial archives, through the making of technical copies, fragmentation into computational units (tokens) and storage on servers for the training and refinement of large language models (LLMs) or predictive artificial intelligence systems, is undeniably an act of reproduction and use of protected intellectual works." He drew a line between that and search: "This is not a matter of transient indexing of links to direct traffic, but of the actual incorporation of the authors' expressive substance into the respondent's databases."

Two further points close off the defences a platform might raise. Article 31 of the same law provides that authorisation for one form of use "does not extend to any other," so publication on the open web does not license reuse for training. And Brazil has no equivalent of the European Union's text and data mining exceptions. "Brazilian law provides no general text and data mining exception for the commercial training of artificial intelligence models," the decision says, and the statutory limitations in Article 46 contain nothing permitting "the indiscriminate extraction of private news portals to feed systems that operate as competitors to, or functional substitutes for, the media outlets themselves."

The judge added the international three-step test, from Article 9(2) of the Berne Convention and Article 13 of the TRIPS Agreement: any exception must be confined to special cases, must not conflict with normal exploitation of the work and must not unreasonably prejudice the rightholder. "The conduct described on the part of the respondent infringes each of those criteria," he wrote. He also cited a December 19, 2024 ruling of the court's 2nd Private Law Chamber, Civil Appeal 1003782-77.2023.8.26.0394, which held that the liability of an internet application provider for copyright infringement is strict.

The contract reasoning

The second line of argument concerns the form of the agreement. The Commercial Terms are "a typical contract of adhesion, drafted unilaterally by the supplier of the technological infrastructure and imposed on a 'take it or leave it' basis," according to the decision. Article 424 of the Civil Code voids clauses in such contracts that require the adhering party to waive in advance a right arising from the nature of the transaction. "Requiring media outlets to hand over their copyright assets free of charge and universally as a toll for remaining connected to the channels through which today's public consumes news constitutes a manifest abuse of rights and unjust enrichment," the judge wrote, citing Articles 187, 421 and 422.

The more technical point concerns third-party material. Clause 1 requires the business to declare that it holds all rights needed to grant the licence, including for derivative works. News sites carry agency photographs, licensed illustrations and contributors' texts whose use was authorised only for publication. That, in the court's description, puts media companies in "an inescapable legal trap: either they breach the platform's terms because they lack the power to license works belonging to others, or they become exposed to recourse liability towards the original copyright holders." The judge concluded that the object of the clause is unlawful "because performance is legally impossible," under Articles 104(II) and 166(II) of the Civil Code.

That argument is not specific to journalism. Any business website that displays licensed photography or third-party content would face the same warranty under the clause. The relief, however, covers only the members of the three associations.

Why the court acted before hearing Meta

Interim relief in Brazil requires a likelihood of the right and a risk of harm. On the second, the judge focused on how machine learning works. Once protected material has been ingested for pre-training or fine-tuning, "data incorporated into the synaptic weights and mathematical parameters of models can hardly be segregated, audited or removed afterwards, creating a fait accompli," according to the decision.

He rejected two alternatives that would have left the terms in place. Leaving Meta's platforms is not viable, he wrote, because "a substantial proportion of Brazilian society primarily uses the respondent's social networks and messaging applications to access daily news," citing studies filed by the associations that the decision does not name. Withdrawal would cause "an immediate collapse in editorial reach."

The second alternative was the one Meta itself had offered. When it postponed the terms in October, a Meta spokesperson told Reuters that content owners would still be able to use robots.txt to signal their preferences. The court was not persuaded. "The exclusion protocol is a mere technical cooperation directive without binding legal force, whose effectiveness depends on the behaviour of the crawler," the decision says, and its generalised application "may impair the generation of hyperlink previews and the ordinary indexing of articles which the outlets themselves have a legitimate interest in preserving." On Meta's platforms, link previews are generated by fetches from FacebookExternalHit, so a blanket block can cost a publisher the cards that accompany shared links.

The question of whether a single file can carry a training preference is live elsewhere. On October 30, Google and Bing backed one robots.txt rule to opt out of AI training, alongside OpenAI and Anthropic; Meta was not among the companies that announced support. The São Paulo decision treats any such signal as cooperation rather than a legal safeguard.

On reversibility, the judge said the measure "merely preserves the state of affairs that has prevailed unchanged between the parties for years under the contractual terms in force since March 2024." Should Meta's clauses later be found lawful, "the respondent may exercise its contractual prerogatives at the appropriate time."

The size of the penalty

The decision explains the BRL 1 million daily figure in a single sentence. "I justify the high amount of the daily penalty by the fact that the respondent has shown itself to be particularly reluctant to comply with decisions handed down in other proceedings, choosing when to comply and when not to comply," the judge wrote. The decision does not identify those proceedings.

Brazilian authorities have penalised Meta over its terms before. In January, the competition authority CADE imposed interim measures that suspended Meta's ban on third-party AI chatbots on WhatsApp, with daily fines of R$250,000 for non-compliance. On April 23, CADE announced that its Tribunal had unanimously upheld that preventive measure and the daily fine against WhatsApp LLC and Facebook Serviços Online do Brasil Ltda., after the rapporteur concluded that the companies had not complied with the Tribunal's decision, according to CADE. The São Paulo penalty is four times CADE's daily rate. Whether the judge had the CADE case in mind is not stated.

How it compares with Oslo, London and Paris

The São Paulo order is the latest in a sequence of moves against the same terms. The Oslo District Court acted on October 28 on an application from the Norwegian Media Businesses' Association on behalf of ten companies, including NRK, TV 2 and Schibsted, and, according to AFP, without hearing Meta. The association did not publish the written decision when it announced the ruling. Swedish news agency TT reported that the Oslo suspension holds until the court examines whether the terms amount to an abuse of a dominant position.

The Brazilian applicants are three associations rather than individual companies. In a joint note dated November 3, reproduced in full by the trade site Tudo Rádio, ABERT, ANJ and ANER described the decision as the first known court ruling in the world granted to a coalition of media associations against Meta's new rules. ANJ's executive president, Marcelo Rech, called the injunction a sign that there are limits to the abuse of economic power by a large digital platform, and a first step towards platforms and AI models refraining from appropriating third parties' content, according to the note. Neither Tudo Rádio nor Times Brasil, which also reported the decision on the evening of November 3, carried a response from Meta.

The legal routes differ. In the United Kingdom, the Competition and Markets Authority opened a Competition Act investigation on October 30 into whether Meta's existing terms or revisions to them amount to an abuse of dominance. In Brussels, the European Publishers Council asked the European Commission to halt the terms before November 20. In the United States, the News/Media Alliance gave Meta until October 26 to answer on the crawling terms. The São Paulo judge reached his result without ruling on dominance. The associations' application describes Meta's "undisputed dominant position," but the grounds of the order are the Copyright Law, the Civil Code and procedural rules on jurisdiction and collective actions.

France sits on a separate track. On October 29, Meta lost its appeal in Paris and must share revenue data with French press groups, in a neighbouring-rights case that predates the new terms. Together, the cases put Meta in front of courts or regulators in at least four countries over its relationship with news publishers within seven days.

What the order does not settle

The decision is provisional by design. It was issued "inaudita altera parte," without Meta being heard, and none of the company's arguments has yet been tested. Meta can contest within five days of service, and the associations must file the main action within 30 working days. The findings on copyright and contract law are made at the level of plausibility required for urgent relief, not as a final ruling on the merits.

Its reach is limited to the members of ABERT, ANJ and ANER. A Brazilian retailer, advertiser or independent publisher outside those associations is not covered, and the decision does not address the terms as they apply to other business users in Brazil or elsewhere.

The order's wording also raises a practical question. Item c bars Meta from using any automated software to "access, index, cache, analyse, collect, retrieve or crawl" members' domains, without limiting that prohibition to collection for AI. Read literally, it reaches the same automated fetches that the decision elsewhere describes as producing link previews the outlets want to keep. The decision does not say how the two fit together.

Finally, the timing may already have moved. If Meta's consultation runs past November 20, the clause the court suspended will not have taken effect anywhere on that date. The order nonetheless stands as a court-imposed condition on whatever version Meta eventually brings into force in Brazil, at least for these members, until the main action is decided.

Why this matters for marketers and publishers

Meta's Commercial Terms are not a publisher contract. They apply to any business that runs ads, manages a Page or uses Meta Business Suite, which is why PPC Land's earlier reporting found that Meta granted itself the right to crawl advertisers' websites for AI. The Brazilian order protects only one group of those businesses, but its reasoning addresses the clause's mechanics in general terms: acceptance by continued use, warranties over material a business does not own, and the limits of robots.txt as a refusal mechanism.

For publishers that depend on Facebook, Instagram and WhatsApp for referral traffic, the second item of the order matters as much as the first. It separates the decision to refuse the crawling licence from the decision to stay on the platforms, which is the link Meta's terms make and the one every objector in this dispute has challenged. When Meta first postponed the terms, PPC Land noted that the date changed but the clause barely did. A São Paulo judge has now put a price on applying it unchanged to Brazil's organised press: BRL 1 million a day.

Timeline

Summary

Who: Judge Guilherme Madeira Dezem of the 44th Civil Court of the Central Civil Forum in São Paulo, ruling on an application from ABERT, ANJ and ANER, the national associations of Brazilian broadcasters, newspapers and magazine publishers, against Facebook Serviços Online do Brasil Ltda., Meta's Brazilian subsidiary.

What: An interim precautionary order, granted without hearing Meta, that suspends the website-crawling part of Clause 1 of Meta's revised Commercial Terms for the associations' members, requires Meta to keep their Facebook, Instagram, WhatsApp and Threads accounts active if they refuse the terms, and bars Meta from crawling their sites or using their content for Meta AI, under a penalty of BRL 1 million a day capped at BRL 100 million.

When: Signed on November 3, 2026, ahead of the November 20 effective date Meta announced on October 23. Meta has five days from service to contest; the associations have 30 working days to file the main action.

Where: São Paulo, Brazil. The court declared ineffective the terms' clause sending disputes to courts in California, and the order covers the domains of the associations' members across Brazil.

Why: The court found it likely that crawling news sites to build AI is unauthorised reproduction under Brazil's Copyright Law, which has no text and data mining exception, and that imposing the licence through a take-it-or-leave-it contract, with acceptance by continued use and warranties over third-party content, is abusive under the Civil Code. It acted before hearing Meta because training on scraped material cannot be undone and because leaving Meta's platforms or relying on robots.txt were not, in its view, workable alternatives for the press.