The recently approved Senate Bill 2,338/23, which regulates the use of Artificial Intelligence (AI) in Brazil, faced criticism from legal experts, academics, and corporate representatives during the debate “Decoding Copyright in the Age of Artificial Intelligence.” As the inaugural event in ABPI’s AI & IP Forum series, it took place on December 11, filling a conference room at the INNSiDE São Paulo Iguatemi Hotel in Itaim Bibi.
The bill, approved on December 10, is a substitute proposal by Senator Eduardo Gomes (PL-TO), based on a draft bill introduced by Senate President Rodrigo Pacheco and developed by a committee of legal experts. The legislation now heads to the Chamber of Deputies for review. Among its provisions, the bill includes measures to protect the rights of creators of content and artistic works.
Expert Opinions on the Bill
Keynote speaker Max Sills, General Counsel at MidJourney, joined the event remotely under the moderation of ABPI’s Second Vice President, Tatiana Campello. Sills criticized the Brazilian bill for discouraging data training, a process critical to enabling AI models to recognize patterns. He argued for an approach aligned with international jurisdictions:
“We want AI that maximizes creativity while respecting rights holders, harmonized with international laws.”
In the first panel, “Regulatory Approaches to AI: How to Regulate a Constantly Evolving Technology,” moderated by ABPI Study Committee Coordinator Fernanda Magalhães, Celina Bottino, Director at ITS Rio, emphasized the need for broader discussions on the bill. Bottino argued that the legislation, inspired by European AI regulations, does not adequately reflect Brazil’s realities. Drawing on her studies of regulations in countries like Japan, the United States, and the United Kingdom, she highlighted differing approaches worldwide.
Eduardo Paranhos, leader of ABES’s (Brazilian Software Companies Association) AI Working Group, criticized adopting European legislation as a model for Brazil. He advocated for Japan’s approach, which integrates AI regulation into existing legal frameworks. According to him, the bill’s provisions on AI model training make it “the most restrictive legislation in the world.”
Microsoft Director Christian Perrone pointed out that AI will impact all industries and businesses:
“This legislation will lay the foundation for future laws.”
Legal Challenges in AI-Generated Works
During the panel “Legal Safeguards: Authorship and Ownership of Inputs and Outputs in Generative AI,” moderated by Tatiana Campello, Bruegel researcher Bertin Martens noted that the copyright industry accounts for 4% of global economic activity. He warned that restricting access to training data could negatively affect the remaining 96% of the economy. Martens advocated for eliminating copyright from AI-generated content to benefit the economy:
“If we limit AI training, media content production will decline.”
Federal Judge Caroline Tauk provided an insightful overview of AI, raising questions about the authorship of AI-generated creations, particularly in artistic production. She explained that recognizing a human as the author would depend on their creative contribution to the final work:
“This is the crux of the matter, and we still don’t have an answer. Currently, International Law does not recognize machines as authors.”
Professor Luca Schirru of PUC-RJ addressed the use of training data (inputs) in AI. He criticized the bill for restricting research access and noted that top-ranking innovative countries are more permissive, allowing data mining for research purposes:
“The approved text is highly problematic. Brazil now has one of the most restrictive laws globally regarding research.”
Balancing Copyright and Technological Development
In the panel on “Copyright and AI Training – How to Balance Protection and Technological Development,” moderated by José Eduardo Pieri, co-coordinator of ABPI’s Digital Law & Data Privacy Study Committee, Allan Rocha de Souza, president of IBD Autoral, stated that Bill 2.338/23, by establishing a remuneration obligation (Article 64) limited to copyright and related rights holders, overlooks the distinction between rights holders and authors or artists and has the potential to exacerbate the imbalance of power.
Pamela Samuelson, Professor at the University of California and Co-Director of the Berkeley Center for Law & Technology, provided a detailed overview of U.S. lawsuits involving training data in generative AI. Samuelson explained that generative AI is not the first disruptive technology to face copyright litigation in the U.S. She predicted mixed outcomes in court but noted that many “fair use” defenses for training models are plausible and successful:
“Brazilian policymakers might find a better way to address training data issues, avoiding years of costly litigation like in the U.S.”
AI Legislation and Business Decision Impacts
The final panel, “Bill 2,338 and Its Expected Business Impacts,” moderated by ABPI Council Member Filipe Fonteles Cabral, provided a pragmatic perspective on AI use in businesses.
Andriei Gutierrez, Director of Kyndryl Latin America, criticized the bill’s lack of distinction between computational models and AI systems:
“Governance should focus on the system. Regulating the model makes no sense as it restricts technology and development.”
He emphasized the need to mature the debate to provide legal certainty and foster economic growth.
Samanta Santos de Oliveira, Head of Mercado Livre’s Data Protection Committee, discussed Articles 6 and 8 of the bill, which address human oversight in AI systems. She differentiated between human review and intervention, cautioning that monitoring every stage of the system could become unfeasible. In Mercado Livre’s ecosystem, she explained, 135,361 new listings are created every hour in Brazil, with over two million related to Terms & Conditions violations.
Caroline Rocabado, Legal Coordinator at Dasa, highlighted the dispersed and inflated nature of Brazil’s current healthcare model. Only through technology, she argued, can advancements in product research and development be achieved, enabling faster and more accurate decision-making to benefit the population:
“Using algorithms in healthcare not only enhances diagnostics but also improves care, reduces errors, and minimizes waste through data analysis.”
The organizing committee of the ABPI’s AI & IP Forum series is formed by Tatiana Campello, Erika Silla, Fernanda Magalhães, Filipe Fonteles Cabral and José Eduardo Pieri.
