AI & Intellectual Property 2026 Comparisons

Mexico has not yet enacted specific AI-related legislation. Nonetheless, the legal framework has been amended by including provisions related to certain AI-related activities, and there is Mexican Supreme Court case law that expressly recognises human authorship in cases using AI-assisted material, which is currently shaping the AI landscape in the country. An example of this is the inclusion of potential trade mark infringements which could be carried out by using artificial intelligence (AI), following a recent amendment in the Federal Law on the Protection of Industrial Property (LFPPI).

Moreover, the Federal Copyright Law (LFDA) amended the protection of image and voice, which already required explicit consent, by explicitly mentioning that it also covers results generated by AI. Any use other than that agreed upon requires new authorisation and new compensation.

In addition, a new amendment recognises the right to prevent the identity theft of performances by AI when clones are generated, or when the voice is simulated in an identifiable way. This is designed to prevent deepfakes – with the exception of parody, satire or creative imitation that does not seek to replace the performer.

There have also been amendments in the Federal Labour Law to regulate professional services contracts for artists and performers through the specification of the conditions and remuneration for the use of their image or voice through AI systems. Furthermore, severance in termination of contracts when it constitutes substitution of a human over AI technologies shall be increased.

In the Federal Criminal Code, amendments include the use of AI non-consensual sexual content as a crime.

Finally, there is jurisprudence issued by the Mexican Supreme Court regarding the granting of authorship over works only for human creations. All other content – even AI-assisted or generated by AI technologies – will not qualify for copyright.

In the absence of a Mexican legal framework on AI, the jurisprudence issued by the Mexican Supreme Court – which has interpreted the existing requirement in the LFDA for human authorship for copyright protection – has become the operative standard. 

Mexico has been holding back on the creation of a specific AI legal framework due to the absence of regional AI regulations covering this area. Changes have been consistent with the internal set of changes that AI has posed for authorities, primarily in the copyright sector.

Mexico has not yet formulated definitions and legal treatment for such AI concepts. However, the standardised treatment when considering the outputs generated by such AI technologies is that no copyright protection is granted, as they lack human authorship as per the interpretation of the Mexican Supreme Court.

Currently, all general intellectual property authorities in Mexico are responsible for reviewing AI-related cases within their specific sectors of operation. The National Institute of Copyright (INDAUTOR) oversees cases involving copyright, while the Mexican Institute of Industrial Property (IMPI) handles matters concerning industrial property.

According to the Mexican LFDA, computer programs, including AI programs, are protected in the same terms as literary works. This protection extends to operating programs as well as application programs, whether in the form of source code or object code. Exceptions include those computer or AI programs that are intended to cause harmful effects to other programs or equipment, or that violate the rights of third parties.

The LFDA expressly grants copyright protection to computer programs, whether in the form of source code or object code. Moreover, there is an impediment to copyright protection over ideas themselves, formulas, solutions, concepts, methods, systems, principles, discoveries, processes and inventions of any kind. In this sense, isolated prompts, simple descriptions, processes or functionalities would not be granted copyright protection. However, if they could constitute any other expression under artistic or literary works, this could endorse their protection as copyrights. Examples of this could be a book including a set of prompts, or a full prose text explaining the complete discovery of certain results.

In Mexico, all inventions must have a “human creation” core. In other words, the category of inventor is reserved exclusively for humans. With this in mind, there is a structural problem regarding inventions created by AI, because if they consist of mathematical methods, results in computer programs or forms of presentation of information, for instance, they are not patent-eligible under the LFPPI.

In Mexico, a trade secret refers to any information of industrial or commercial application that is kept confidential by the person who exercises legal control over it. The first element for considering this type of protection would be the legal control over such information. Specifically, AI models are a service created by their developers – in this sense, their terms and conditions over the treatment of this information would first need to be taken into consideration to establish the level of control over the information intended to be protected under a trade secret.

Moreover, typical contractual obligations such as confidential clauses, agreements, non-disclosure agreements and flagging of content as confidential would be deemed necessary to preserve the confidentiality and restrict access to said information.

The LFDA requires human authorship for copyright protection. In this regard, all datasets, databases, corpora, annotations, labels, embeddings or synthetic datasets created by the model creator would not be granted copyright protection. Nonetheless, the LFDA recognises the protection of databases when they involve a creative disposition of information, as collections, by virtue of their selection or the arrangement of their content or subject matter.

In addition, the LFDA states that if databases are not original they will not acquire copyright protection but a five-year exclusive-use right. In the same sense, it gives the holder rights over reproduction, reordering/adaptation, distribution and public communication, and over the results of that reordering. This is relevant for embeddings, indexes and derived datasets. Anti-circumvention rules are also provided in Article 112 and Articles 114 bis–114 quáter), which add an access-control layer against scraping past paywalls or digital rights management.

Mexico has a civil law perspective when it comes to the protection of copyrights. With that in mind, all usages of copyrighted material must comply with the prior approval of authors/rights-holders, unless such uses could be endorsed by the following limitations or exceptions provided in the LFDA:

• quotation of texts, provided that the amount used cannot be considered a simulated and substantial reproduction of the work’s content;
• reproduction of articles, photographs, illustrations and commentaries relating to current events, published in the press or broadcast on radio or television, or any other means of dissemination, unless expressly prohibited by the copyright-holder;
• reproduction of parts of the work for criticism and scientific, literary or artistic research;
• reproduction, once and as a single copy, of a literary or artistic work for the personal and private use, without commercial intention;
• reproduction, communication and distribution through drawings, paintings, photographs and audiovisual means of works visible from public places; and
• publication and performance of artistic and literary works for non-profit purposes for people with disabilities.

Only educational or research institutions, or institutions not engaged in commercial activities, are able to benefit from these limitations/exceptions. Furthermore, these limitations/exceptions must always be considered under the three-step test given by the Berne Convention – that is to say, use under these circumstances must mean that the normal exploitation of the work is not affected, nor be for commercial purposes, always citing the source and without altering the work.

Mexico does not have a text and data mining (TDM) concept available as a limitation/exception over the use of copyrighted material by AI technologies. Nevertheless, a key feature for the operation of the set of limitations/exceptions provided in the LFDA, if carried out by legal entities, is educational or research purposes or being institutions not engaged in commercial activities. To date, there been no relevant case law addressing these matters.

In Mexico, the practical effects of the use of content to train AI models have not yet reached case law. There have been discussions in the collective management associations area – specifically those in charge of the protection of music publishers, songwriters, lyricists, composers and book writers – to try to settle blanket licences for the use of their works as objects of training data for AI technologies. However, these are only statements made publicly by the collection rights management associations, and are not grounded claims in court or before any Mexican authorities.

Mexico follows the civil law approach for the protection of copyright. In this sense, the authorisation of authors/rights-holders is always mandatory for the use of copyrighted content, unless limitations/exceptions could be applicable.

Considering the foregoing, in the realm of AI, any reservation over the use of works, water marks, terms and conditions in websites, or machine-readable opt-outs would express the intention of the author/rights-holders to avoid usage of their works, and must be respected.

Mexican law does not currently impose specific obligations on AI developers to document training data sources (or the related). These matters remain subject to the ordinary rules applicable to the relevant IP rights, as well as to contractual and confidentiality obligations. Maintaining records of datasets, permissions and model development may nevertheless be important in practice. Where datasets, servers, developers and users are located in different jurisdictions, the place of infringement would generally depend on the specific act and the applicable territorial rules, rather than simply on where the relevant server or dataset is located, though the location where the system operates would also matter.

In Mexico, there is currently no specific legal provision regulating an infringement based on AI training or similar or by acts carried out automatically by an AI agent. Consequently, it is necessary to establish that the model developer or provider copied, reproduced or used the copyright or IP right without consent. The analysis would depend on the acts involved, including whether protected works were reproduced, transformed, communicated to the public or otherwise used without authorisation.

Ultimately, the claimant would need to establish that the conduct falls within one of the infringement grounds provided for under Mexican IP legislation. The fact that the conduct involves AI does not create an additional or separate requirement for establishing infringement; rather, the same legal standards would apply to the relevant act, regardless of whether it was performed by a person directly (without AI) or through an AI system.

Mexican law and case law currently provide limited guidance on these issues, and the authors are not aware of any pending Mexican case that has established a definitive standard related to this matter.

In Mexico, there is currently no case law addressing whether model weights, parameters, embeddings, caches or other internal AI artefacts may constitute infringing copies, adaptations or extracted parts of protected works or databases. Mexican courts have likewise not yet developed specific criteria for assessing memorisation, regurgitation, substantial similarity or de minimis copying in the AI context. These issues would therefore be assessed under the general copyright rules according to the LFDA governing reproduction, transformation and other unauthorised uses of protected works and databases.

The question would be whether the particular use results in an unauthorised copy or reproduction. There is also no established Mexican rule providing that the reproduction or retention of a small portion of a work is, by definition, de minimis and therefore non-infringing; this would have to be assessed on a case-by-case basis.

Mexican copyright law is based primarily on civil law principles and does not recognise “contributory” or “vicarious” infringement as separate grounds of liability, as is the case in common law jurisdictions. Therefore, a model provider would not be liable simply because its system is used by a user, agent or downstream application to commit an infringement. Liability would instead depend on the provider’s own conduct and whether it falls within any of the infringement provisions under the LFDA. The provider’s authorisation, knowledge, involvement or control may be relevant when assessing its conduct, but these factors do not operate as independent grounds of liability equivalent to contributory or vicarious infringement.

As mentioned previously (see 2.4 Trade Secrets and Confidential Information), trade secret protection under the LFPPI covers confidential information of industrial or commercial application that is subject to legal control by the person who lawfully controls it. The unauthorised disclosure, use or appropriation of such information may therefore give rise to liability. In the AI context, the infringement may arise at the point at which confidential information is provided to, uploaded to or otherwise made available to an AI system without authorisation; it would not always be necessary to wait until the system produces an output incorporating the information, as may be the case with other forms of IP infringement where the infringing act is tied to the subsequent use.

A key issue will be the measures adopted by the trade secret holder to preserve the confidentiality of the information. If an AI system subsequently reproduces, infers or reveals confidential information, this may give rise to additional concerns that may be independent from the liability of the person who initially provided the information to the system.

AI developers and model providers may rely on the exceptions and limitations generally available under Mexican copyright and IP legislation. Their availability, however, does not necessarily mean that they will be effective in the particular circumstances. The LFDA is a matter of public order, and certain rights and statutory provisions cannot simply be waived or excluded by contract or by specific terms and conditions.

There is currently no significant Mexican case law specifically addressing the application of these defences in the AI context. In practice, a relevant defence may instead be that the conduct does not fall within any of the infringement provisions established by law in the first place. 

Under Mexican law, an AI-generated output may infringe copyright where it reproduces or otherwise incorporates protected elements of a pre-existing work. This may be the case where the output is substantially similar to a protected work or where a prompt, reference image or other input results in the reproduction or transformation of protected content. The fact that the output was generated through AI would not, in itself, prevent the copyright owner from asserting infringement. At the same time, purely AI-generated content is not itself eligible for copyright protection, as Mexican law requires a human author.

The mere imitation of an artistic style, genre, voice or technique, without reproducing protected elements of a particular work, would generally not amount to copyright infringement under Mexican law. The analysis would therefore focus on whether the AI output appropriates protected content from an existing work, rather than the fact that the output resembles the general style or technique of another creator.

Where the output imitates a particular artist’s style, voice or technique without literal copying, however, the line may be less clear and would ultimately depend on the circumstances of each case. As a general matter, such imitation, standing alone, would not rise to the level of copyright infringement.

In Mexico, a user, deployer or customer may incur IP infringement even where the infringing output was generated unknowingly or without an intention to infringe. The fact that an AI system produced the content would not shield the person who subsequently uses, publishes or commercialises it from infringement. This is particularly relevant where the output is used commercially or for profit, although infringement is not necessarily limited to such cases. However, in this case it will be relevant to also assess the AI system’s terms and conditions to confirm user liability.

The LFPPI expressly provides that trade mark and unfair competition infringements may also be sanctioned when carried out through the use of AI. Accordingly, AI-generated outputs that reproduce or imitate third-party trade marks, logos or other protected signs may give rise to liability under the applicable infringement provisions.

The fact that the content was generated by AI would not change the infringement analysis. Rather, the relevant question would be whether the use of the output falls within any of the infringement or unfair competition provisions established under the LFPPI. This is particularly relevant following the recent express inclusion of AI in the LFPPI, which makes clear that the use of AI does not exempt conduct that would otherwise constitute an infringement.

In Mexico, there is no established distinction between “direct” and “indirect” infringement as separate causes of action in the way that these concepts are understood in other jurisdictions. In fact, the important matter is whether the conduct falls within one of the infringement provisions of the LFPPI. As mentioned previously, the LFPPI also expressly provides that conduct will be sanctioned even if there is use of AI.

Accordingly, the mere generation of a product design, technical instructions, software code, manufacturing parameters, chemical or biological candidates, or other technical output by an AI system would not necessarily constitute infringement. The risk would generally arise when the output is used in a manner covered by the applicable infringement provisions, such as:

• manufacturing products protected by a patent or industrial design without the right-holder’s consent or the relevant licence;
• offering for sale, placing into circulation or using products incorporating a patented invention without such consent or licence; or
• using a patented process without the corresponding authorisation.

Thus, the focus under Mexican law is less on the AI-generated nature of the output itself and more on the subsequent conduct and whether it falls within the acts expressly prohibited by the LFPPI.

As mentioned previously, there have been amendments in the LFPPI to expressly include trade mark infringements as also being penalised if carried out through the use of AI.

Copyright protection is limited to works resulting from human creativity. In its recent decision in the amparo appeal of June 2025, the Supreme Court held that content generated by an AI system cannot be protected as a copyrighted work because authorship under Articles 3 and 12 of the LFDA must be attributed to a natural person. Importantly, the Court reached this conclusion even though the claimant had provided instructions and parameters to the AI system, including his own photographs.

Mexico does not recognise copyright or a separate copyright-like right in AI-generated works where there is no human author. Under the LFDA, authorship must be attributed to a natural person, and the Supreme Court recently confirmed this – ie, that AI-generated works do not qualify for copyright protection as a general rule. Accordingly, where there is no human author, no copyright ownership, term or scope of protection arises.

As mentioned previously, the Supreme Court recently confirmed that copyright authorship must be attributed to a natural person. Therefore, the use of an existing work, reference images or other protected material as the basis for an AI-generated output does not, by itself, create joint authorship.

Mexican law does not currently impose an obligation to disclose the use of AI when applying for copyright, patent, design or other IP rights. However, where AI involvement is relevant to the existence of the claimed right, disclosure would be consistent with the principle of good faith that applies to legal proceedings and registration procedures. This is particularly relevant for copyright, given the Supreme Court’s recent ruling that authorship must be attributed to a natural person and that AI-generated material is not eligible for copyright protection.

Accordingly, failure to disclose AI involvement would not necessarily invalidate a registration or right. However, if a right was granted on the basis that the claimed work satisfied the statutory requirements when, in fact, it was AI-generated and therefore not eligible for protection, its validity could be challenged on the basis that it was granted contrary to the applicable provisions of the LFDA.

In Mexico, an AI system cannot be named as an inventor or co-inventor. The LFPPI expressly refers to the natural/physical person who makes the invention as the holder of the corresponding right. The use of AI, however, would not necessarily prevent a natural/physical person from being recognised as the inventor where AI is used as a tool in the inventive process, much in the same way that the use of computer-aided design software such as AutoCAD or other technical tools does not, by itself, displace human inventorship. Mexican law does not establish a specific threshold of human contribution for AI-assisted inventions.

The availability or use of AI does not, by itself, alter the standards for assessing inventive step or the person skilled in the art. The analysis remains focused on the state of the art and on whether the claimed invention would have been obvious to a person skilled in the relevant field. As AI becomes more widely available, its use may become increasingly relevant to this analysis, though its availability alone should not be enough to establish obviousness or lack of inventive step.

AI-generated disclosures and technical information may be relevant as prior art if this was made available to the public before the invention filing date. In any event, a patent may be declared invalid where, among other grounds, the subject matter is not considered an invention, the invention is not patentable, or it lacks novelty, inventive step or industrial applicability.

In Mexico, AI-generated designs and product appearances may be protected under the ordinary IP regimes, provided that the applicable requirements of the LFPPI are met.

Trade dress can also be protected – the LFPPI expressly recognises the combination of operational or visual elements such as size, design, colour, arrangement, labelling, packaging or ornamentation that, when combined, distinguishes goods or services in the marketplace. Other elements such as graphical interfaces, icons, avatars and virtual goods may likewise be protected. Copyright protection, however, would not extend to material generated by AI, in light of the Supreme Court’s recent position requiring a human author.

Under Mexican law, names, logos, slogans, sounds, motion marks, and other signs generated with the assistance of AI may be registered as trade marks, provided they meet the requirements established by the LFPPI. The law does not make registration conditional upon the sign having been created by AI; the key factor is that it can distinguish products or services.

Therefore, the analysis must be conducted under the general trade mark regime, particularly regarding distinctiveness. This includes verifying that the sign does not violate any legal prohibitions, is not generic or merely descriptive (or otherwise lacks distinctiveness), and does not fall under any of the grounds for registration refusal set forth in the LFPPI.

Ownership of an AI-assisted trade mark belongs to the natural or legal person who applies for and obtains the registration, not to the AI. The LFPPI expressly allows any natural or legal person to obtain exclusive rights to a trade mark through registration.

AI cannot be an author or holder of copyright. With regard to attribution, or the right of paternity, a person cannot present as their own a work that, although generated with the assistance of AI, reproduces, transforms or adapts protected elements of a pre-existing work created by another author. In such a case, the rights of the author of the pre-existing work may be affected, and the person using AI cannot claim as an original creation something that incorporates protected elements belonging to another author.

With regard to the right of integrity, the use of AI to modify, transform or intervene in a pre-existing work, when the modification made through AI constitutes a distortion, mutilation or alteration of the work that affects the author’s interests or reputation, may entitle the author to exercise the rights available to them against such modification.

Finally, in cases of false attribution, if a person presents as their own a work that was actually created by another human being, this may infringe that person’s right to recognition of authorship.

In Mexico, with respect to rights concerning a person’s name, image, physical appearance, voice, performance, personality or digital replica, protection against their generation or use through AI derives primarily from the personal image right and personality rights, as well as (depending on the circumstances) from rights relating to privacy, identity, reputation, copyright and related rights.

Personality rights are constitutionally protected in Mexico. These rights derive from the principle of human dignity. Depending on the circumstances, they include the right to a name, personal image, identity, honour and privacy. In particular, the personal image right has been recognised as a human, fundamental and personal right, which allows individuals to decide how their image is represented and how and for what purposes third parties may use it. It has also been recognised that an image may have economic value and may be subject to commercial exploitation. In this regard, the protection of these elements may be understood as protection derived from human and personality rights, which may be complemented by different legal regimes depending on the circumstances of the case.

Of particular relevance to AI is Article 87 of the LFDA, which provides that the image – including the voice – of performing artists, as well as their characters, may only be used or published with their express consent. This provision expressly establishes that such protection extends to results generated by AI systems or any other technology.

In Mexico, there is no formal distinction between an AI-related IP dispute and an ordinary IP dispute, nor is there a specific court or authority with exclusive jurisdiction over AI-related matters. Instead, Mexican authorities address these issues within their respective areas of competence.

The main authorities dealing with IP matters are the IMPI and the INDAUTOR, which would therefore handle IP matters involving AI in the same manner as other IP disputes. Any challenge to their decisions would likewise follow the ordinary procedures applicable to IP matters. For example, administrative decisions may be challenged before the Federal Court of Administrative Affairs and, subsequently, before the Federal Circuit Courts. This was the procedural route followed in a case that ultimately reached the Supreme Court and resulted in its recent ruling that works generated exclusively by AI cannot be protected by copyright because authorship must be attributed to a natural person.

Under Mexican law, there is no specific evidentiary procedure for AI-related information such as training data, prompts, logs, model weights or tool-call histories. Such evidence may be obtained through the ordinary means available in contentious proceedings, including documentary and expert evidence. Where the relevant conduct can be reproduced, the interaction with the AI system may be replicated and documented, including through an attestation of facts before a notary public, supported where appropriate by an IT expert report or opinion.

Confidential technical information, trade secrets and other sensitive material may be submitted on a confidential basis and receive specific treatment by the court or authority. This may include restricting access to the information while allowing it to be considered as evidence in the proceedings.

In Mexico, the provisional measures available in AI and IP disputes are the same measures available in other IP proceedings. Under Article 344 of the LFPPI, the IMPI may order, among other measures:

• the withdrawal or seizure of infringing goods;
• the suspension or cessation of infringing acts; and
• the blocking, removal or disabling of access to infringing content in digital environments.

The latter may be particularly relevant in AI-related disputes, for example, where infringing content is generated or made available through an AI system.

In a final decision, the authority may also order the infringing activity to cease. Depending on the circumstances, this could extend to measures such as:

• stopping the use or training of a model;
• removing infringing material from a dataset;
• disabling access to certain outputs; or
• taking other measures necessary to prevent the continuation of the infringement.

Mexican law provides for monetary compensation for IP infringements, which shall in no case be less than 40% of the legitimate value indicator selected by the affected right-holder. Under the LFPPI, this indicator may be based on:

• the value of the infringed products or services, calculated by reference to their market price or suggested retail price;
• the profits that the right-holder would have ceased to receive as a result of the infringement;
• the profits obtained by the infringer as a result of the infringement; or
• the price that the infringer would have paid to the right-holder for a licence.

Mexican courts have also developed the concept of punitive damages as part of the broader principle of integral reparation, although its application depends on the circumstances of the particular case.

In Mexico, the written formality in licence agreements is mandatory. In addition, they must include an economic incentive and, if this is exclusive, it must be indicated as such; in absence of an economic agreement between the parties, the Mexican authorities are entitled to establish it.

They must also be temporal; in the absence of timing, they are considered as being for a five-year term. They also need to include proportional participation of authors in revenue or a fixed, determined remuneration – this is an unwaivable right as per the LFDA. Lastly, if the licence agreement is certified by a notary public and recorded in the Public Registry of INDAUTOR, it will be enforceable.

Moral rights cannot be waived, so no attribution clause can be drafted as a waiver. If any derived work could be created, this must have the authorisation of the author of the first work in order for the latter to be commercialised.

All prerogatives in the licence are construed restrictively; anything not named is not granted.

One of the key clauses in a licence agreement of this type is evidencing the chain of title of the work to be licensed. An express certificate of recordal of the work in the name of the licensor recognising their authorship, or previous agreements whereby they had assigned or delivered the work on a work-for-hire basis, will be valuable for licensees to further rely on the commercialisation of the project in question.

In January 2026, Mexico introduced the “Chapultepec Principles”, an ethical framework establishing guidelines for the responsible implementation of AI technologies in alignment with international best practices. It is important to note, however, that these principles are non-binding and serve purely as guidance rather than enforceable regulation.

Moreover, the Mexican government, through its Agency for Digital Transformation and Telecommunications (ATDT), created the “Artificial Intelligence Factory” project, which involves the creation of a supercomputer named “Coatlicue” that will have the capacity to execute 314 trillion operations per second, making it the most powerful in Latin America.

Furthermore, the recent amendments to the LFPPI and the LFDA evidence the attempt to incorporate “artificial intelligence” into the Mexican legal framework; nevertheless, the country still lacks a formal definition of the term or specific provisions governing its scope.

In addition, on 17 June 2026, the Mexican Congress introduced a bill proposing the enactment of the General Law for the Regulation and Ethical Use of Artificial Intelligence in Mexico. Although the initiative is still under legislative review, it represents Mexico’s first comprehensive attempt to establish a dedicated regulatory framework for AI. Among its most notable provisions are:

• the creation of a specialised supervisory authority;
• the adoption of a risk-based approach for AI systems;
• the establishment of specific obligations for developers and deployers;
• mandatory impact assessments for high-risk systems;
• transparency requirements; and
• the recognition of fundamental rights as the guiding principle for the development and deployment of AI technologies.

According to the Organisation for Economic Co-operation and Development (OECD), Mexico is one of the members that does not have a public financial scheme dedicated to AI development. While there have been improvements in the public use of AI within the government agencies in the country, as well as research and adoption of AI in the private and public spheres, the lack of a uniform legal framework does not incentivise private investment.

Nonetheless, the recent designation of the IMPI as the International Searching Authority (ISA) and International Preliminary Examining Authority (IPEA) under the Patent Cooperation Treaty (PCT) by the World Intellectual Property Organization (WIPO), as well as the regional European AI legal framework that imposes transparency obligations for flagging deepfakes and AI-generated content under the Code of Practice on marking and labelling of AI-generated content (as a transparency obligation of the AI Act), might provide incentive for the legal changes expected in the country.