Brazil’s AI Bill (PL 2338/2023): Status and What Applies Today

Last reviewed: October 22, 2026. Bill 2338/2023 is the Brazil AI bill: the proposed legal framework for artificial intelligence. The Senate passed it on December 10, 2024, and it has been in the Chamber of Deputies since March 2025. It is not law, and the Chamber committee has not yet filed an opinion on it. This article sets out where the bill stands, why its timetable keeps slipping, what the Senate-approved text would require and which Brazilian rules already apply to AI products today.

Where the Brazil AI bill stands

  • Senate: approved on December 10, 2024 and sent to the Chamber of Deputies. According to the timeline published by the news outlet Antihype, it reached the Chamber on March 17, 2025.
  • Chamber committee: a special committee on artificial intelligence handles the bill. The Chamber’s records show the committee was created on April 4, 2025 and Deputy Aguinaldo Ribeiro (PP-PB) designated rapporteur on May 20, 2025.
  • Current status: when we checked the Chamber’s record on October 22, 2026, the bill was marked as awaiting the rapporteur’s opinion, with no opinion, substitute text or amendments recorded. Thirty-seven other bills are attached to it, the latest on September 2, 2026.
  • Executive’s bill: the Chamber’s records show that the Executive’s Bill 6237/2025, submitted on December 8, 2025, was attached to Bill 2338/2023 by the Mesa Diretora on March 19, 2026, under priority procedure, and received by the special committee on April 6, 2026.

Why the timetable keeps slipping

Every date given for a vote has passed without one. The sequence reported so far:

  • November 25, 2025: first date announced for the rapporteur’s report, with a vote in December. Neither happened.
  • December 2025: the vote moved to February 2026, and the Executive sent its governance bill (6237/2025).
  • March 25, 2026: the rapporteur promised a committee vote by early May.
  • May 19 and 27, 2026: the dates then expected for the committee and the floor passed without a vote.
  • August 24, 2026: the rapporteur told Mobile Time that work would resume at full force only after the elections (first round on October 4, runoff on October 25), with a vote expected at the end of the year.

Two further points matter for planning. The rapporteur intends to change the text the Senate approved, according to Antihype. And because the bill returns to the Senate after the Chamber votes, Mobile Time reports that he is discussing copyright and other open points with Senator Eduardo Gomes, whom it identifies as the Senate rapporteur. A Chamber vote at the end of 2026 would therefore not by itself make the bill law. A company that is waiting for final text before acting is likely to wait for a long time.

What the Senate-approved text would do

The points below come from press reports of the text the Senate passed in December 2024. The Chamber may change any of them, so check the current text before relying on it.

  • Prohibited uses: reported examples include real-time remote biometric identification in publicly accessible spaces (with exceptions such as searching for missing persons and enforcing court orders), systems that assess personality traits or criminal behavior to predict crime or reoffending, public-sector systems that rank people in an illegitimate or disproportionate way for access to public policies, and systems that produce or facilitate child sexual abuse material.
  • High-risk uses: press reports list, among others, systems for traffic control, water and energy supply networks, access to education and student monitoring, worker recruitment and management, prioritizing essential public services, criminal investigations by the justice system, health diagnostics and autonomous vehicles in public areas. A provision that would have classified platform content recommendation as high risk was removed in the Senate, under an agreement among party caucuses.
  • Rights of affected people: a right to information about interactions with AI. For high-risk systems, a right to an explanation, to contest decisions and to request human review. Human oversight of high-risk systems would also be required, except where it is proven impossible or disproportionately burdensome.
  • Penalties: administrative sanctions including warnings, fines, suspension and a prohibition on data processing. The fine would reach up to R$50 million per infraction or up to 2% of the group’s gross revenue. Liability for damages follows the Civil Code.
  • Authority: a national AI regulation and governance system (SIA), coordinated by the ANPD, the data protection authority.
  • Copyright: developers and distributors would have to remunerate holders of protected content used to develop AI, with the amounts negotiable between the parties and an exception for non-commercial research. This is one of the points still being negotiated.

What already applies to AI products in Brazil

An AI product does not need the bill to be regulated in Brazil. The data protection law applies now, and the regulator is already preparing to apply it to AI. For how it reaches companies abroad, see Brazil’s LGPD for foreign companies.

  • Reach (LGPD, article 3): the law applies to processing aimed at people located in Brazil, regardless of where the company is headquartered.
  • Automated decisions (article 20): a person can ask for review of decisions made solely by automated processing that affect their interests. The controller must give clear information on the criteria used, subject to commercial and industrial secrecy, and the ANPD may audit for discriminatory aspects.
  • Liability (article 42): a controller or processor that causes damage by violating the law must repair it. The Civil Code’s general rule that whoever causes damage by an unlawful act must repair it (article 927) also applies.
  • Sanctions (article 52): fines of up to 2% of the company’s, group’s or conglomerate’s revenue in Brazil in its last fiscal year, limited to R$50 million per infraction.
  • Regulator: the ANPD’s first AI regulatory sandbox, which tests transparency and explainability of automated decisions under article 20, is in its final testing phase with three companies in medical diagnosis, transport and financial-sector security. The ANPD expects to publish the full results by the end of October 2026. It has also announced a second sandbox, on children’s digital protection (ECA Digital), for launch by the end of 2026, and is building an internal AI center.

What this means in practice

These examples illustrate how the rules fit together. They are not advice on any real product.

  • A recruiting tool sold to Brazilian employers. The LGPD already applies, including a candidate’s right to ask for review of an automated decision. Under the Senate text, worker recruitment and management appear among high-risk uses, so explanation, contestation and oversight duties could follow if that text survives.
  • A generative AI product trained on creative works. The copyright remuneration rule is still under negotiation. Contracts with rights holders and with customers should not assume either outcome.
  • A foreign vendor whose only Brazilian link is an API used by local companies. The LGPD may still apply, depending on the data and on the vendor’s role. Contracts should say who answers data subject requests and who bears the cost of a violation.

Questions to ask while the bill is pending

  1. Does our AI system process personal data of people in Brazil, and on what legal basis?
  2. Do we make decisions solely through automation, and could we explain the criteria on request?
  3. Do our vendor and customer contracts allocate the cost of a data protection violation?
  4. Does any of our use cases match a prohibited or high-risk use in the Senate text?
  5. Who tracks each new version of the bill’s text, and who decides when we adapt?
  6. Have we read the ANPD’s sandbox results once they are published?

Related guide: Tech & data in Brazil.

Further reading in Portuguese

Our Portuguese articles PL 2338/2023: o marco legal da IA no Brasil and Marco legal da IA no limbo follow the bill for Brazilian readers.


This article is for informational purposes only. It does not constitute legal, tax or investment advice, does not create an attorney-client relationship, and is not a solicitation of clients. See our Disclaimer.

Sources


About the author

Luiz Guedes is a Brazilian attorney (OAB/PB 11.005) based in João Pessoa, with more than 25 years of experience in corporate, tax and labor law and a doctorate in Economic Law from the Federal University of Paraíba (UFPB). He writes LawGeek International to explain Brazilian law and regulation to foreign investors, founders and builders in tech and crypto.

Academic profile (Lattes) · About LawGeek · Contact

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top