São Paulo court bars Meta from using Brazilian press content to train AI, sets R$1m daily fine
3 нояб. 2026 г. · Собрано автоматически из открытых источников. Человеком не проверялось.
Этот материал доступен только на языке оригинала (английский).
What happened
On 3 November 2026, Judge Guilherme Madeira Dezem of the 44th Civil Court of the Central Civil Forum in São Paulo granted a provisional (precautionary) injunction against Facebook Serviços Online do Brasil Ltda., Meta's Brazilian subsidiary, over Meta's revised Commercial Terms for businesses. The order suspends the website-crawling part of Clause 1 of the revised terms — due to take effect on 20 November 2026 — for members of Brazil's three national press associations: the Brazilian Association of Radio and Television Broadcasters (ABERT), the National Association of Newspapers (ANJ) and the National Association of Magazine Publishers (ANER).
What the order requires
Meta must not use automated software (web scrapers, robots, bots, spiders, crawlers or test credentials) to access, index, cache, analyse, collect, retrieve or crawl the member outlets' domains, web pages or URLs, or use that content to provide, develop, research, maintain, train, fine-tune or feed its AI systems (Meta AI), until final judgment. Meta must also keep the members' accounts, profiles, pages, groups and business tools on Facebook, Instagram, WhatsApp and Threads fully active, with no differentiated, restrictive or punitive treatment for refusing the new terms. Non-compliance carries a daily penalty of BRL 1 million, capped at BRL 100 million.
The court's reasoning
The judge found it likely that scraping news sites to build AI models is unauthorised reproduction under Brazil's Copyright Law (Law No. 9.610/1998, Article 29), which requires prior and express authorisation for reproduction, including storage in a database or on a computer. The decision notes Brazil has no general text and data mining exception for commercial AI training, and that publication on the open web does not license reuse for training. The judge also found the clause abusive as a take-it-or-leave-it contract term: it made continued access to Meta's platforms conditional on surrendering content free of charge, with acceptance by continued use and no partial opt-out. He rejected robots.txt as a sufficient refusal mechanism, calling it a technical cooperation directive without binding legal force, and held that the harm would be irreversible once material was ingested into model weights.
What this does and does not change
Nothing is finally decided. The order was granted without hearing Meta (inaudita altera parte); Meta has five days from service to contest the precautionary application, and the associations have 30 working days to file the main action (case no. 4214874-21.2026.8.26.0100). The relief covers only members of the three associations, not other Brazilian businesses. It is the second court order against the same terms in a week, after an Oslo court temporarily blocked them on 28 October 2026 on an application from Norwegian media groups. The UK Competition and Markets Authority opened a Competition Act investigation into the terms on 30 October 2026.
Источники
- Brazilian media win first round against Meta over AI training demand — Malay Mail (secondary)
- Meta faces R$1m daily fine if it scrapes Brazilian media sites for AI — PPC Land (secondary)