A technology company builds a generative AI system that produces music, images, and written content at industrial scale. Its business model depends on owning – or at minimum controlling – the intellectual property in that output. Then Portuguese counsel delivers an unwelcome assessment: the existing body of Portuguese copyright law was not designed with autonomous machines in mind, and the question of who owns AI-generated works remains doctrinally unresolved. The commercial opportunity is real; the legal foundation underneath it is not.
AI-generated works occupy a contested space under Portuguese intellectual property legislation, which conditions copyright protection on a human author whose creative choices are reflected in the work. Where a generative system produces output with minimal human direction, that condition may not be satisfied, leaving the output unprotected or owned by a party the developer did not intend. Portuguese courts have not yet resolved the central doctrinal questions, and the EU AI Act compliance regime adds a further regulatory layer that intersects with, but does not replace, copyright doctrine. Businesses that rely on AI-generated content must audit their IP position before their competitive advantage evaporates in a dispute they could not have anticipated.
This analysis covers the doctrinal foundations of copyright law as applied to AI output in Portugal, the competing interpretive approaches emerging in European legal practice, the gap between the letter of the statute and actual enforcement conditions. Cross-border implications for EU and international clients, strategic recommendations for securing enforceable IP rights. Additionally, the regulatory outlook through the lens of the EU AI Act.
Doctrinal foundations: what Portuguese intellectual property legislation requires
Portuguese intellectual property legislation is built on a continental civil law model. Copyright protection attaches to original works that are the expression of the author's intellectual creation. The author, in the traditional formulation, is a natural person. This requirement is not an accident of drafting. It reflects a philosophical commitment, shared across most civil law systems, to the idea that creative works carry a trace of their human originator's personality.
The practical consequence is significant. An AI system does not have legal personality under Portuguese law. It cannot hold rights, assume obligations, or be identified as an author within the meaning of intellectual property legislation. This position is consistent with the approach taken under Portuguese civil law more broadly, which reserves legal personality for natural persons and legal entities constituted under corporate legislation. A limited liability company incorporated under Portuguese corporate legislation – the Código das Sociedades Comerciais (CSC) – can own intellectual property by assignment or transfer, but it cannot originate authorship in the copyright sense. The same logic applies, with even greater force, to a software system.
The originality threshold matters equally. Under Portuguese intellectual property legislation, a work qualifies for protection if it reflects the author's own intellectual choices. Courts applying EU-derived copyright doctrine – following harmonisation driven by the InfoSoc Directive and subsequent instruments – have tied originality to the expression of free and creative choices. Where an AI system generates output by identifying statistical patterns across a training corpus and applying probabilistic outputs, it is difficult to identify a human creative choice that is expressed in the resulting work. The machine has processed; it has not created in the legal sense.
Two scenarios sit at opposite ends of the spectrum. First, a human uses an AI tool as a sophisticated instrument – selecting parameters, curating outputs, making editorial decisions about which results to use and how to combine them. Here the human contribution may be sufficient to satisfy the originality threshold. Second, a fully automated pipeline generates output without meaningful human input at any stage beyond the initial deployment of the system. Here the human contribution may be so attenuated that no copyright subsists in the output at all. The commercially important cases cluster in the large middle ground between these poles, and that middle ground is where Portuguese doctrine is least settled.
Database rights offer a partial alternative. Portuguese intellectual property legislation implements the EU Database Directive, which confers a sui generis right on the maker of a database who has made a substantial investment in obtaining, verifying, or presenting its contents. This right is economic in character and does not require human authorship in the copyright sense. A company that assembles a large corpus of AI-generated content within a structured database may be able to claim database rights over the collection, even if individual elements within it are unprotected. However, database rights cover the investment in the database structure, not the underlying content, and they expire after a fixed term. They are a supplement, not a substitute, for copyright protection.
For legal entities operating in Portugal, the interaction with Portuguese corporate legislation is also relevant. Rights that subsist in works created by employees in the course of employment generally vest in the employer under intellectual property legislation, subject to contractual arrangements. But this rule presupposes that a human employee created the work. Where an AI system deployed by the company generates the output, the employment-based attribution rule does not apply in its direct form. Companies relying on employee-generated AI output should document the human editorial contribution with care, precisely because that documentation may later become the only evidence that copyright subsists at all.
Competing interpretations and the gap between statute and practice
Portuguese courts – including the Tribunal da Relação (Court of Appeal) and ultimately the Supremo Tribunal de Justiça (Supreme Court of Portugal) – have not yet produced a settled line of authority on AI-generated works. The doctrinal vacuum means that the analysis in any specific dispute will be built from first principles, drawing on the text of intellectual property legislation. EU-derived harmonisation measures. Additionally, comparative sources from jurisdictions that have confronted the question more directly.
The competing interpretive approaches that have emerged in European legal scholarship and early judicial commentary can be grouped into three camps.
The first approach – call it the strict human-authorship position – holds that copyright simply cannot attach to output that lacks a human author in the traditional sense. On this view, AI-generated works enter the public domain immediately upon creation. The practical effect is that competitors can freely copy the output. For a business whose revenue depends on exclusive rights in that output, this is a direct and measurable commercial loss.
The second approach focuses on the human-AI interaction. It treats the human who designed, trained, selected, and directed the AI system as a qualifying author, provided the system functioned as an instrument of that human's creative vision. The analogy invoked is the camera: early photography raised the same authorship question, and courts ultimately held that the photographer's choices of angle, light, timing, and subject matter were sufficient to satisfy the originality threshold. Proponents of this approach argue that a comparable analysis should apply to AI systems operated by humans with identifiable creative intentions.
The third approach, more radical, would extend a form of copyright – possibly with a reduced term or with restrictions on moral rights – to AI systems or their operators on a strictly economic basis. This approach finds very limited traction in Portuguese doctrine, because it would require legislative change. Portuguese intellectual property legislation, as currently written, does not accommodate it.
The gap between statute and practice is visible in several ways. Practitioners in Portugal report that registrations of AI-assisted works are being accepted by the relevant authorities without systematic scrutiny of the human-authorship question. Registration is not constitutive of copyright under Portuguese law – copyright subsists on creation, without formality – but registration creates an evidentiary presumption. A company that registers an AI-generated work today may find that presumption challenged in litigation tomorrow, when the counterparty adduces evidence of the automated production process.
The Tribunal da Relação has addressed the concept of originality in the context of computer-generated output in tangential ways. In cases involving software and database compilations. However, without directly resolving the authorship question for fully autonomous generative systems. Legal practitioners note that when the authorship question does reach the Supreme Court of Portugal in a well-developed case. The outcome will depend heavily on the factual record of human involvement in the creation process. which is precisely the record most businesses fail to build.
There is also a material gap at the level of enforcement. Even where a company asserts copyright in AI-generated output and a defendant contests the claim, the cost and complexity of establishing the evidentiary record in Portuguese civil proceedings can be substantial. Technology licensing disputes that involve AI-generated content must address the threshold authorship question before reaching the substantive merits of the licence claim. This procedural burden can make enforcement uneconomical for lower-value works, effectively leaving the rights undefended in practice even if they subsist in theory.
For a broader view of how enforcement strategies in intellectual property cases are structured in Portugal. The firm's analysis of intellectual property law in Portugal provides the procedural and strategic context within which AI-related copyright disputes would be resolved.
The EU AI Act compliance dimension and its intersection with IP doctrine
The EU AI Act imposes obligations across the AI value chain that have direct relevance to intellectual property strategy, even though the Act does not itself resolve copyright ownership. AI Act compliance is now a structured legal requirement for providers and deployers of AI systems operating in the EU market, including Portugal.
The most significant intersection between AI Act compliance and IP doctrine concerns transparency and documentation. The EU AI Act requires providers of high-risk AI systems to maintain technical documentation that describes the system's design, development process, training data, and intended purpose. General-purpose AI models face additional transparency requirements, including disclosure of training data summaries and compliance with copyright law in the use of training data.
This documentation requirement, designed for safety and accountability purposes, also creates a record of the AI system's operation that can be used – or used against a party – in an IP dispute. If a company's technical documentation reveals that training data included third-party copyrighted works without licence. The company faces infringement exposure on the input side, separate from and prior to any question about ownership of the output. Algorithmic accountability obligations under the AI Act thus produce legal risk that extends into the IP domain.
The training data question is particularly acute in Portugal because Portuguese intellectual property legislation has implemented the EU text and data mining exceptions narrowly. Research organisations benefit from a broad exception for text and data mining for scientific research purposes. Commercial operators face a much narrower residual exception that applies only where rights holders have not expressly reserved their rights. Many major content providers have now implemented such reservations through standard website terms. A commercial AI system trained on web-scraped data without a systematic rights-clearance programme is likely to have infringed copyright on a substantial scale. Creating a liability exposure that sits at the foundation of the entire output portfolio.
Software liability under Portuguese law adds a further dimension. Digital services built on AI-generated output may themselves carry liability for third parties who rely on that output. Under Portuguese digital services legislation and the broader EU digital services regime, platform operators face obligations around content and accuracy that interact with the question of whether AI-generated content is properly identified as such. Misrepresentation of AI-generated content as human-authored content can give rise to liability independent of the IP questions.
The CAAD – the Centro de Arbitragem Administrativa (Administrative Arbitration Centre) – handles disputes involving public-sector bodies and regulatory authorities in Portugal. Tax and regulatory dimensions of AI deployment, including whether software licensing arrangements involving AI systems are structured correctly for tax purposes, may be reviewed at CAAD in the context of administrative challenges. Businesses that structure technology licensing arrangements for AI-generated content should ensure that the tax and regulatory treatment of those arrangements has been considered alongside the IP strategy.
For clients building or deploying AI systems in Portugal and across the EU. The firm's dedicated practice on AI and technology law in Portugal covers the full scope of AI Act compliance obligations, software liability. Additionally, regulatory interaction with IP rights.
To discuss how the EU AI Act compliance regime and IP ownership questions apply to your specific AI deployment in Portugal, contact us at info@ferrazwhitmore.com.
Cross-border implications for European and international clients
Portugal is an EU member state, which means that Portuguese intellectual property legislation operates within the EU harmonisation framework. The copyright directives that have shaped EU member states' domestic laws apply uniformly across the single market – at least in theory. In practice, the AI authorship question has not yet been resolved at EU level, and divergence between member states' approaches creates material risks for businesses that generate and deploy AI-generated content across multiple jurisdictions.
The Spanish approach to AI-generated works, for example, has developed along broadly similar doctrinal lines to Portugal, given the shared civil law tradition and parallel implementation of EU directives. Clients operating in both markets can benefit from a coordinated IP strategy that applies consistent documentation standards across both jurisdictions. A comparative perspective on how Portugal's approach relates to that of neighbouring civil law systems is developed in the firm's analysis of AI-generated works and IP in Spain.
The UK presents a contrast that is instructive for clients with operations on both sides of the Channel. English copyright legislation contains an express provision for computer-generated works, attributing authorship to the person who made the necessary arrangements for the creation of the work. This provision – unusual in the common law world – provides a degree of legal certainty that Portuguese law does not currently offer. A business that structures its AI output operations through a UK entity may be able to establish cleaner copyright ownership under English law. Subject to the territorial limitations of that copyright and the choice of law rules that would govern a cross-border dispute.
The EU's approach to the AI authorship question is still forming. The EU AI Act does not amend copyright legislation. The European Commission has indicated that the copyright dimensions of AI-generated content will be addressed through separate legislative or soft-law instruments, but no definitive measure is in force. In the interim, the analysis in each member state defaults to domestic copyright doctrine applied to EU harmonisation requirements. This produces a patchwork of legal uncertainty that is commercially disadvantageous for businesses that need to operate at EU scale.
Enforcement of IP rights in AI-generated works across borders raises further complications. A Portuguese court applying Portuguese intellectual property legislation may reach a different conclusion on the authorship question than a German or French court applying its own domestic law to the same work. Where the AI system was developed in one jurisdiction, the training data resided in another. Additionally. The output was distributed in a third, the choice of law analysis alone can become a significant piece of litigation.
The New York Convention framework is relevant where IP disputes are resolved through international arbitration rather than national courts. Technology licensing agreements that include arbitration clauses can designate a neutral seat and a predictable procedural regime, reducing the jurisdiction risk. However, arbitral awards in IP disputes face enforcement challenges where the underlying rights are contested in the enforcement jurisdiction – which is precisely the scenario that AI authorship disputes create.
For clients with cross-border AI deployments, the strategic implication is that IP ownership should be established as clearly as possible at the point of creation. Using the most favourable applicable law. Additionally, documented in a form that is usable in any jurisdiction where enforcement may be needed. Waiting for the law to settle is not a viable commercial strategy when competitors may be building defensible IP positions today.
To explore a tailored cross-border IP strategy for AI-generated content across Portugal and the EU, reach out to info@ferrazwhitmore.com.
Strategic recommendations for securing enforceable IP rights
The legal uncertainty does not mean that businesses must operate without IP protection. It means that protection must be built deliberately, through a combination of legal structuring, technical documentation, and contractual arrangements that do not rely solely on the unresolved copyright question.
Document the human creative contribution at every stage. The single most important defensive measure is a contemporaneous record of the human decisions made in the AI-assisted creative process. This record should capture: the parameters set before generation, the editorial choices made among multiple outputs, the modifications applied to raw AI output, and the human judgment exercised in selecting and combining materials. This documentation serves dual purposes. It supports a copyright claim based on human authorship of the selection and arrangement, and it provides the technical documentation that AI Act compliance requires.
Structure IP ownership through assignment and work-for-hire arrangements. Where AI-generated content is produced by employees or contractors, the contractual chain of IP assignment should be explicit and complete. Relying on statutory employment provisions that presuppose human authorship is insufficient. Contracts should include specific assignment clauses addressing AI-assisted output, fallback provisions for works where copyright may not subsist, and representations regarding compliance with training data licensing obligations.
Use trade secret protection as a complement to copyright. Where copyright cannot be reliably established. Trade secret law under Portuguese and EU legislation may protect the AI model itself, the training data compilation. Additionally, the proprietary processes used in generation. Trade secret protection requires reasonable confidentiality measures – access controls, confidentiality agreements, and internal policies – but it is not dependent on human authorship. It also has no fixed expiry term. For businesses whose competitive advantage lies in the AI system rather than in any individual output, trade secret protection may be more commercially valuable than copyright in the output.
Register works where possible and maintain the evidentiary record. Registration under Portuguese intellectual property legislation creates an evidentiary presumption of authorship and subsistence. While registration does not resolve the authorship question in a contested case, it shifts the initial burden and provides a formal record of the claim. Companies should register AI-assisted works where the human contribution is credibly documented, and should maintain that documentation alongside the registration.
Clear training data rights before deployment. The input-side liability exposure – copyright infringement in the training data – is a more immediate and better-established legal risk than the output-side authorship question. A comprehensive training data audit should identify the sources of training data, the basis on which each category was used, and the applicability of any statutory exceptions. Where the audit reveals gaps, those gaps should be remediated before the system is deployed commercially. The cost of a training data licensing programme is substantially lower than the cost of defending an infringement claim at scale.
Prepare for the regulatory trajectory. The EU legislative process on AI and copyright will produce new rules within a commercially relevant timeframe. Businesses should monitor the European Commission's work programme on AI and creativity, track the positions being taken by major rights-holder organisations in consultation processes. Additionally. Structure their AI IP strategy to be adaptable rather than locked into the current doctrinal position. Companies that have built clean documentation practices and clear contractual structures will be better positioned to adapt when legislative clarity arrives.
Outlook: regulatory trajectory and what to monitor
The regulatory trajectory for AI-generated works in Portugal and across the EU points toward increasing formalisation of the rules, but not necessarily toward resolution of the core authorship question in the near term. Several developments merit close attention.
At EU level, the AI Act's phased implementation schedule means that the transparency and documentation obligations for general-purpose AI models will become fully applicable progressively. Providers of foundation models used in commercial content generation must begin compliance programmes now. The interaction between AI Act compliance documentation and IP rights management has not yet been systematically addressed by regulators, but it is becoming an operational reality that businesses must manage.
The European Court of Justice's development of the originality doctrine in copyright cases continues to provide the interpretive baseline for Portuguese courts. Any ECJ ruling that bears on the concept of creative choices or human authorship in the context of automated processes will be applied in Portuguese proceedings. Practitioners in Portugal monitor ECJ copyright case law closely as a leading indicator of how domestic courts will resolve contested questions.
The Portuguese legislature has not introduced specific AI authorship legislation, and no bill is currently before the Assembleia da República that would address the question directly. The legislative approach in Portugal has historically followed EU harmonisation measures rather than anticipating them. This means that domestic legislative change is unlikely before the EU adopts a position, which in turn means that the doctrinal vacuum is likely to persist for a commercially significant period.
The positions of major creative industries – music, film, publishing, and software – in Portuguese and EU consultations will shape the eventual legislative outcome. Rights-holder organisations have consistently argued for strong human-authorship requirements and for copyright protection in training data. Technology companies have argued for broader text and data mining exceptions and for recognition of AI-assisted authorship. The outcome of this political contest will determine whether Portuguese intellectual property legislation moves toward the UK model of computer-generated works protection or toward a strict human-authorship standard that leaves AI output unprotected.
In the interim, the practical advice for businesses operating in Portugal is to treat the current period as a window in which IP positions can be staked out. through documentation. Registration, contractual structuring. Additionally, trade secret protection. before courts and legislators close off the options. Lost opportunity in IP is rarely recoverable. A competitor that builds a defensible IP position in AI-generated content today, when the law is uncertain, may hold an entrenched advantage by the time the law settles.
Self-assessment checklist for AI and IP strategy in Portugal
This analysis is applicable to your situation if one or more of the following conditions are met:
- Your business generates commercial content – text, images, audio, code, or other works – using AI or AI-assisted tools, and that content forms part of your product or revenue model.
- You operate in Portugal or distribute AI-generated content to the Portuguese or EU market.
- You license, assign, or monetise rights in AI-generated works.
- You have deployed or are developing a generative AI system trained on third-party data.
- You are entering into technology licensing arrangements that include AI-generated output.
Before initiating a structured IP strategy for AI-generated content in Portugal, verify the following:
- Have you documented the human creative contribution to AI-assisted works in a contemporaneous and retrievable record?
- Have you conducted a training data audit to identify the source and licensing basis of all data used to train your AI systems?
- Do your employment and contractor agreements include explicit IP assignment provisions for AI-generated and AI-assisted output?
- Have you assessed whether trade secret protection should be applied to the AI model, the training corpus, or the generation methodology?
- Have you reviewed your technology licensing agreements for adequacy in the context of uncertain copyright subsistence?
- Are you monitoring EU AI Act compliance milestones that are relevant to your AI system's risk classification?
Frequently asked questions
Q: Can a Portuguese company own copyright in content generated entirely by an AI system, without human creative input?
A: Under Portuguese intellectual property legislation as currently interpreted, copyright requires a human author whose creative choices are expressed in the work. Where output is generated entirely autonomously by an AI system, with no identifiable human creative contribution, the weight of doctrinal opinion holds that copyright does not subsist. The company may seek protection through trade secret law or database rights in appropriate cases, but direct copyright ownership in fully autonomous AI output is not reliably available under current Portuguese law. This position may change when the EU adopts specific measures on AI authorship, but no such instrument is yet in force.
Q: How long does it typically take to resolve an IP dispute involving AI-generated content before Portuguese courts?
A: First-instance proceedings in Portuguese civil courts in IP disputes typically take between one and three years to reach judgment. Depending on court workload, the complexity of the evidentiary record. Additionally, whether expert witnesses are required. An appeal to the Tribunal da Relação adds further time. The authorship question in AI-generated works adds a novel doctrinal element that may require more extensive argumentation than a standard copyright dispute, lengthening the process. For this reason, practitioners in Portugal recommend resolving IP ownership questions contractually before disputes arise, rather than relying on litigation to establish rights retrospectively.
Q: Does AI Act compliance in Portugal address who owns the copyright in AI-generated output?
A: The EU AI Act does not amend intellectual property legislation and does not allocate copyright ownership in AI-generated works. AI Act compliance obligations – covering transparency, documentation, algorithmic accountability, and risk management – operate as a regulatory overlay on top of existing IP law. Compliance with the AI Act is necessary but not sufficient for IP protection. It does, however, produce documentation that can support or undermine an IP claim in litigation, making AI Act compliance practices directly relevant to IP strategy even though the two regimes are formally separate. Engaging a lawyer in Portugal with experience across both AI regulation and IP law is the most efficient way to manage this intersection.
About Ferraz & Whitmore
Ferraz & Whitmore is an international law firm based in Lisbon, advising business clients across 46 jurisdictions on AI and technology law, intellectual property protection, and cross-border regulatory compliance. As a law firm in Portugal that combines Portuguese civil law expertise with English common law tradition, the firm advises technology companies, investors. Additionally. Institutional clients on AI Act compliance, software liability, technology licensing, digital services regulation. Additionally, IP strategy for AI-generated content. Our team has advised on AI and IP matters across both civil law and common law systems, working with clients from market entry through to enforcement and dispute resolution. The firm's Lisbon base provides direct access to Portuguese and EU regulatory processes, while our common law expertise supports enforcement and arbitration strategies in English-speaking jurisdictions. Ferraz & Whitmore participates in cross-border practice groups focused on AI regulation and intellectual property, covering matters before Portuguese courts, the EU institutions, and international arbitral bodies. To discuss your AI and IP strategy in Portugal or across the EU, 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.