Brazil is trying to write AI copyright rules before the rest of the region does, and the fight over how to do it is now centred on money. PL 2338/2023, approved by the Senate in December 2024 and currently before a special committee in the Câmara dos Deputados, would require AI companies to pay rights holders whenever copyrighted material is used to train commercial systems. Creators would get an explicit right to opt their work out entirely, developers would have to disclose summaries of their training datasets, and violations could draw fines of up to R$50 million, roughly $8.9 million.
The bill has split Brazil's tech and cultural sectors cleanly down the middle. Musician Marisa Monte and a coalition of cultural organisations back the royalty requirement as basic protection for Brazilian creators whose work is already being scraped into foundation models with no compensation. The Brazilian Software Association and agribusiness groups are pushing back hard, arguing Brazil should instead adopt the EU's opt-out model, which does not require payment by default.
The industry's core argument is about capital flight risk: trade groups warn the copyright-fee regime could redirect AI development offshore, jeopardising roughly R$23 billion, about $4.1 billion, in government investment already committed to computing infrastructure and Portuguese-language foundation models under the Brazilian AI Plan and the Redata data-centre tax-incentive programme. Free use would remain available for non-commercial research, journalism, and library or museum use, but for commercial AI developers building in Brazil, the calculus now includes a licensing bill that does not yet have a final price tag.
