Sao Paulo Judge Orders Meta to Stop Scraping News for AI or Pay 1 Million Reais a Day

A Sao Paulo court has ordered Meta to stop scraping Brazilian news outlets for AI training and suspended the business terms that demanded it, with a fine of 1 million reais a day. It is a preliminary injunction, and Meta has five days to respond.

Nov 4, 2026
Sao Paulo Judge Orders Meta to Stop Scraping News for AI or Pay 1 Million Reais a Day
The grand columned hall of the Palace of Justice in Sao Paulo, Brazil

A Sao Paulo judge has told Meta it cannot make Brazilian news outlets hand over their work for AI training as the price of staying on Facebook and Instagram. On November 3, Judge Guilherme Madeira Dezem of the 44th Civil Court granted a preliminary injunction suspending the crawling clause in Meta's new business terms. The fine for ignoring it is 1 million reais a day, capped at 100 million.

This is a provisional order, not a final verdict. According to AFP, the judge ruled without waiting for Meta because the terms were due to take effect on November 20 and the damage would be irreversible once a model had been trained on the content. The suit came from three trade bodies, ANJ for newspapers, ABERT for radio and TV, and ANER for magazines. The order protects their members, not every outlet in Brazil. Meta has five days after notification to respond.

The reasoning is what other publishers will read closely. In the decision itself, the judge treats AI scraping as copying. Article 29 of Brazil's Copyright Law requires prior and express permission to reproduce a work or store it on a computer. Here is how he describes a training crawler, in our translation from the Portuguese.

"Automated scraping of news portals, broadcaster websites and editorial archives, through technical copies, fragmentation into computational units (tokens) and storage on servers for training large language models, is an undeniable act of reproduction. 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." Judge Guilherme Madeira Dezem, Sao Paulo state court

That last sentence draws the line between search and training. A crawler that indexes a page and sends readers to it is one thing. A crawler that digests the page into a model and sends nobody anywhere is, in this judge's view, another. He also dismissed robots.txt as a safeguard, calling it a technical courtesy with no binding legal force.

Then comes the leverage. Meta must keep member outlets' Facebook, Instagram, WhatsApp and Threads accounts running, with no punishment if they refuse the terms. It must keep bots off their sites and their content out of Meta AI training until final judgment. Quitting the platforms is no real option, the judge wrote, citing studies filed by the associations. He called the demand for free and universal surrender of copyrighted works as a toll to stay connected a manifest abuse of rights, in AFP's translation.

The fine came with a pointed explanation. The judge wrote that Meta has been especially reluctant to comply with decisions in other cases, choosing when to obey (our translation). He did not say which cases. Trade site PPC Land, which read an English translation of the order, notes the order bars automated access to members' sites without limiting that ban to AI, so read literally it could catch fetches that build link previews.

It told CNN Brasil it will not comment on the ruling. Its last public word came on October 28, when spokesperson Andy Stone said Meta would delay the updated terms if a consultation with publishers runs past November 20. The order does not mention that pause, according to PPC Land. A pause Meta controls is a promise. An injunction is a constraint.

ANJ chief Marcelo Rech called the injunction "a very positive sign that there are limits to the abuse of economic power," per AFP. The associations say it is the first such ruling for a coalition of media groups, and O Globo reports a Norwegian court suspended the same terms six days earlier. Britain's competition regulator opened its own probe on October 30.

What stands out is the route. Brazil did not need a dominance finding or a new AI law. A judge used copyright and contract rules already on the books and said tokenizing a news archive is copying it. Meta can still fight this, and the findings are only at the level of proof needed for urgent relief. But a company that wanted publishers to say yes by default now faces an order saying silence is not consent.