Brazil Court Blocks Meta From Scraping News for AI Training Without Permission
A São Paulo judge temporarily banned Meta from forcing media outlets to hand over their content for AI training to keep their accounts.
Policy · Source: AFP via Pakistan TV
What happened
A São Paulo state court ordered Meta to stop scraping content from Brazilian media outlets to train its AI models. Three national press associations filed a civil suit against the tech giant over its new business terms. The associations involved are the National Association of Newspapers, the Brazilian Association of Radio and Television Broadcasters, and the National Association of Magazine Editors. The controversial rules were set to take effect on November 20. They would have allowed Meta to use automated programs like web scrapers and spiders to copy articles, photographs, and videos directly from publishers' websites. This included content published entirely outside of Meta platforms.
Meta made this a strict take-it-or-leave-it deal. If publishers wanted to keep their accounts active on Facebook, Instagram, WhatsApp, and Threads, they had to give Meta free access to their copyrighted work. Judge Guilherme Madeira Dezem called this a manifest abuse of rights. He stated that demanding the free surrender of copyrighted works as a toll to reach the public is unacceptable. He issued a preliminary injunction immediately, noting that the damage to media outlets would be irreversible once the AI models were trained on their data.
The court explicitly banned Meta from penalizing or restricting media outlets that reject the new terms. If Meta breaks the rule, it faces a daily fine of one million reais, which is roughly 180,000 dollars. The total penalty is capped at 100 million reais. Meta now has five days to present its defense. This Brazilian ruling follows a similar move in Norway, where a court temporarily suspended the same rules on October 28 after ten media groups took legal action.
Key facts
- November 20 — When Meta's new business terms were set to take effect
- 1 million reais — The daily fine Meta faces for non-compliance
- 100 million reais — The maximum cap for the fines
- 5 days — Time Meta has to present its defense
- October 28 — When a Norwegian court similarly suspended Meta's rules
Why it matters
The era of scraping everything by default is hitting a massive legal wall. If you build AI models, you can no longer assume public data is free data for you to take. Courts across different countries are starting to view mandatory data surrender as a clear abuse of economic power. This ruling sets a major global precedent for coalitions of content creators to fight back against forced scraping terms from big tech platforms. It proves that publishers do not have to choose between keeping their social media distribution and protecting their intellectual property.
Big tech companies will soon have to negotiate actual licensing deals instead of just updating their terms of service. This creates immediate friction for model training pipelines that rely on massive, free data ingestion. However, it also opens up a lucrative market for verified, opt-in data marketplaces. Startups that figure out how to properly compensate creators while feeding hungry AI models will have a massive advantage. Those relying on brute-force scraping will face constant lawsuits and operational blocks.
For builders
Scraping as a service gets risky
Courts are actively blocking automated web scrapers, spiders, and crawlers from taking copyrighted media. If your product relies on scraping news sites without explicit permission, you carry massive legal liability. You lose big when publishers unionize and take you to court.
Paywalls and opt-in data win
Media coalitions are successfully defending their intellectual property against tech giants. Builders who create infrastructure for publishers to license their data securely will get paid. AI companies will have no choice but to pay for high-quality, legally cleared training data.
Terms of service are not shields
You cannot force users to surrender their copyright just to use your platform. Judges are striking down these clauses as abusive and anti-competitive. Founders who try to sneak AI training rights into user agreements will face severe regulatory backlash.
My take
I have always said that building AI on stolen data is a fragile foundation. Meta tried to bully publishers by holding their social media distribution hostage, and the Brazilian court rightly shut it down. If we want to build sustainable AI products, we have to pay the people who actually create the data. Forced consent is not consent.
Original reporting: AFP via Pakistan TV. This is my rewrite and opinion.