Case Study: Brazilian Civil Society’s Long Fight for Human Rights in AI Governance

In December 2024, the Federal Senate in Brazil approved AI-related legislation, Bill No. 2338/2023. This comprehensive lawsets out principles, rules, and mechanisms for regulating AI in a way that is aligned with fundamental rights and democratic values. It adopts a risk-based approach to AI governance, similar to the EU AI Act, establishing three tiers of AI systems.¹⁹³ Systems that pose excessive risk shall be prohibited; high risk shall be heavily regulated; and other systems must meet basic requirements of regulation. The legislation also mandates assessments of algorithmic impact for high-risk systems and human oversight and strict requirements for AI use in the public sector. It further establishes several rights for individuals interacting with AI systems, including the right to explanation and the right to contest decisions that significantly impact them.

This law was first proposed in 2023 but had to be reviewed and voted on by the House of Representatives and adopted by the President. It is still unclear when it will be adopted and how the bill will look before it is adopted. However, the law is considered a win for civil society in Brazil which has been advocating for human rights-based regulation of AI.¹⁹⁴

In fact, another bill that was first proposed to regulate AI in the country, draft Bill no.21/2020 adopted “a very light touch and minimalistic approach” with “several substantial gaps, being either laconic or vague as regards some key elements such as risk assessment and oversight mechanisms”.¹⁹⁵ This bill “ was really favoured by the industry,” said Yamin Curzi , a professor at the Centre for Technology and Society at Fundação Getulio Vargas (FGV) Law School in Rio de Janeiro. “But of course, civil society pushed a lot against this, because it would not give the process of accountability and the due process that we needed to have, actually, governance of the AI systems. We would have only the industry regulating itself. So then, civil society pushed against this. We had a lot of push by civil society organizations back then and a lot of debates, also in academia”. As a result of the civil society pressure , including Coalizão Direitos na Rede (CDR), a coalition of 51 academic and civil society organisations in defense of digital rights,¹⁹⁶ Brazil’s Senate launched multistakeholder public hearings and consultations for a substitute bill.

As part of its efforts, CDR, which has conducted campaigning and advocacy on AI regulation in the country, published petitions,¹⁹⁷ sent letters to lawmakers,¹⁹⁸ submitted input for an alternative bill,¹⁹⁹ and engaged directly with members of parliament. ²⁰⁰ CDR has also conducted legal analysis and campaigned for a ban on the use of facial recognition technology in public spaces.²⁰¹

While civil society has been successful in pushing back against the initial AI draft bill and shaping a more robust alternative, a long road still lies ahead, as the tech industry and lawmakers can still undermine human rights-based regulation of AI. Already the version of Bill No. 2338/2023 that was approved by the Senate in December 2024 had been re-shaped. For example, under pressure from technology companies, algorithmic systems to recommend content were not listed under the high-risk category and most references to ‘information integrity’ or ‘freedom of expression’ were removed after Big Tech companies and members of the opposition equated such references with ‘censorship’.²⁰² Pressure came not only from Big Tech companies, as Laís Martins, a Brazilian investigative journalist based in São Paulo, wrote for Policy Tech Press: “Advocates for a weaker regulatory approach (or no regulation at all) were joined by Brazilian industry groups and data center executives, all echoing a big tech narrative that regulation would hinder innovation. To some extent, their efforts were successful in weakening what started out as a very ambitious bill”.

“Both bills are still alive because people are trying to push for AI regulation also with this principled logical approach still nowadays…Bill 2338/2023 is being discussed. It was approved in the National Chamber, but because when the Senate approves a bill, it goes back to the National Deputy Chamber for revisions, and then again goes to the Senate, to be sent out to the federal government for final ratification, before it is signed to be enforced. They try to retrieve some of the due process rules because of the lobby of big tech, especially by Microsoft, but also, AI enterprises and associations of industries, especially linked to the agribusiness here in Brazil because we have tons of AI deployments in agribusiness”, asserted Professor Curzi.

While civil society continues the fight,²⁰³ they face obstacles in their advocacy efforts: “most of these [government] agencies are based in Brasilia, the capital of Brazil. Civil society is not there. Civil society is in Rio, in Sao Paulo, in the Northeast but not in Brasilia. It is really expensive to have a lawyer. It is really expensive for them to do advocacy all the time. And for the enterprises, it is not. They go all the time there, they have this capacity of being in private spaces with the regulators..and they established links to and relationships with these people [regulators and lawmakers]”, Curzi declared.

In the meantime, Brazilians belonging to excluded groups, including Black and Indigenous people, continue to face discrimination in AI systems. Facial recognition cameras are widely deployed in public and private spaces with little protection, transparency, and oversight, thus posing risks to privacy, freedom of expression, and freedom of association. Brazil’s civic space was rated as obstructed by the CIVICUS Monitor in 2025, a rating it has retained since 2018.