Brazil AI liability civil courts

Brazil's Two-Track AI Liability Regime Sets Up a Collision Between the Civil Code and the AI Framework Bill

PL 4/2025's Book VI and PL 2338/2023 would govern the same algorithmic harm under different rules — and the rapporteurs' May–June reports will decide how messy that gets.

Brazil's AI liability calendar — and what overlaps i… People of Internet Research · Brazil May 6, 2026 Sectoral report deadline Sen. Portinho's partial report on … Jun 3, 2026 General report deadline Sen. Veneziano's consolidated PL 4… Dec 10, 2024 PL 2338 Senate approval AI Framework bill, pending Chamber… R$50M Max sanction under PL 2338 Per-violation cap in the risk-tier… peopleofinternet.com

Key Takeaways

Brasília is about to lock in two parallel answers to the same legal question — who pays when an algorithm causes harm — and the second answer is being drafted before the first has even passed the Chamber of Deputies. On May 6, 2026, Senator Carlos Portinho (PL-RJ) is scheduled to deliver the sectoral partial report on civil liability inside the Senate's Temporary Commission for the Civil Code reform (CTCivil); on June 3, 2026, general rapporteur Veneziano Vital do Rêgo (MDB-PB) is due to consolidate that text into the full opinion on PL 4/2025, with the commission targeting closure by the end of June. Buried in Book VI ("Digital Civil Law") of that 900-plus-article reform is a new civil-liability regime for artificial intelligence — a regime that overlaps with, but does not align to, the AI-specific PL 2338/2023 that the Senate already approved on December 10, 2024 and that is still awaiting a vote in the Chamber.

What Book VI actually does

The most consequential provision in Book VI — circulating as Article 2.027-AL in successive working drafts — establishes that every algorithmic operation must be legally anchored to a natural or legal person, and that liability for AI-caused harm is governed by the Civil Code's principle of full reparation (reparação integral). In plain terms: there is no "the model did it" defence. A developer, deployer, or operator is always on the hook, and the measure of damages is whatever it takes to put the victim back in the position they would have occupied absent the harm — material loss, moral injury, and consequential damages included.

Layered on top are duties of transparency, non-discrimination, traceability, and human oversight, folded into the existing civil-law concepts of good faith and the social function of contracts. The drafters' explicit theory is that AI does not need a sui generis liability statute; it needs to be readable by the Civil Code that Brazilian first-instance judges already use every day.

The steelman for doing this in the Civil Code

There is a serious case for the rapporteurs' approach. Brazilian civil courts are going to adjudicate AI harms whether or not Congress finishes PL 2338/2023 — the Chamber has now sat on that bill for over a year. If the Civil Code stays silent, judges will improvise: some applying the Consumer Defense Code's strict-liability regime, others the general fault-based articles 186 and 927, and others the ANPD's data-protection rules by analogy. That fragmentation is exactly the kind of jurisprudential noise that scares investors more than any single rule. Anchoring every algorithmic act to an identifiable person also closes the "agentic AI" loophole that European and US commentators are now raising — the worry that fully autonomous systems will be used to launder responsibility through corporate shells.

Why the parallel track is still a problem

The difficulty is that PL 2338/2023 is not silent on liability either. The Senate-approved text carries its own liability provisions attached to a risk-tiered framework (excessive, high, non-high) with sanctions of up to R$50 million per violation, coordinated by the ANPD through the proposed National AI Regulation and Governance System (SIA). Legal analysts of Book VI have already identified three concrete risks from running both regimes side by side: normative duplication (two statutes prescribing different duties for the same conduct), interpretive conflicts (a high-risk classification under PL 2338 that does not map cleanly to the Civil Code's good-faith standard), and legal uncertainty about which regime a plaintiff can invoke first.

This matters because civil courts move faster than regulators. If PL 4/2025 passes first — which the June timetable makes plausible — Brazilian plaintiffs will sue under the Civil Code long before the ANPD has issued a single implementing rule under PL 2338. The de facto AI liability regime in Brazil will then be whatever the tribunais de justiça of São Paulo, Rio, and Minas Gerais say it is, applying a full-reparation standard with no risk-tier filter and no safe harbour for compliant operators. Sub-rapporteur Portinho himself flagged the underlying anxiety at the March 19 CTCivil hearing, warning that vague new liability concepts risk "banalisation" and increased judicialisation.

What a proportionate fix looks like

The rapporteurs do not need to abandon Book VI. They need to coordinate it. Three edits, all achievable inside Senator Portinho's May 6 sectoral report, would meaningfully reduce the collision risk without weakening victim protection:

None of this is deregulation. It is the difference between a country whose courts can confidently adjudicate AI harm and a country whose first AI ruling becomes a warning to every founder choosing where to incorporate. The CTCivil has six weeks to choose which one Brazil becomes.

Sources & Citations

  1. Senado Federal — PL 2338/2023 tracking page (Marco Legal da IA)
  2. Congresso Nacional — PL 4/2025 bicameral tracking page
  3. Senado Notícias — CTCivil debate on civil liability (Mar 19, 2026)
  4. RNA Advogados — Analysis of Article 2.027-AL and Book VI Digital Civil Law
  5. Data Privacy Brasil — Technical analysis of PL 2338/2023 as approved by Senate