Is Brazil Getting Tougher on Arbitral Awards? Evidence from the Court of Justice of São Paulo

Arbitration is the resolution mechanism of choice for high-value commercial disputes in Brazil, and São Paulo state is its centre. Its court, the Court of Justice of São Paulo (TJSP), is the largest in Brazil and hears more arbitration-related cases than any other in the country. It is also highly specialised, with first-instance courts dedicated to business and arbitration disputes and two chambers reserved for business law-related appeals.

A few years ago, the Brazilian legal press suggested that the relationship between the TJSP and arbitration was souring. Losing parties appeared to be challenging arbitral awards more frequently, and the courts appeared increasingly willing to set them aside. Nonetheless, relevant voices in Brazilian arbitration described the debate as ‘a lot of smoke and little fire’. Our empirical research set out to test what the TJSP’s decisions actually show, first for 2011–2022 and then, in an update, up to August 2025.

As in many other jurisdictions, judicial review is not an appeal on the merits. Brazilian law limits judicial review of arbitral awards, protecting the finality bargained for by commercial parties. An award may only be annulled on the grounds set out in Article 32 of the Brazilian Arbitration Act (Lei nº 9.307/1996), either through an independent set-aside action brought within 90 days of notification of the award or as a defence in enforcement proceedings. The grounds include an invalid arbitration agreement, an award rendered by someone who could not act as arbitrator, an award that exceeds the scope of the arbitration agreement, missing formal requirements, and a breach of due process. An error in applying substantive law is not a ground for annulment.

To examine how the TJSP applies these standards in practice, we collected every appellate decision publicly available in the Court’s case-law database whose summary addressed the annulment of an arbitral award. We did so using an automated Python scraper, retrieving appellate judgments issued between February 2011 and August 2022. We then manually examined each decision and coded it through a structured protocol, recording the outcome (annulled in full, in part, or upheld) and the legal basis. In 2025, we extended the dataset to August 2025, coding the new judgments with a large language model.

Across the whole period, from 2011 to August 2025, the data temper the narrative of judicial hostility but reveal significant judicial intervention. Of the 200 appellate decisions on annulment, most (71.5%) upheld the arbitral award intact. Yet 28.5% annulled the award in full or in part, a frequency that challenges the conventional description of annulment in Brazilian scholarship as an ‘extremely exceptional’ remedy. Until 2022, partial annulments were more common than total ones (16.9% against 10.1%). Since then, total annulments have predominated. In partial annulments, the TJSP struck out specific components, such as a set-off or an award of default interest, while preserving the rest of the tribunal’s decision, in line with the principle that the valid part of an act is not tainted by the invalid part (utile per inutile non vitiatur).

Nonetheless, the data confirmed that practitioners’ perceptions of a shift were rooted in genuine movement. Appellate challenges began climbing in 2018, and the number of awards set aside grew sharply from 2019. The three years from 2019 to 2021 accounted for 34 annulments, 85.0% of all annulments between 2011 and 2022. That surge, however, came from a single source. Twenty-eight of the 40 annulments (70.0%) arose from disputes between medical cooperatives within a single cooperative healthcare network, which has its own arbitral chamber for commercial disputes among its members. Outside that network, the TJSP annulled awards in 11.8% of its decisions on annulment claims (12 of 102), and the 2019–2021 peak largely disappears. 

The picture since 2022 is different. Between September 2022 and August 2025, the TJSP decided 52 annulment claims and set aside 17 awards in full or in part (32.7%), a similar overall rate to before. Only five of those annulments involved the cooperative network. Excluding decisions concerning the cooperative network, the share of decisions annulling an award more than doubled, from 11.6% to 27.3% (12 of 44). Most recent annulments were total, and more than half rested on breaches of due process. The cooperative wave has passed, but judicial intervention appears to have spread into the wider commercial caseload.

The findings hold three lessons for commercial parties arbitrating in Brazil. First, headline numbers on annulments of arbitral awards can mislead in both directions. A cluster of cases in one industry inflated the 2019–2021 surge, and the end of that cluster has since masked a rise in annulments in ordinary commercial cases. Second, due process has become the main point of failure, and tribunals seated in São Paulo should treat equal treatment and the right to be heard as the most likely grounds of judicial review. Third, partial annulment preserves the enforceable core of awards, but it leaves the final word on the severed issue with the court. Article 33(2) of the Arbitration Act already allows courts to order the tribunal to render a new award, and it is worth asking whether that route should be used more often.

The findings also hold one lesson for researchers and legal scholars. In our latest study, we used a large language model (Gemini 2.5 Pro) to re-code our original judgments. It matched our manual classification in 96.2% of the coded variables (Cohen’s kappa of 0.92). It did so in under 90 minutes and for about USD 5, and it reproduced our headline result, placing the annulment rate outside the cooperative network at 11.6% against our 11.8%. Generative AI made the 2025 update feasible. Periodic empirical monitoring, now cheap enough to run routinely with automation and large language models, is a new way to detect a genuine shift in judicial attitudes and distinguish it from a cluster of casesin one sector.

So, is Brazil getting tougher on arbitral awards? Until 2022, it was not. Since then, outside the healthcare cooperatives, the early evidence suggests that it may be.

The authors’ complete articles (in Portuguese) can be accessed here and here.

Camilo Zufelato is a Professor of Civil Procedure at the Ribeirão Preto Law School, University of São Paulo.

Victor Dantas de Maio Martinez is a lawyer in Brazil and a PhD candidate in Civil Procedure at the Ribeirão Preto Law School, University of São Paulo.

Fernando Barroso is a lawyer in Brazil and holds an MSc in Law and Finance from the University of Oxford.