A software company based in Bogotá deploys a generative AI tool that produces marketing copy, graphic designs, and product documentation daily. The company registers the outputs under its corporate name with the Dirección Nacional de Derecho de Autor (Colombia's National Copyright Directorate). Months later, a competitor reproduces those materials. When the company attempts to enforce its rights, it encounters a question that Colombian copyright legislation has not yet answered with precision: who. If anyone, is the author of a work produced autonomously by a machine?
Under Colombian intellectual property legislation, copyright protection attaches to original works created through human intellectual effort. AI-generated works occupy an unresolved doctrinal space: no statute expressly grants or denies protection to machine-produced outputs, and no Colombian court has issued a binding ruling on the question. Businesses relying on AI-generated content as a commercial asset face real, immediate risk of holding rights that may prove unenforceable.
This analysis examines the doctrinal foundations of Colombian copyright law, the competing interpretations practitioners advance, the gap between formal rules and daily practice. The cross-border dimensions that affect international clients. Additionally, the strategic steps available to businesses operating in this environment.
Doctrinal foundations: what Colombian copyright law actually says
Colombia's copyright system sits within the civil law tradition. Intellectual property legislation in Colombia draws heavily from the Andean Community's supranational rules – particularly Decision 351 of the Andean Community, the regional copyright regime that governs all member states including Colombia. That regional instrument defines an "author" as the natural person who creates a work. The word "natural" carries doctrinal weight. It excludes legal persons from original authorship. It also, by clear implication, excludes artificial systems.
Domestic copyright legislation reinforces this position. The principal statute governing author's rights in Colombia conditions protection on an original creation of the human mind. "Original" in Colombian doctrine means that the work reflects the personality of the creator – a test borrowed from the French-influenced civil law tradition. A generative AI model, however sophisticated, has no personality in the legal sense. It produces outputs through statistical inference, not through the kind of individual expression that copyright doctrine was designed to reward.
This doctrinal baseline creates an immediate practical problem. If no human is the author of an AI-generated work, the work falls into the public domain at the moment of creation. Any party – including direct competitors – may freely copy, distribute, and adapt it. For a business that has invested substantially in developing or deploying an AI system to produce commercial content, that outcome represents a significant lost opportunity.
Colombian legislation does provide for related rights – protections that sit beside copyright and cover producers, broadcasters, and performers. Some practitioners argue that a company deploying an AI tool could claim a producer's right over the resulting work, by analogy with the protection afforded to producers of sound recordings. The analogy is imperfect. Producer's rights in Colombian legislation are designed for industries with defined production processes, not for the open-ended output of generative systems. Courts have not validated this approach, and relying on it involves real doctrinal uncertainty.
A separate line of argument focuses on the human-in-the-loop. Where a human operator provides detailed prompts, selects among outputs, arranges and edits generated material. Additionally, exercises genuine creative judgment throughout the process. Practitioners contend that the resulting work is a human-authored work assisted by a machine. not a machine-generated work in the pure sense. This argument has surface plausibility. It mirrors the accepted treatment of photographs, films, and computer-assisted design, all of which involve mechanical processes guided by human choices. However, it introduces a spectrum problem: there is no agreed threshold that distinguishes "genuinely creative human involvement" from "mere instruction of a machine." Different examiners at the Dirección Nacional de Derecho de Autor may apply different standards.
Competing interpretations and the gap between statute and practice
Three interpretive positions circulate among Colombian practitioners. Understanding each – and the conditions under which courts might adopt one over the others – is essential for building a defensible IP strategy.
The first position holds that AI-generated works are simply unprotectable under current law. The statute says what it says: authorship requires a human creator. No registration, no contract, and no corporate structure can supply what the law withholds. On this view, businesses seeking to protect AI-generated content must resort entirely to trade secret law, contractual confidentiality obligations, and practical barriers to copying – not copyright. Trade secret protection under Colombian commercial legislation is available and can be effective, but it requires genuine secrecy. Once an AI-generated work is published or commercialised, trade secret protection dissolves. The business is left exposed.
The second position takes a more pragmatic approach. The Dirección Nacional de Derecho de Autor currently accepts registrations of works described as "created with the assistance of AI tools," provided a human author is named. In practice, registration proceeds on the basis of the applicant's representation. The directorate does not conduct a substantive inquiry into the degree of human creative involvement. This administrative tolerance does not resolve the underlying legal question. A registration can be challenged and annulled. But it does provide a procedural starting point – and it signals, at minimum, that the administrative body has not closed the door entirely to AI-assisted works.
The third position looks to software liability and technology licensing doctrine for analogies. Under Colombian intellectual property legislation, software is protected as a literary work. The person who commissions the software owns the economic rights, by default, if the work is created under an employment or service contract. Some practitioners extend this logic to AI outputs: the company that trains, fine-tunes, or substantially customises an AI model may be characterised as the "author by employment" of the outputs that model generates. This argument is creative but contested. The software-as-literary-work doctrine was designed for human programmers whose code reflects intellectual choices. Applying it to the outputs of a trained neural network stretches the analogy significantly.
Where does this leave the practitioner? The honest answer is that Colombian courts have not yet issued published, binding guidance on any of these positions. Decisions by first-instance and appellate civil courts in Colombia are not formally binding on other courts. the system follows civil law doctrine rather than the common law rule of precedent. Though the Corte Suprema de Justicia (Supreme Court of Justice) and the Consejo de Estado (Council of State) do establish interpretive lines that lower courts follow in practice. Until one of these higher bodies issues a reasoned opinion on AI authorship, practitioners must advise clients based on doctrinal analysis, administrative practice, and comparative law – all of which point in different directions.
The gap between statute and practice is most visible in the registration process. The directorate registers AI-assisted works. Courts may not protect them. A business that invests in registration and then discovers, during litigation, that its rights are unenforceable has lost time, money, and – potentially – its entire competitive advantage in a product line. Algorithmic accountability for that outcome rests with no single actor. It is a systemic gap.
For a tailored strategy on AI-generated content protection in Colombia, reach out to our AI and technology law practice in Colombia for an initial assessment.
Training data, infringement exposure, and the software liability dimension
The authorship question is only one side of the AI-and-IP problem in Colombia. The other side concerns liability: when an AI system generates content that reproduces, imitates, or is substantially similar to a protected work, who bears responsibility?
Colombian copyright legislation does not include a broad text-and-data-mining exception. Training an AI model on a corpus of protected works – images, texts, musical compositions – involves reproduction of those works, at minimum during the training process. Whether that reproduction qualifies as an infringement depends on whether a recognised exception applies. The principal exceptions in Colombian copyright law cover private use, educational use, and quotation. None of them maps cleanly onto commercial AI training at scale.
This creates software liability exposure for Colombian businesses that train or fine-tune AI models on third-party content without obtaining licences. It also creates exposure for businesses that import or deploy AI models trained elsewhere, if those models were trained on works protected under Colombian law or under an Andean Community decision that Colombia enforces. The liability chain can extend from the original model developer to the Colombian operator of a digital service built on that model.
Digital services built on generative AI in Colombia operate within a regulatory environment that is still forming. There is no dedicated AI Act equivalent in Colombian law at the time of writing. The government has published discussion documents on AI policy, and legislative proposals have circulated in the Colombian Congress, but none has been enacted into law. Businesses seeking AI Act compliance as a single reference framework will not find one in Colombia. Instead, they must navigate a combination of copyright legislation, data protection legislation, consumer protection rules, and the general civil law of obligations – each applying to different aspects of the AI development and deployment chain.
Technology licensing is the most immediate practical tool available. A well-drafted technology licence can allocate risk between an AI model provider and a Colombian deployer, specify which training data was used and how rights were cleared. Define the scope of permitted use. Additionally, establish indemnification obligations if third-party claims arise. Colombian courts enforce commercial contracts under general civil and commercial legislation. A governing law clause selecting Colombian law, combined with an arbitration clause referencing the Centro de Arbitraje y Conciliación de la Cámara de Comercio de Bogotá (Bogotá Chamber of Commerce Arbitration Centre). Provides a predictable and enforceable dispute resolution mechanism.
Businesses that fail to address these licensing questions at the contracting stage frequently discover the omission only when a dispute arises. By that point, the absence of clear ownership documentation, the lack of training data provenance records, and the ambiguity about which party bears infringement risk combine to make the litigation position difficult to defend.
Cross-border implications for Americas clients
Colombia is a party to the Andean Community Decision 351 framework, the Berne Convention, the TRIPS Agreement, and the United States-Colombia Trade Promotion Agreement. Each of these instruments shapes the cross-border dimension of AI-generated works in different ways.
The Berne Convention requires member states to protect foreign works on the same terms as domestic works. A US company whose AI-generated works receive some form of protection in the United States – even limited, administrative protection – cannot automatically rely on that status in Colombia. The question of whether protection exists is governed by Colombian law, not by the law of the country of origin. A business that has invested in an IP portfolio in the United States may hold assets that have uncertain or no value in the Colombian market.
The US-Colombia Trade Promotion Agreement contains intellectual property chapters that broadly align Colombian standards with US norms. However, neither agreement addresses AI authorship directly. The treaty text predates the current generation of generative AI by years. Practitioners seeking to invoke treaty protections for AI-generated works must argue by analogy – a weaker position than relying on an express provision.
The Andean Community framework is particularly relevant for businesses operating across Colombia, Peru, Ecuador, and Bolivia. Decision 351 defines authorship and protection at the supranational level. Any interpretation by Colombian courts that diverges significantly from Andean Community doctrine risks conflicting with supranational law. The Tribunal de Justicia de la Comunidad Andina (Andean Community Court of Justice) has jurisdiction to issue interpretive rulings on Decision 351. Additionally. Domestic courts are obliged to request such rulings in cases involving the interpretation of Andean norms. An AI authorship case in Colombia could, in principle, trigger a reference to the Andean Court – producing a ruling with effect across the entire Andean Community.
For businesses with operations in multiple Andean jurisdictions, this supranational dimension matters strategically. A ruling that denies copyright protection to AI-generated works under Decision 351 would bind all member states simultaneously. Conversely, a ruling that finds some form of protectable interest in AI outputs would represent a significant development for the entire region.
The comparison with other jurisdictions is instructive. Courts and legislators in the United States, the United Kingdom, and the European Union have each grappled with AI authorship from different starting points. The UK's copyright legislation contains an explicit provision addressing computer-generated works – a provision that has no Colombian equivalent. The EU's approach under its evolving AI regulation emphasises transparency and accountability rather than authorship. For a comparative view of how these questions are resolving in the United States, our analysis of AI-generated works and intellectual property in the United States provides a useful reference point.
The civil law tradition in Colombia also shapes the cross-border analysis in ways that common law practitioners sometimes overlook. Colombia does not apply the "work for hire" doctrine in the same form as the United States. Economic rights can be transferred by contract, but moral rights – the right of attribution and the right to object to derogatory treatment of a work – are inalienable under Colombian law. An AI-generated work, if it has no human author, has no moral rights to protect. This is a doctrinal difference with practical implications: a Colombian court may refuse to apply moral rights doctrine to AI-generated content even if a foreign jurisdiction would treat those rights as subsisting in a human operator or developer.
To discuss how intellectual property rules apply to your cross-border AI operations in Colombia, contact our team at our intellectual property practice in Colombia.
Strategic recommendations and the regulatory outlook
Businesses operating with AI-generated content in Colombia face a situation where waiting for legal certainty is itself a strategic choice – and not necessarily a wise one. The following recommendations reflect current doctrine, administrative practice, and the likely trajectory of Colombian AI regulation.
Document human creative involvement at every stage. The strongest copyright claim available under current Colombian law rests on demonstrable human authorship. Businesses should maintain records of the prompts used, the editorial choices made, the selections among AI outputs, and any modifications applied. These records support the argument that the final work reflects human creativity rather than autonomous machine generation. The more granular the documentation, the stronger the claim.
Register with the Dirección Nacional de Derecho de Autor, but do not rely on registration alone. Administrative registration provides evidentiary weight and a starting point for enforcement. It does not cure underlying doctrinal uncertainty. Treat registration as one layer of a multi-layered protection strategy, not as a definitive resolution of the authorship question.
Deploy trade secret protection for AI outputs that are not publicly disclosed. Where commercial content is used internally or shared only under confidentiality obligations, trade secret protection under Colombian commercial legislation provides meaningful coverage. The trade secret regime does not require human authorship. It requires genuine secrecy, reasonable protective measures, and commercial value – all of which a well-managed AI development programme can satisfy.
Audit training data provenance before deploying or licensing AI models in Colombia. The absence of a text-and-data-mining exception in Colombian copyright legislation means that unlicensed use of protected training data is a real infringement risk. A provenance audit – examining what data was used to train a model, on what terms, and with what rights clearances – should be conducted before commercial deployment. This is equally important for businesses importing AI models developed elsewhere.
Draft technology licences that allocate AI-specific risks expressly. A licence agreement that addresses training data, output ownership, infringement indemnification, and regulatory compliance in Colombia-specific terms provides far better protection than a generic software licence. The governing law and dispute resolution clauses are particularly important given the current state of uncertainty.
On the regulatory horizon, Colombia is likely to adopt dedicated AI legislation within the next several years. Legislative proposals circulating in 2025 and early 2026 suggest a risk-based approach broadly influenced by the EU's model, though calibrated to Colombia's economic conditions and legal tradition. Businesses that build compliance infrastructure now – including documentation practices, licence structures, and risk allocation mechanisms – will be better positioned to adapt to a new regime than those that wait. The lost opportunity is not merely theoretical: early movers in setting industry practice tend to influence the regulatory debate itself.
Algorithmic accountability is increasingly a theme in Colombian regulatory discussion. Proposals under consideration would require transparency about how AI systems make decisions in certain high-risk contexts. While the initial focus is on public sector and financial services applications, the accountability logic extends naturally to content generation. Businesses that can demonstrate responsible AI deployment practices – including clear ownership structures, provenance documentation, and user-facing disclosure – will face less regulatory friction as the legislative environment develops.
Frequently asked questions
Q: Can an AI system hold copyright over a work it generates in Colombia?
A: No. Under Colombian copyright legislation, authorship requires a human creative mind. An AI system has no legal personality and cannot hold rights. The question courts and practitioners are still working through is which human – developer, operator, or user – holds any protectable interest in the output.
Q: How long does it take to register an AI-assisted work with the Colombian copyright authority?
A: Registration with the Dirección Nacional de Derecho de Autor typically takes several weeks from the date of application, depending on documentation and the examiner's workload. Registration is declarative rather than constitutive under Colombian law, so rights arise at creation – but registration provides critical evidentiary weight in any dispute.
Q: Is it a misconception that training an AI on third-party content is automatically lawful in Colombia?
A: Yes, this is a common misconception. Colombian copyright legislation does not contain a broad data-mining or text-and-data-mining exception comparable to those found in some other jurisdictions. Using third-party protected content to train an AI model without authorisation may constitute infringement, regardless of whether the resulting output reproduces the original work verbatim.
About Ferraz & Whitmore
Ferraz & Whitmore is an international law firm based in Lisbon, advising business clients across 46 jurisdictions. Our AI and technology law practice supports companies deploying generative AI systems in Colombia and across the Americas, covering copyright registration strategy, technology licensing, training data audits, and regulatory compliance. Engaging a lawyer in Colombia with cross-border experience in both civil law and common law systems is particularly valuable when AI-generated works cross jurisdictional lines. As an international law firm in Colombia and across Latin America. We combine Portuguese civil law expertise with English common law tradition to provide practical guidance on the full spectrum of digital services and algorithmic accountability questions that AI deployment raises. Our attorneys have advised on technology licensing and IP strategy matters across civil law and common law systems throughout the Americas and Europe. To discuss how AI regulation and intellectual property law apply to your specific operations in Colombia, 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.