São Paulo court bars Meta from scraping press sites to train AI

An injunction suspends a clause that traded access to Meta's platforms for site scraping. It covers only members of three associations, but the reasoning matters to every company.

On Tuesday (the 3rd), the 44th Civil Court of São Paulo suspended the part of Meta’s new commercial terms that tied the use of Instagram, Facebook, WhatsApp and Threads to permission for automated collection of websites, known as scraping. The injunction, requested by Abert, ANJ and Aner, the Brazilian associations of radio and TV broadcasters, newspapers and magazines, forbids Meta from scraping its members’ sites and from using their protected content in Meta AI. It also orders that their accounts stay active, without penalty. The fine is R$ 1 million (Brazilian reais) per day, capped at R$ 100 million. Conjur and Gazeta do Povo detailed the case.

The scope of the ruling

It is a provisional, first-instance decision, issued without hearing Meta, which said it will not comment. It holds until the final judgment and was written for the members of the three associations, not for companies in other sectors. On Monday, we reported here that these terms are also under investigation in the United Kingdom and that Meta announced a pause. The Brazilian ruling mentions only the postponement to November 20.

What goes beyond the press is the reasoning of judge Guilherme Madeira Dezem. In his view, collecting protected works to train AI is an act of reproduction, and being available on the internet doesn’t put them in the public domain. He also states that Brazilian law has no general exception for this kind of commercial training, and he found the California jurisdiction clause in the terms ineffective in Brazil.

What this changes for your company

1. Separate what’s yours from what belongs to others. According to the ruling, the clause makes the company declare that it holds the rights needed to license the content. A stock photo, a partner’s article and a supplier’s material may not be yours to license. Map this before accepting terms like these.

2. Don’t treat robots.txt as legal protection. The judge accepted the argument that this file, which guides search robots, is a technical directive with no binding force. Whatever matters needs access control and a rule in the contract.

3. Look at the other side too. If your company trains or feeds an AI with content collected from the web, the same argument can be used against you. Ask your legal team to check where that data comes from.

With 29 years of programming, I would start with a simple spreadsheet: where each piece of data that goes into your AIs comes from and who authorized that use.


Originally published on IWA7 Insights, IWA7 Digital’s blog, which is now part of Escovando Bytes.

Keep readingSee all →