A technology company based in São Paulo trains a generative AI model on a corpus of Brazilian musical compositions. The model produces original-sounding scores. The company registers them, licenses them internationally, and begins collecting royalties. Twelve months later, the original composers' estates file infringement claims. The company's legal team discovers that its entire licensing structure rests on a question Brazilian courts have never fully resolved: who, if anyone, owns the output of an AI system under Brazilian intellectual property legislation?
Brazilian intellectual property legislation does not recognise AI systems as authors or rights holders. Ownership of AI-generated works must be claimed through human authorship doctrines, contractual assignment, or employer-employee rules – none of which map cleanly onto autonomous generative outputs. Brazil's evolving AI governance legislation is moving through its legislative process but has not yet resolved the core question of authorship. Leaving businesses that rely on AI-generated content exposed to ownership gaps, infringement risks, and technology licensing disputes.
This analysis examines the doctrinal foundations of Brazilian intellectual property law as they apply to AI-generated content, surveys the gap between statutory text and emerging practice. Addresses the cross-border implications for companies operating across the Americas. Additionally, offers strategic recommendations for businesses building AI-dependent content strategies in Brazil.
Doctrinal foundations: authorship, originality, and the human-creativity requirement
Brazilian intellectual property legislation is built on a classical civil law conception of authorship. Rights vest in the natural person who creates a work through the expression of intellectual effort and individual creative choice. This principle runs through the entire body of copyright law and has been consistently reinforced by courts at all levels, including the Superior Tribunal de Justiça (Superior Court of Justice of Brazil).
The originality threshold in Brazilian doctrine requires that a work bear the imprint of its author's personality. This is not a high bar for human creators. For AI outputs, however, it poses a structural problem. A generative AI system does not have a personality. It produces outputs through statistical inference over training data. There is no volitional creative act in the sense that Brazilian doctrine presupposes.
Three doctrinal positions have emerged in academic commentary and early administrative practice. The first treats AI-generated works as public domain from the moment of creation, on the basis that no eligible author exists. The second applies a work-for-hire analogy: the human or entity that commissions or operates the AI is treated as the functional author. In the same way an employer holds rights over work produced by an employee within the scope of employment. The third takes a tool-use approach: the AI is treated as a sophisticated instrument, and the human who directed its use is the author of the resulting work.
Each approach has genuine support in Brazilian legal commentary. Each also has meaningful weaknesses. The public domain approach denies protection to commercially valuable output and creates a free-rider dynamic that could disincentivise investment in AI development. The work-for-hire analogy requires a human operator whose creative contribution is identifiable – which is difficult to establish when a model operates autonomously on a vague prompt. The tool-use approach collapses under scrutiny when the human's contribution is minimal or purely technical.
Courts in Brazil have not yet produced a definitive ruling on which approach governs. Practitioners advising technology companies in Brazil note that the tool-use framing currently has the most practical traction in registration proceedings, but it is by no means settled.
The gap between statute and practice: registration, enforcement, and algorithmic accountability
The absence of a clear statutory rule creates a concrete problem at the registration stage. Brazil's national intellectual property authority processes copyright registrations without applying a standardised test for AI involvement. Applicants who disclose AI assistance face inconsistent treatment. Some applications have been accepted on the basis of human editorial curation. Others have been delayed or rejected where the AI's role appeared dominant.
This inconsistency matters commercially. A registration that is later challenged as invalid provides no protection against infringement. Companies that have built licensing programmes on unregistered or invalidly registered AI-generated content face the prospect of losing their entire revenue stream without notice.
The software liability dimension compounds the risk. Where an AI system produces content that infringes a third party's copyright – by reproducing protected material from its training data without authorisation – the question of who bears liability is equally unsettled. Brazilian civil liability doctrine applies a fault-based standard in most cases, but also recognises strict liability in specific contexts. Practitioners in Brazil note that the strict liability provisions applicable to defective products and digital services are being tested in early cases involving AI outputs. With courts applying algorithmic accountability reasoning to attribute responsibility to the deploying company rather than the developer.
For international businesses, this creates a technology licensing exposure that is easy to overlook. A company that licenses AI-generated content to a Brazilian distributor may find itself liable for infringement claims originating in the training data, even where the licence agreement contains standard warranties of title. Brazilian courts have shown a willingness to pierce through contractual chains to reach the party with effective control over the AI system.
The digital services dimension adds further complexity. Platforms that host or distribute AI-generated content may face secondary liability under Brazilian legislation governing digital services, particularly where they have knowledge of infringing material and fail to act promptly. The safe harbour provisions available to platforms under Brazilian internet and digital services legislation are narrower than their US equivalents and are subject to conditions that AI-content distributors frequently fail to meet.
To explore how Brazilian AI and technology law applies to your content strategy and platform operations, contact us at info@ferrazwhitmore.com.
Training data, input rights, and the infringement risk hidden in the model
One of the least-appreciated risks for companies deploying AI systems in Brazil concerns not the output but the input. The question of whether training an AI model on copyrighted works constitutes an act of reproduction requiring authorisation is analytically prior to any question about output ownership. Without a lawful training process, the entire output is potentially tainted.
Brazilian intellectual property legislation does not contain a broad exception for AI training. The exceptions that exist are narrowly drawn and tied to specific purposes: private copying, quotation for criticism, use for educational purposes, and certain research activities. None comfortably covers the systematic ingestion of protected works at scale to produce a commercial AI model.
Practitioners in Brazil note that rightsholders – including collecting societies representing musical works, publishers holding literary rights, and visual artists' organisations – have begun monitoring AI developers' use of their members' works. Early demand letters have been sent. Formal proceedings before Brazilian courts are anticipated. The exposure is not theoretical: it is a timed risk that increases as AI model deployment in Brazil expands.
The doctrinal argument available to AI developers rests primarily on a reading of the transformative use doctrine imported from comparative law analysis. Brazilian courts have not formally adopted the US fair use framework. However. Some academic commentary argues that Brazilian intellectual property legislation permits a form of transformative analysis as part of the three-step test applicable to copyright exceptions. This argument is unproven in Brazilian courts and carries significant litigation risk.
A more defensible approach involves licensing training data directly from rightsholders or their representative organisations. Several collecting societies operating in Brazil have begun developing AI-specific licensing products. Technology licensing agreements that include training data rights are still an emerging category, but they offer the clearest path to a defensible position under current Brazilian law.
For companies that have already deployed models trained on Brazilian-market content without clearing rights, the recommended step is a training data audit. This involves mapping the provenance of training inputs, identifying works that may be protected under Brazilian intellectual property legislation, and assessing the realistic exposure before litigation begins. Early engagement with rightsholders produces materially better outcomes than responding to formal proceedings.
Cross-border dimensions: Americas clients and the regulatory divergence risk
For companies operating across the Americas – particularly those with operations spanning Brazil, the United States, Mexico, and Colombia – the divergence between jurisdictions on AI-generated intellectual property creates a specific strategic risk. A content portfolio that is legally protected in one jurisdiction may be unprotected or infringing in another.
The United States has moved toward a clear rule: the Copyright Office will not register works produced by AI without human authorship. However. Will register works where a human author made creative contributions and the AI served as a tool. This approach is relatively well-documented through Copyright Office guidance and early court decisions. For a detailed comparison with the US position, see our analysis of AI-generated works and intellectual property in the United States.
Brazil is at an earlier stage. Its approach is less settled and more fragmented between registration practice, court decisions, and pending AI governance legislation. For companies that want to exploit AI-generated content commercially in both jurisdictions, this means that a single content strategy will not work. Separate legal analysis is required for each market.
The cross-border technology licensing dynamic introduces an additional layer. A licence granted in the United States over AI-generated content – assuming US protection applies – does not automatically confer protection in Brazil. The licensee in Brazil must independently assess whether the work qualifies for protection under Brazilian intellectual property legislation and whether the licensor had valid rights to grant in the first place. Many cross-border technology licensing agreements in the AI content space currently fail to address this gap explicitly, leaving both parties exposed.
AI Act compliance considerations are also relevant for companies with European operations that distribute AI-generated content into the Brazilian market. The EU's AI Act imposes obligations on high-risk and general-purpose AI systems that may include disclosure requirements relevant to copyright. A company managing algorithmic accountability obligations in Europe while simultaneously deploying AI-generated content in Brazil faces a dual compliance challenge that requires coordinated legal strategy across both systems.
Mexico presents a distinct variant of the same problem. Mexican intellectual property legislation, which shares civil law foundations with Brazil's, similarly centres authorship on human creative contribution. However, Mexican courts have not produced even the tentative administrative guidance visible in Brazil, making the position in Mexico even less certain. For companies with multi-jurisdiction Americas content strategies, Brazil – despite its current doctrinal uncertainty – is actually the most developed Latin American market on this question.
For a tailored strategy on AI-generated content protection and technology licensing across Americas jurisdictions, reach out to info@ferrazwhitmore.com.
Strategic recommendations and the approaching legislative inflection point
Brazil's AI governance legislation has been advancing through its legislative process for several years. The most recent drafts reflect a broad regulatory ambition: disclosure obligations for AI-generated content, algorithmic accountability requirements for developers and deployers, and provisions addressing AI's relationship to existing intellectual property rights. The legislation has not yet been enacted in final form, but its passage is widely anticipated within the near term.
When enacted, Brazil's AI governance rules are likely to change the terrain in at least three ways. First, mandatory disclosure of AI involvement in content creation will create a documentary record that affects registration proceedings and litigation strategy. Second, algorithmic accountability provisions may formalise the strict liability approach that some courts have been applying informally. Third, provisions addressing training data may either clarify the exception question or impose new licensing obligations – creating both risk and opportunity depending on a company's position in the value chain.
Companies that wait for final legislation before addressing their AI content exposure are likely to find themselves in a reactive position. The legislative process has been long enough that early movers who have built compliant structures will have a meaningful advantage. Those who have relied on ambiguity as a substitute for strategy will face compressed timelines to remediate.
Practitioners in Brazil advising technology companies on AI content matters recommend a phased approach. In the near term, the priority is a rights audit covering both training data provenance and existing output portfolios. This audit should assess which works might qualify for registration under current practice, which carry infringement risk from training data, and which are most commercially significant. The audit output frames the remediation priority.
In the medium term, technology licensing agreements – both incoming and outgoing – should be reviewed and updated to address AI-generated content explicitly. Standard intellectual property warranties in commercial agreements typically do not cover AI training data infringement. Updating these warranties, and obtaining appropriate representations from AI model providers, closes a gap that courts are beginning to exploit.
The intellectual property practice in Brazil at Ferraz & Whitmore supports clients through both the rights audit process and the structural remediation that follows. For companies at an earlier stage, a preliminary legal assessment of their AI content exposure in Brazil can be arranged within a short timeframe.
The longer-term strategic question is whether to engage proactively in the Brazilian legislative process. Collecting societies, platform operators, AI developers, and content-dependent businesses all have divergent interests in how the pending AI governance legislation is shaped. Companies with significant exposure should consider whether participation in consultation processes – through industry associations or direct engagement with relevant legislative bodies – is warranted. The legislative text, once enacted, will be difficult to reverse.
Frequently asked questions
Q: Can an AI system hold intellectual property rights over a work it generates in Brazil?
A: No. Under Brazilian intellectual property legislation, authorship rights vest exclusively in natural persons. An AI system has no legal personality and cannot hold or exercise intellectual property rights. Rights over AI-generated output must be claimed by a human or legal entity through contractual or statutory mechanisms.
Q: How long does it take to register an AI-assisted work with the relevant Brazilian authority, and what does it cost?
A: Registration timelines vary by category. Literary and artistic works registered with the designated national authority typically require several weeks to a few months for processing. Government fees are modest but vary by work type. Legal fees for preparing the supporting documentation – particularly where human creative contribution must be demonstrated – depend on the complexity of the AI-human interaction involved.
Q: Is it a misconception that training an AI on copyrighted Brazilian content is automatically lawful?
A: Yes, this is a common misconception. Brazilian intellectual property legislation does not contain an explicit broad exception for AI training data. Using protected works as training input without a licence or applicable statutory exception exposes companies to infringement claims. The position remains unsettled as Brazil's AI governance legislation moves through its legislative process, but relying on an assumption of lawfulness carries meaningful legal risk.
About Ferraz & Whitmore
Ferraz & Whitmore is an international law firm based in Lisbon, advising business clients across 46 jurisdictions. Our team combines Portuguese civil law expertise with English common law tradition to deliver cross-border legal solutions in AI and technology law. Intellectual property protection. Additionally, technology licensing. with particular depth across Iberian and Latin American markets. We advise technology companies, international investors, and in-house legal teams who need results-oriented counsel on AI-generated content strategy, algorithmic accountability, and digital services regulation in Brazil and across the Americas. As a law firm in Brazil-facing matters, our Americas practice draws on direct experience with Brazilian intellectual property proceedings, AI governance developments, and cross-border licensing structures. Engaging a lawyer in Brazil with cross-border experience is especially valuable when AI content strategies span multiple jurisdictions with divergent rules. The firm's AI and technology law practice covers AI Act compliance obligations in Europe alongside parallel advisory work in civil law jurisdictions including Brazil, Portugal, Spain, and Mexico. To discuss your AI content exposure in Brazil or across the Americas, contact us at info@ferrazwhitmore.com.
Disclaimer: This publication is provided for informational purposes only and does not constitute legal advice. The information herein should not be relied upon as a substitute for professional legal counsel tailored to your specific circumstances. Ferraz & Whitmore assumes no liability for actions taken or not taken based on the contents of this material. For advice regarding your particular situation, please contact info@ferrazwhitmore.com.