Artificial Intelligence 2026 Comparisons

Any views expressed in this publication are strictly those of the authors and should not be attributed in any way to White & Case LLP.

To date, there is no AI-specific law in Mexico. AI is indirectly regulated through the application of existing general laws, depending on how AI systems are used.

In cases where AI systems process personal data, the Mexican data protection law (Ley Federal de Protección de Datos Personales en Posesión de los Particulares, or LFPDPPP) applies, regardless of the technology used, which directly requires compliance with data protection principles and data subject rights. In practice, predictive AI systems, including those used for digital onboarding, profiling, risk assessment and credit scoring, will often fall within this regime.

Depending on the context, AI systems may also be subject to sector-specific regulation. In financial services, the use of AI by financial entities for risk assessment, fraud detection or credit scoring, among other processes, results in the application of financial regulatory laws. The use of AI in interactions with end-users triggers obligations under consumer protection laws (Ley Federal de Protección al Consumidor, or LFPC, and Ley de Protección y Defensa al Usuario de Servicios Financieros, the “Consumer Laws”), particularly in relation to transparency, misleading practices and liability for defective or unsafe products and services. Employment law may also be relevant where AI is used in hiring or workplace monitoring, for example where AI tools are involved in selection processes, performance evaluation or monitoring of employees’ activities.

Generative AI involves distinct considerations and is still developing under IP law. Recent case law from the Mexican Supreme Court (Suprema Corte de Justicia de la Nación, or SCJN) (Amparo Directo 6/2025, paras. 52–54, 58) held that copyright protection in Mexico is reserved for works created by human authors and therefore denied protection to content generated by an AI system, reasoning that AI lacks the individuality and originality inherent to human creativity, experience, perception and feelings (see 4.1 Precedent-Setting Judicial Decisions). In addition, the April 2026 reform to the Industrial Property Law (Ley Federal de Protección a la Propiedad Industrial, or IPL) extended existing infringement provisions to conduct carried out through AI systems.

Finally, contract law will govern the allocation of risk between parties involved in the development and use of AI systems. Civil and product liability principles may apply where AI systems cause harm, raising questions of fault, causation and foreseeability, particularly in contexts involving increased autonomy of such systems. Criminal law may apply in cases involving misuse of AI, such as fraud, forgery or money laundering.

According to Statista (March 2026), Mexico’s AI market is valued at around USD2.78 billion in 2026, representing growth of roughly 41% compared to 2025, and is expected to reach approximately USD12 billion by 2032.

Key industry applications across different AI architectures include (Statista, March 2026; all market sizes and growth rates are rounded, year‑on‑year figures for 2026 versus 2025):

• Traditional machine learning (ML) (ie, use of algorithms to enable computer systems to learn from data) is the largest AI segment in Mexico, at USD1.12 billion (48% growth);
• computer vision (ie, applications that enable computers to interpret and understand digital images and video data) reached USD352 million (23% growth);
• natural language processing (NLP) (ie, applications that enable computers to understand, interpret, and generate human language) reached USD659 million (42% growth);
• generative AI accounted for USD355 million, approximately 12.8% of Mexico’s AI market, and is forecast to exceed USD1 billion by 2032;
• agentic AI systems such as autonomous and sensor technology (ie, machines and systems that operate independently by using sensors, AI and ML to respond to changes in their environment), valued at USD288 million; and
• AI robotics (ie, combination of AI, ML, and engineering to create intelligent machines that can perform tasks autonomously), valued at USD362 million and showing the highest growth rate at 61%. According to the AI Index Report 2026 (Stanford HAI), Mexico installed 5,600 industrial robots in 2024, with installations growing 4% between 2023 and 2024.

According to Endeavor (2024), key AI industry applications include business intelligence and analytics, ML, robotic process automation (RPA) and robotics, all primarily used to optimise and automate business processes. ML, NLP, and predictive analytics are among the principal AI applications used by Mexican companies, which is consistent with the rapid expansion of these segments reflected in Statista’s 2026 data. Endeavor further reported that investment in adaptive AI and chatbots/NLP reflects the importance of improving user experience and personalised interaction, while interest in RPA and robotics underscores the demand for efficient and automated solutions across industrial sectors.

Mexican authorities – particularly at the federal level – have implemented measures to facilitate the adoption and advancement of AI for both industrial and governmental purposes, focusing on attracting strategic investment, strengthening institutional capacity and building tools to support and automate projects. According to the Ministry of Economy, Mexico has supported a USD1 billion investment plan by Flex for AI data centre construction between 2026 and 2028 in several cities, a strategy expected to create more than 5,000 jobs and position Mexico as a strategic location for technology and data-centre operations. According to the AI Index Report 2026, Mexico had 173 data centres as of 2025.

On the institutional side, November 2024 amendments to the Organic Law of the Federal Public Administration (Ley Orgánica de la Administración Pública Federal) created the Ministry of Science, Humanities, Technology and Innovation (Secretaría de Ciencia, Humanidades, Tecnología e Innovación, or SECIHTI) as a dedicated ministry responsible for national science, technology and innovation policy, empowered to propose fiscal incentives, financial support and subsidies for strategic projects.

The Federal Government has also announced a public centre for Training in AI, targeting approximately 25,000 graduates per year across 20 specialisations in AI and related fields, which would be the largest public AI training institution on the continent.

As mentioned in 1.1 General Legal Background, Mexico has not enacted a dedicated AI law, and AI-related activities are currently addressed through existing general laws. According to Pedro Salazar, a constitutional law scholar and researcher at the Instituto de Investigaciones Jurídicas (IIJ) at the Universidad Nacional Autónoma de México, approximately 200 AI-related initiatives have been presented in recent years, around 70 of which remain pending. He has noted that, as of April 2026, only one had advanced significantly: a bill to amend the Federal Copyright Law (Ley Federal del Derecho de Autor, or LFDA) and the Federal Labour Law (Ley Federal del Trabajo, or LFT) to strengthen protection of voice actors and dubbing professionals against unauthorised use of their voices through AI systems. This was approved by the Chamber of Deputies on 7 April 2026, and by the Senate on 15 April 2026.

On the executive side, the same amendments discussed in 2.2 Involvement of Governments in AI Innovation established the Agency for Digital Transformation and Telecommunications (Agencia de Transformación Digital y Telecomunicaciones, or ATDT), which leads federal digital transformation alongside SECIHTI.

At the parliamentary level, the Mexican Senate established a specific Senate AI Commission (Comisión de Análisis, Seguimiento y Evaluación sobre la Aplicación y Desarrollo de la Inteligencia Artificial en México). According to its mandate, the commission’s role is to analyse the impact of AI in different sectors, promote its equitable, widespread and accessible deployment and work towards a comprehensive regulatory framework.

The federal judiciary has also started to define parameters for the use of AI in proceedings. In January 2026, two binding precedents (jurisprudencia) from a federal court held that AI tools in judicial processes must comply with minimum standards on proportionality and safety, personal data protection, transparency, explainability and human supervision, and that judges must disclose AI use and treat it as an auxiliary tool rather than a substitute for deliberation. The court expressly referred to EU ethical guidelines on AI, the EU AI Act and UNESCO’s recommendation on the ethics of AI, and upheld the use of AI tools to calculate guarantees in amparo proceedings as a valid mechanism to improve consistency and efficiency while preserving the core of judicial decision-making.

At the sub‑national level, some states have begun to add AI‑related language to existing state legislation, for example creating aggravating circumstances when crimes are committed using AI as a tool and introducing statutory definitions of AI (such as the Criminal Code of Quintana Roo). These are isolated, uncoordinated measures rather than part of a national AI framework.

As mentioned, no AI-specific legislation has yet been enacted in Mexico.

In January 2026, SECIHTI and the ATDT issued the non-binding “Chapultepec Principles“, an ethical and good-practices declaration intended to guide public policies, regulation and institutional instruments throughout the AI lifecycle, with an advisory character rather than creating legally enforceable obligations.

The principles set out guidelines for ethical and responsible AI. They state that:

• AI should expand, not reduce, human rights;
• all AI-assisted decisions must have identifiable human responsibility;
• decisions which cannot be explained should not be automated;
• AI use should be transparent;
• its use should be governed by a combination of citizens, academia, communities, social organisations and the private sector;
• AI should be used to expand rights and reduce inequalities; and
• An assessment of potential social, territorial and environmental impacts should be undertaken before deploying AI in public policy and services.

Further, the principles treat AI as part of a national development strategy. They highlight the need to align AI infrastructure with Mexico’s long-term priorities, to strengthen education and interdisciplinary research, to reflect the country’s cultural and linguistic diversity in AI systems, and to treat data as a public good managed with safeguards for quality, representativeness, privacy and cybersecurity.

There is no applicable information in this jurisdiction.

There is no applicable information in this jurisdiction.

From a data protection perspective, Mexico has no specific AI data processing provisions, but the LFPDPPP applies fully where AI involves or relies on personal data. It requires controllers to define and disclose purposes in the privacy notice, distinguishing those that require the consent of the data subject (ie, the person whose data is being processed), and to limit processing to those purposes, unless new consent is obtained for any additional purpose. As a general rule, processing requires the data subject’s consent, which may be express or tacit, with express consent required for financial data, and express written consent required for sensitive data, subject to limited statutory exceptions (eg, where processing is required by law, arises from a legal relationship or follows prior dissociation). Controllers must also ensure that personal data is accurate, complete, correct and up to date for the purposes for which it was collected, and block and delete it once it is no longer necessary for those purposes, subject to legal retention periods.

See 17.2 AI Deployment and Data Subject Rights for further details on automated decision‑making and data subjects’ rights, 17.1 AI Training and Data Protection for data principles and anonymisation and 17.3 AI Data Governance and Cross-Border Transfers for data governance and security obligations.

Mexican law does not contain specific text and data mining (TDM) or web scraping‑related provisions. From a data protection perspective, the lawfulness of scraping‑based training depends on whether the data qualifies as personal data, whether it comes from publicly available sources and whether the controller has complied with notice and consent requirements under the LFPDPPP.

For a discussion of recent IP developments, see 4.1 Precedent-Setting Judicial Decisions and 16. Intellectual Property.

Key AI-specific legislative proposals currently pending are as follows:

• A constitutional reform bill. This bill proposes amending Article 73, Section XVII of the Constitution to empower Congress to legislate on AI and enact the General AI Law. If approved, Congress would be required to harmonise secondary legislation within 90 days and enact a General AI Law within 180 days.
• A bill to enact the National Law Regulating the Use of AI. This bill proposes a national AI framework built around a National AI System, a National AI Centre and a National Council. It introduces a three-tier risk classification (unacceptable, high and limited risk), mandatory labelling of AI-generated content, restrictions on public sector AI use and guiding principles, including transparency, privacy by design and non-discrimination.
• A bill to enact the National AI Law. This bill proposes a national framework covering AI development, deployment and administrative sanctions, with emphasis on human rights protection, gender perspective, non-discrimination, neuro-rights and the creation of a dedicated regulatory agency under SECIHTI.

In April 2026, the Senate AI Commission announced it was developing both a legislative bill and a national AI strategy intended to align government, private sector and academic efforts. No text has been published.

The following breaks down a key judicial decision pertaining to generative AI and IP rights:

• The court considering the decision was the SCJN;
• the matter on the docket was Amparo Directo 6/2025, considered on 2 July 2025;
• the questions under consideration were twofold: whether artwork created by generative AI should be protected under the LFDA and whether moral rights can be recognised in favour of a generative AI system;
• the court adjudicated: AI systems cannot satisfy the authorship requirements of the LFDA, as AI lacks the individuality, originality, creativity, experience, perception of surroundings and feelings intrinsic to human beings; that moral rights – being non-waivable and inalienable – are personal to human authors and cannot vest in artificial entities; and
• the court rejected constitutional challenges to the applicable provisions and dismissed the relevance of foreign precedents favouring AI authorship on territoriality grounds.

The regulatory agencies playing a leading role in AI include:

• the ATDT: responsible for federal government technology policy, data and information analysis standards and coordination of government technology procurement;
• the SECIHTI: responsible for national science, technology and innovation policy, ensuring public access to the benefits of technological development and fostering collaboration between government, academia and the private sector;
• the Mexican Institute of Copyright (Instituto Nacional del Derecho de Autor): responsible for the protection of works of art, computer programs, databases and other copyrightable content;
• the Mexican Institute of Industrial Property (Instituto Mexicano de la Propiedad Industrial, or IMPI): responsible for granting and enforcing patents, trademarks and other industrial property rights;
• the Secretariat of Anticorruption and Good Governance (Secretaría Anticorrupción y Buen Gobierno, or SABG): responsible for overseeing, interpreting, enforcing data protection law and issuing guidance on privacy and information security;
• the National Antitrust Commission (Comisión Nacional Antimonopolio, or CNA): responsible for guaranteeing free competition, investigating and sanctioning monopolistic practices and unlawful concentrations and issuing measures to remove barriers to competition;
• the Office of the Attorney General of the Republic (Fiscalía General de la República): the autonomous body responsible for the prosecution of federal crimes;
• the Mexican Banking and Securities Commission (Comisión Nacional Bancaria y de Valores, or CNBV),: responsible for authorising the contracting of technology services by certain financial entities;
• the Mexican Insurance and Bonds Commission (Comisión Nacional de Seguros y Fianzas), responsible for prudential and regulatory oversight of the insurance sector;
• the Ministry of Health (Secretaría de Salud): responsible for national public health policy and coordination of the National Health System; and
• the Federal Commission for the Protection against Sanitary Risks (Comisión Federal para la Protección contra Riesgos Sanitarios, or COFEPRIS): responsible for authorising and registering medical drugs and health products.

See 3.3 Jurisdictional Directives.

The authors have not identified any notable enforcement actions targeting AI systems to date. However, the April 2026 reform to the IPL expressly states that existing administrative infringements (eg, unfair competition, misuse of trademarks, patents or trade secrets) are also sanctionable when committed through the use of AI and may be investigated ex officio or at the request of an interested party by the IMPI.

Mexico has not issued any AI-specific standard or norm under the Mexican Quality Infrastructure Law (Ley de Infraestructura de la Calidad), which regulates standardisation activities in Mexico. International standards such as ISO/IEC 42001 (Artificial Intelligence Management System) are nonetheless used as reference frameworks by both the public and private sector. Mexico is an observer member of the ISO committee on AI and has created a National Quality Infrastructure System under this law to coordinate technical regulation and promote the use of standards.

At the skills and workforce level, the Council for Standardisation and Certification of Labour Competencies (Consejo Nacional de Normalización y Certificación de Competencias Laborales) issues official certifications of workers’ skills. It has begun adopting AI-related competency standards, including on the implementation of generative AI technologies in logistics supply chains, the development of educational materials with generative AI tools and the basic use of generative AI tools for creating digital content. Note that these competency standards do not directly regulate AI systems or impose binding obligations on companies.

ISO standards (such as ISO/IEC 42001) are often used as reference frameworks by both the private and public sector. At this time, it is still too early to predict whether conflicts with jurisdictional law will arise.

The Mexican government has begun to use AI in areas such as tax administration and judicial processes. The Economic Commission for Latin America and the Caribbean (ECLAC, 2026) reported that the Mexican Tax Administration Service (Servicio de Administración Tributaria, or SAT) has intensively used big data and AI technologies since 2017 to analyse electronic invoices and identify tax evasion schemes, despite the absence of a national AI strategy. SAT’s 2024 Master Plan described the deployment of graph analytics and ML models to classify high-risk taxpayers, detect complex evasion and avoidance networks and identify inconsistencies in e-invoices across several high-risk sectors.

In the judicial branch, a federal court has issued binding precedents on the use of AI tools in judicial processes, requiring minimum standards on proportionality and safety, personal data protection, transparency and explainability, human supervision and the preservation of decision-making. The court has upheld their use as auxiliary tools for numerical calculations in amparo proceedings (see 3.1 General Approach to AI-Specific Legislation).

In addition to the judiciary’s own use of AI tools in court proceedings (see 3.1 General Approach to AI-Specific Legislation), the following decisions summarise how AI has been conceptualised in statutory and criminal contexts:

Case 1

• The court considering the case was the SCJN;
• the matter on the docket was 66/2024 (unconstitutional action; acción de inconstitucionalidad);
• the issue under consideration was assessing whether the Sinaloa Criminal Code’s definition of AI for deepfake‑enabled sexual‑intimacy offences violates the principle of legality (taxatividad) by not being technically precise; and
• in its decision, the SCJN upheld this functional, context‑specific definition as sufficiently clear (see also 12.3 Deepfakes and Synthetic Media).

Case 2

• The court considering the case was the SCJN;
• the matter on the docket was 80/2024 (unconstitutional action; acción de inconstitucionalidad);
• the issue under consideration was assessing whether a statutory reference to “use of artificial intelligence” in a provision protecting children from digital violence violated the principles of legality and legal certainty; and
• in its decision, the SCJN ruled that laws are not dictionaries, and a general reference to AI is compatible with legality in a rapidly evolving technological context, especially when seeking to protect children from violence facilitated by digital tools.

Case 3

• The court considering the case was the SCJN;
• the matter on the docket was 104/2024 (unconstitutional action; acción de inconstitucionalidad);
• the issue under consideration was assessing whether the definition of AI as a general aggravating circumstance in the Criminal Code of Quintana Roo violates the principles of legality and legal certainty (taxatividad); and
• in its decision, the SCJN's draft judgement proposed striking down the provision, but the court failed to reach the required majority of votes, so the provision remains in force.

The use of AI in national security and defence is now expressly addressed in the Law on the National System of Investigation and Intelligence in Public Security, in force since July 2025. This statute creates a National System (Sistema Nacional de Investigación e Inteligencia en Materia de Seguridad Pública), coordinated by the Ministry of Security and Citizen Protection (Secretaría de Seguridad y Protección Ciudadana) and operated by the National Intelligence Centre (Centro Nacional de Inteligencia), which manages a Central Intelligence Platform (Plataforma Central de Inteligencia) interconnecting multiple federal, state and private databases and systems.

The law sets forth that data and intelligence products generated through the system and the platform must be used for criminal analysis to prevent and prosecute crimes, namely particularly high‑impact offences committed by organised crime, and to support criminal investigations. For processing and analysis, the law authorises the use of systems and programs that receive, convert, organise, classify and interrelate all types of data, expressly including automation programs and AI tools, to generate security‑intelligence strategies, actions and data protection frameworks.

In the authors’ view, key legal challenges around generative AI in Mexico result from the need to fit these systems and their outputs and uses into existing frameworks in the absence of AI-specific legislation. There is also the practical difficulty of determining who is responsible for which part of the AI lifecycle, as the process can span across several jurisdictions and through opaque technical arrangements. In this context:

• There is currently no dedicated regulatory approach to foundation models or general‑purpose AI. No guidelines, registration, notification, classification or pre‑market requirements apply specifically to foundation model developers, deployers or providers;
• neither the LFDA nor the IPL expressly recognise a fair use doctrine or an equivalent open-ended exception: training on protected works without authorisation carries infringement risk, and no statutory exception applies specifically to AI training. The SCJN has confirmed that AI-generated works are not protectable under the LFDA, holding that authorship is intrinsic to the human condition (see 4.1 Precedent-Setting Judicial Decisions and 16. Intellectual Property);
• training and deploying LLMs on datasets containing personal data triggers the application of the LFPDPPP. AI use must comply with data protection principles, obtain consent as required, keep the use within the purposes consented by the data subject under the privacy notice, obtain additional consent when new processing purposes arise and respect the rules on automated decision‑making where applicable. (see 17. Data Protection);
• defamatory, infringing or otherwise harmful outputs are addressed through existing civil, consumer and IP liability rules (see 10. Liability for AI). In practice, allocating responsibility along these chains may raise evidentiary and jurisdictional challenges, particularly given the technical opacity of some systems and the potential for cross-border processing; and
• there is no general, AI‑specific labelling, watermarking or explicit disclosure regime for synthetic media or AI‑generated content under Mexican law. Any transparency duties are indirect, arising from data protection, consumer protection and advertising rules.

AI-assisted tools for legal research, drafting and contract review are increasingly used in Mexico, but no AI-specific professional conduct rules exist. General duties of competence, confidentiality and diligence govern their use, raising considerations around the accuracy of AI-generated outputs, the handling of client data and unauthorised practice boundaries.

The theory of liability for personal injury or commercial harm derives fundamentally from Article 1910 of the Federal Civil Code (FCC) and its state law equivalents, which provide that any person who acts unlawfully or contrary to good morals (buenas costumbres) and causes damage shall repair it. In the AI context, this encompasses liability for negligent design, deployment or supervision of AI systems. Furthermore, Article 1913 of the FCC provides that anyone who causes damage via devices, instruments or substances that are dangerous per se, by virtue of speed, flammability, electricity or any analogous cause, shall pay for the damage they cause, even if they did not act unlawfully.

Additionally, Article 1916 of the FCC establishes liability for moral damage (daño moral), understood as harm to a person’s feelings, honour, reputation, private life or physical appearance and image. Section IV covers acts that offend honour, attack private life or violate a person’s image, so AI‑generated outputs such as defamatory content, synthetic media or unauthorised uses of a person’s image may give rise to a moral damage claim independently of any economic loss. Indemnification is set by the judge, considering the rights affected, the degree of responsibility and the economic circumstances of both parties.

The opacity of AI systems and the potential involvement of multiple parties across development and deployment chains may present practical challenges in establishing causation and attributing fault in claims under the frameworks described above. These questions remain largely untested under Mexican law.

Where relevant insurance is available, it may help to cover damages under these liability frameworks.

For risk allocation, see 13.2 AI Supply Chain Accountability and Due Diligence.

AI‑related liability is currently governed by existing legal frameworks. For example, the IPL now expressly provides that administrative infringements (eg, unfair competition, misuse of trademarks or trade secrets) apply equally when the conduct is carried out through the use of AI.

The bills mentioned in 3.7 Proposed AI-Specific Legislation and Regulations generally do not contain detailed provisions on liability allocation across the AI supply chain. An exception is the bill to enact a National AI Law, which proposes an administrative infringement regime for certain AI‑related conduct, such as:

• deploying uncertified AI systems;
• using high‑risk systems without authorisation;
• using AI in ways that harms life, health, privacy or cognitive integrity; or
• using AI for fraud, extortion, identity theft or criminal processing of sensitive personal data.

Sanctions would range from warnings and fines to suspension or withdrawal of systems or services and closure of platforms or services. It is unclear whether this bill will advance, given the number of AI‑related initiatives that have not advanced in Congress (also see 3.1 General Approach to AI-Specific Legislation).

Agentic AI systems are governed by general principles of contract law, civil liability, privacy and consumer protection, applied on a case-by-case basis depending on how the system is used, the domain and the harm caused.

Where agentic AI systems process personal data or make automated decisions affecting individuals, the LFPDPPP applies, including its rules on consent, purpose limitation and automated decision-making (see 17.2 AI Deployment and Data Subject Rights). There are no statutory human oversight, logging, auditability or explainability requirements for agentic systems. In domains such as health and financial services, sector-specific regulation applies (see 15.2 Financial Services and 15.3 Healthcare).

The general civil framework described in 10.1 General Theories of Liability applies to harm caused by autonomous AI systems, with Articles 1910 and 1913 of the FCC being the primary bases for fault-based and strict liability claims respectively. Liability allocation across the AI supply chain is primarily addressed contractually; for a full discussion, see 13.1 AI Procurement Standards and Contracting.

Autonomous AI systems pose special challenges for causation and evidence. The causal chain between design decisions, deployment choices and potential harm may involve multiple parties across different jurisdictions, and the opacity of the systems can make it difficult to identify the point of failure or the responsible actor. These questions remain largely untested under Mexican law.

According to the Organisation for Economic Cooperation and Development, algorithmic bias in AI can be characterised as “systematically better or lower AI algorithmic performance leading to some harm against one person or sub-population group”.

Consumer areas in which bias can create significant risk are those that could be associated with an individual’s access to opportunities, well-being or physical integrity. No fairness testing, auditing or bias mitigation obligations are imposed by Mexican law on private entities that develop and deploy AI systems. In the absence of binding requirements, best practice references include ISO/IEC 42001 and the Chapultepec Principles‘ call for prior impact assessments.

Potential liability for companies due to unmanaged algorithmic bias that causes discrimination can be categorised as follows:

• Civil liability for discriminatory AI outputs may arise under the FCC’s general civil liability framework, including Article 1916 where the discrimination causes harm to a person’s honour or reputation, and Article 1913, which imposes strict liability on anyone who causes damage through inherently dangerous devices, instruments or substances, even where no unlawful conduct is established (see 10.1 General Theories of Liability);
• administrative liability under the Federal Law to Prevent and Eliminate Discrimination may result in monetary compensation to the affected party, the posting of notices at the company’s premises identifying it as a place where discrimination occurred and publication of the resolution in official and media outlets. These measures are imposed by the National Council to Prevent Discrimination and may carry significant reputational consequences for the legal entity involved; and
• criminal liability is established under the Federal Criminal Code and state criminal codes. It is important to note that legal entities – as well as individuals – can be criminally liable under Article 421 of the National Code of Criminal Procedure.

Biometric technologies and emotion recognition practices are regulated on a sector-specific basis, with the LFPDPPP providing the general data protection baseline. The LFPDPPP does not expressly list biometric data among the categories of sensitive personal data. However, personal data qualifies as sensitive where it:

• Affects the most intimate sphere of the data subject;
• could give rise to discrimination if improperly used; or
• could entail a serious risk for the data subject. Statutory examples include: racial or ethnic origin, present or future health status, genetic information, religious, philosophical or moral beliefs, political opinions and sexual preference, but this list is illustrative rather than exhaustive.

In the PANAUT decision (acción de inconstitucionalidad 82/2021), the SCJN recognised that biometric data processed to uniquely identify individuals may qualify as sensitive personal data, and as such, requires enhanced protection and safeguards defined in law, including clear rules on purposes, retention and access. Under non‑binding guidance from the SABG, this classification is further justified by the unique and irreversible nature of biometric data, which means that, in case of a breach, the potential harm can be long-lasting or permanent. Where biometric data qualifies as sensitive, the enhanced protection regime under the LFPDPPP, as further clarified by the SABG’s guidance, includes, as applicable:

• Express written consent;
• reinforced administrative, technical and physical security measures;
• risk analyses or data protection impact assessments, where appropriate; and
• strict compliance with the principles of proportionality, purpose limitation and data minimisation.

Under Article 58, Sections XIII and XV and Article 59, Section III of the LFPDPPP, collecting or transferring personal data without the required express consent, as well as collecting personal data in a deceptive or fraudulent manner, may be subject to fines up to USD2.1 million, which is particularly relevant in the AI context for training or operational data. Articles 62 to 64 set forth criminal sanctions of up to five years’ imprisonment for misuse of personal data, with penalties doubled where sensitive personal data is involved.

An example of industry-specific considerations concerns banks, which are permitted to use biometrics for authentication purposes under Article 1, Section LXVI (d) of the General Rules Applicable to Banks (Circular Única de Bancos, or CUB), subject to strong cybersecurity measures. Compliance with KYC or AML obligations under financial regulation does not alter the sensitive nature of the data, its legal classification, or the enhanced obligations that apply to its processing. While the exceptions to consent under Article 9 of the LFPDPPP – such as processing necessary information to comply with a legal obligation or an existing legal relationship – may be available in this context, all applicable data protection principles and reinforced security measures continue to apply.

For reference, the EU AI Act defines “deep fake” as “AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful”.

Where deepfakes or synthetic media are created or disseminated using biometric data (eg, facial images or voice samples) that allow unique identification of individuals, such processing is subject to the LFPDPPP and may qualify as sensitive personal data under the criteria described in 12.2 Biometric Technologies and Emotion Recognition. In line with non-binding guidance from the SABG on biometric processing, misuse, loss or unauthorised access to such data can cause serious harm to the data subject’s legal and economic interests, including identity theft, fraud or discrimination, which supports the application of a reinforced protection standard in deepfake contexts.

At the federal level, there is no AI‑specific statute on deepfakes or synthetic media, so harmful uses are generally tackled through existing offences such as cybercrime, fraud and identity theft.

The SCJN has recognised the right to one’s own image as a derecho personalísimo of constitutional rank (P. LXVII/2009), granting its holder the exclusive faculty to decide everything relating to the capture, reproduction or publication of their image; deepfakes produced without consent directly engage this right. Unauthorised use of a person’s image in synthetic media may constitute an infracción en materia de comercio under Article 231 of the LFDA, a procedure before IMPI that must be completed before seeking civil indemnification, and may additionally give rise to a moral damage (daño moral) claim under Article 1916 of the FCC, independently of any economic loss (see 10.1 General Theories of Liability).

States have also begun to legislate specifically on AI‑enabled image and media manipulation.

• In Sinaloa, the Criminal Code (article 185 Bis C) includes within the offence of violation of sexual intimacy the creation and dissemination, without consent, of sexually intimate images, audio or video manipulated using AI to generate false but realistic content. For that purpose, AI is defined as applications, programs or technology that analyse photographs, audio or video and offer automatic adjustments to alter or modify them. The SCJN upheld this context‑specific definition of AI and the provision’s constitutionality in acción de inconstitucionalidad 66/2024, finding that, in this context, a functional, non‑technical definition of AI is sufficiently clear to satisfy the principle of legality (taxatividad), notwithstanding the absence of a single, universal definition of AI (see 7.2 Judicial Decisions).
• In Quintana Roo, the Criminal Code now provides an aggravating factor where any offence is committed using AI as a tool or means, which increases the penalty by up to one half.

There is no AI‑specific transparency or disclosure regime in Mexico for chatbots, AI‑generated content or foundation models.

However, when AI systems process personal data or support decisions affecting individuals, they are subject to the data protection principles under the LFPDPPP, including the information principle, which requires controllers to inform data subjects of the existence and main characteristics of the processing through the privacy notice (see 3.6 Data, Information or Content Laws, 17.1 AI Training and Data Protection and 17.2 AI Deployment and Data Subject Rights).

While non-binding, the Chapultepec Principles (see 3.3 Jurisdictional Directives) address explainability, establishing that decisions which cannot be explained should not be automated and calling for transparency about AI use. In consumer contexts, AI‑driven interactions, recommendations, rankings and advertising must also comply with the consumer laws, which mandate truthful, verifiable, clear and not misleading or abusive information. Depending on the sector and type of processing, additional transparency or disclosure duties may arise under other general or sector‑specific law.

Mexico has not enacted AI-specific procurement or contracting standards; no statutory rules govern the allocation of rights and risks between the parties involved in developing, deploying or using AI systems. The contractual framework therefore becomes the main risk management instrument across the AI supply chain. Best practice considerations include:

• risk allocation and liability caps;
• service level agreements (SLAs);
• data rights and data processing agreements (DPAs);
• IP ownership, compliance warranties;
• audit rights; and
• exit and portability provisions.

Where developers are located outside Mexico, enforcement of contractual remedies may be challenging; insurance coverage and due diligence on counterparty financial standing are therefore also advisable.

In the absence of a statutory allocation framework (see 13.1 AI Procurement Standards and Contracting), accountability tends to fall on deployers, given their direct contractual relationship with customers and the fact that they are typically subject to local jurisdiction. Developers, by contrast, are often located in foreign jurisdictions where enforcement by Mexican authorities may be more complex.

In line with the notion of deployer used in the EU AI Act (ie, the natural or legal person, public authority, agency or other body using an AI system under its authority, except where the AI system is used in the course of a personal non-professional activity), deployers may face greater exposure when AI outputs cause harm to end-users and should conduct due diligence on upstream providers, covering model documentation, known limitations, data provenance and applicable certifications. Contractual mechanisms, including compliance warranties, indemnities for upstream failures, audit rights and cascading sub-processor obligations, are the main risk management tools available in the absence of statutory requirements.

Mexican law imposes no specific supply chain transparency or traceability obligations on AI systems. Where personal data is involved, obligations under the LFPDPPP are triggered (see 17.3 AI Data Governance and Cross-Border Transfers).

Mexican employment law applies regardless of the technology used in hiring or termination decisions. Under Article 439 LFT, when an employer reduces headcount due to the introduction of machinery or new work procedures, it must either reach an agreement with the affected employees or obtain prior Labour Tribunal authorisation. Affected employees are entitled to:

• Four months’ salary;
• 20 days per year of service (or the higher contractual amount); and
• the seniority premium (prima de antigüedad).

Where companies process personal data (including through AI tools) in recruitment or hiring processes (eg, background check data such as criminal records, administrative infraction records or financial history), the SABG has indicated in non-binding interpretive guidance that such data may qualify as sensitive personal data under the LFPDPPP, where their use may generate exclusion, discrimination, stigmatisation or disproportionate impacts on human dignity and restrict access to employment. In which case, the enhanced protection regime applies, including express written consent, purpose limitation and reinforced security measures.

For discrimination risks in automated decision-making, see 12.1 Algorithmic Bias.

It is important to be mindful of possible discriminatory practices (see 12.1 Algorithmic Bias) and to act in general compliance with the LFT. Automated employee performance evaluation, as well as monitoring employees’ work, could impact the indemnification of terminated employees under Article 439 of the LFT (see 14.1 Hiring and Termination Practices). Additionally, employees whose performance or behaviour is subject to fully automated evaluation without human intervention may have the right to object to or request cessation of such processing under the LFPDPPP (see 17.2 AI Deployment and Data Subject Rights).

Generally speaking, there are no clear regulations on the use of AI in digital platform companies such as car services and food delivery. For example, in Mexico City, the Regulations of the Mobility Act (Reglamento de la Ley de Movilidad de la Ciudad de México), which cover the digital platforms of car services, do not refer to AI or concepts like automation, ML, etc.

In practice, however, AI‑driven matching, pricing, rating and fraud‑detection systems used by these platforms are subject to the data protection framework discussed in sections 3.6 Data, Information or Content Laws, 17.1 AI Training and Data Protection and 17.2 AI Deployment and Data Subject Rights.

Financial entities (banks, broker-dealers, e-wallets, crowdfunding entities, etc) are required to give notice to their supervisory authority (the CNBV in the case of the aforementioned entities) to contract technology services with third parties (eg, Chapter XI, Section 3 of the CUB). Although insurance companies are not required to give such notice to their supervisory authority, they must fulfil the requirements established in applicable law (ie, Article 268 of the Insurance and Bond Institutions Law).

Furthermore, third parties contracting technology services with financial entities in Mexico are generally obliged to specify, in the agreements, that they will submit the relevant information required by the applicable financial entity, or by the respective supervisory authority, for the purposes of oversight.

Beyond the cybersecurity considerations associated with the use of AI, deploying this technology in financial services may also give rise to questions about discriminatory outcomes (eg, in relation to credit-scoring or insurance underwriting).

There are no specific regulations governing the use of AI in healthcare in Mexico. AI‑enabled medical technology is subject to the general health and sanitary framework. Under Article 262 of the General Health Law (Ley General de Salud), “medical equipment” comprises equipment, accessories and instruments used for medical or surgical care or for diagnostic, treatment or rehabilitation procedures, which can cover AI‑based systems: such equipment requires prior sanitary authorisation from COFEPRIS and is subject to post‑market “tecnovigilancia” under Article 262 Bis.

Recent amendments (DOF, 15 January 2026) introduced a specific chapter on digital health, defining it as the use of information and communication technologies in health services, including telemedicine, electronic health records and the analysis of large datasets to identify patterns, optimise diagnoses and personalise treatments. Digital health services must ensure confidentiality, personal data protection, secure systems and informed consent for remote care.

AI in healthcare typically involves sensitive personal data, requiring express written consent and heightened safeguards, in line with the data protection rules described in sections 3.6 Data, Information or Content Laws and 17.1 AI Training and Data Protection, and is also subject to institutional bioethics and research ethics oversight under Article 41 Bis.

There are no specific regulations in Mexico governing the use of AI in autonomous vehicles.

In practice, the deployment and testing of autonomous vehicles would be subject to general laws, including traffic rules and those relating to consumer protection and data protection. From a data protection and transparency perspective, these systems may process location, behavioural and potentially sensitive data, for example where driving patterns or travel routes are used by companies (eg, insurers or service providers) to segment users, adjust prices or infer certain sensitive characteristics (eg, regular visits to hospitals). This triggers consent, purpose limitation, data minimisation and security obligations, as well as the special rules for sensitive personal data discussed in sections 3.6 Data, Information or Content Laws and 17.1 AI Training and Data Protection, and may raise concerns about discriminatory or opaque profiling.

Where autonomous vehicles are defective or fail to provide the safety reasonably expected, consumers may seek remedies under the LFPC, and PROFECO may impose precautionary measures (see 15.5 Retail and Consumer). Liability for harm caused by autonomous vehicles would be assessed under general civil principles, including the strict liability framework applicable to dangerous instruments (see 10.1 General Theories of Liability).

The LFPC applies to AI-enabled goods and services on the same basis as any other product or service.

Where goods or services (including AI‑enabled) are defective, unsafe or do not provide the safety that can reasonably be expected (eg, an AI-powered recommendation tool that causes financial or medical harm, a customer-facing chatbot that provides inaccurate product advice or an autonomous vehicle with defective safety systems), consumers may seek remedies such as restitution of the product or service, rescission of the contract or a reduction of the price. This applies in any case of bonification or compensation where there are defects or hidden faults that make the product or service unfit for its ordinary use, diminish its quality or usability or fail to provide the level of safety that would normally be expected from reasonable use. In addition, when AI‑related products or services affect or may affect the life, health, safety or economic interests of a group of consumers, the Consumer Protection Agency (Procuraduría Federal del Consumidor, PROFECO) may impose precautionary measures, including immobilisation or seizure of products, suspension of commercialisation, product withdrawal, suspension of advertising and consumer alerts or recalls. Manufacturers or distributors must inform authorities immediately if they discover a risk.

The LFPC applies to end users (individuals or legal entities) and to persons who integrate goods or services into their value chains, though in the case of legal entities the latter category is limited to certain small enterprises in accordance with the applicable size tests under Mexican law. For general liability theories, see 10.1 General Theories of Liability.

As noted in 2.1 Industry Use, AI robotics is Mexico’s fastest-growing AI segment, and Mexico installed 5,600 industrial robots in 2024. No sector-specific regulatory framework governs the use of AI in industrial and robotics contexts in Mexico. Harm caused by AI-enabled machinery or automated industrial systems may trigger civil liability under Article 1913 of the FCC, on the basis that such systems constitute inherently dangerous instruments. Consumer protection rules may also apply where industrial AI products or services reach end users. See also 10.1 General Theories of Liability and 15.5 Retail and Consumer.

Certain AI system components may be protected under Mexican IP laws through two primary regimes: the LFDA and the IPL.

LFDA

• Computer programs – including AI-related software and code – are protected as works (obras) in both source and object code from the moment of fixation, without registration.
• Datasets and databases may qualify as protected compilations to the extent their selection or arrangement constitutes an intellectual creation; protection does not extend to the underlying data. Methods, systems, inventions and processes are expressly excluded from protection under the LFDA.
• For commissioned works, economic rights vest in the commissioning party. For computer programs created by employees in the course of their employment, economic rights vest entirely in the employer; for other works, rights are presumed equally divided under a written employment contract (unless otherwise agreed in writing).

IPL

• Certain assets not protected under the LFDA – which may include, for example, model weights, training methodologies, and proprietary processes – may qualify as trade secrets under the IPL, provided the information has industrial or commercial application, confers a competitive or economic advantage, and sufficient measures to preserve its confidentiality and restrict access have been adopted.
• The IPL expressly provides that administrative IP infringements, including trade secret misappropriation, are sanctionable even when carried out through the use of AI (see 5.3 Enforcement Actions).
• Registered trademarks and slogans (avisos comerciales) may protect AI-generated outputs (eg, logos), provided they satisfy the usual requirements of distinctiveness and availability. Protection depends on those criteria and on registration, rather than on the identity of the person who created the mark.
• AI-generated inventions raise patentability issues, as Article 46 expressly defines an invention as a human creation (creación humana).
• AI-generated industrial designs face obstacles to registration, as Articles 36, 38 and 39 frame the originating right and the status of designer around natural persons (personas físicas).
• Note that, while the IPL frames inventors and designers as natural persons (personas físicas), patents, utility models, industrial designs and trademarks may ultimately be owned by either natural or legal persons.
• The extent to which AI-assisted inventions and designs may qualify – where a human contribution is sufficiently documented – remains unclear in both cases.

Mexico has yet to define the IP framework applicable to AI training data and AI-generated outputs, as recommended by UNESCO.

Author: please refer to 4.1 Precedent-Setting Judicial Decisions.

Inventor: Article 39 of the IPL states that an inventor is presumed to be one or more individuals.

The authors are not aware of judicial or agency decisions relating to whether AI technology can be an inventor or co-inventor for patent purposes.

For ownership of AI-assisted works and the role of human contribution in qualifying for protection, see 16.1 IP Protection for AI Assets.

• The LFDA does not include a TDM exception, nor does it recognise a general fair use or fair dealing doctrine; limitations and exceptions are statutory and exhaustive. Uses outside those cases require authorisation or may constitute infringement under the LFDA or an infracción en materia de comercio. No specific licensing framework for the use of works in AI training exists; authorisation from rights holders is therefore the primary mechanism for legitimising the use of protected works in training datasets (see 3.6 Data, Information or Content Laws and 8.1 Generative AI: Key Legal Issues and Regulatory Approaches).
• The LFDA additionally prohibits circumvention of technological protection measures (Article 112) and expressly sanctions the commercialisation of circumvention tools and the act of circumvention itself (Articles 232 Bis and 232 Ter); assembling training datasets by bypassing such measures may give rise to additional liability. Where scraped data includes confidential information meeting the IPL’s trade secret definition, its unauthorised acquisition may constitute misappropriation (see 16.1 IP Protection for AI Assets).
• Moral rights under the LFDA are non-waivable (irrenunciables) and inalienable, and remain with the human author regardless of any subsequent use of the work for AI training purposes (see 4.1 Precedent-Setting Judicial Decisions).
• The use of images of real persons to train AI systems without their authorisation may additionally constitute an infracción en materia de comercio under Article 231 of the LFDA.
• Liability for outputs that reproduce or closely resemble protected training data is addressed through existing IP and civil frameworks (see 10 Liability for AI).

The authors are not aware of judicial or administrative decisions specifically addressing the use of protected works for AI training. No specific legislation on AI training data has been enacted. UNESCO has recommended that Mexico define the IP framework applicable to the use of works in AI training (see 16.1 IP Protection for AI Assets).

Please refer to 4.1 Precedent-Setting Judicial Decisions.

The use of protected works to train foundation models is subject to the general copyright analysis under the LFDA; no licensing exception exists for training purposes, and general infringement risk applies to the use of such works (see 16.3 Copyright and AI Training Data). Ownership of outputs generated by foundation models follows the analysis in 16.1 IP Protection for AI Assets.

Companies that integrate or use foundation or open-source AI models in their products or services remain subject to the same IP and civil liability frameworks as any other deployer, but any claim based on outputs that reproduce or closely resemble protected training data would need to be assessed on a case-by-case basis under those frameworks, including the usual requirements on proof of copying, similarity and causation (see 10.1 General Theories of Liability).

The unauthorised use of protected works as training data by large foundation model developers has gained legislative attention, including a Senate colloquium convened to examine the adequacy of existing copyright protections in this context; no specific rules have resulted to date.

The LFPDPPP’s data protection principles: lawfulness, purpose limitation, loyalty, consent, data quality, proportionality, information and accountability apply to AI training involving personal data (see 3.6 Data, Information or Content Laws). Controllers may only process personal data that is necessary, adequate and relevant for the stated purposes. They also must ensure that data is accurate and up to date, and that any subsequent AI training is compatible with the original purpose or supported by a separate lawful basis (eg, compliance with a legal obligation).

AI training involving biometric data requires special scrutiny and triggers the enhanced protections discussed in 12.2 Biometric Technologies and Emotion Recognition. By contrast, transaction, billing or location data may initially be treated as non-sensitive personal data when used, for example, to comply with AML or fraud‑prevention obligations under applicable law, but may be considered sensitive personal data if AI training uses those variables in a way that reveals or infers any sensitive trait, such as health condition (eg, medication purchases), religious or political affiliation (eg, payments to politically-affiliated bodies), other intimate aspects of the data subject or behavioural patterns derived from continuous location tracking (eg, regular attendance at medical facilities, religious sites or political gatherings).

Variables such as gender, ethnicity or health status may give rise to discriminatory processing if used in ways that systematically result in less favourable treatment for certain groups. Even where the training dataset does not explicitly include sensitive attributes, AI models may in some cases learn correlations that could lead to indirectly discriminatory outcomes for women or minority groups (eg, by inferring lower creditworthiness or higher risk from proxies such as postcode, type of employment or consumption patterns). Such risks should be addressed through careful selection and treatment of input variables, bias testing and appropriate mitigation measures.

Where a training dataset includes sensitive personal data, controllers must obtain express written consent, justify any database containing such data for legitimate and concrete purposes aligned with their activities and limit the processing period to the minimum necessary (see 17.2 AI Deployment and Data Subject Rights). Under Article 13 of the LFPDPPP, controllers must also adopt the measures necessary to ensure compliance with data protection principles and guarantee that the privacy notice is respected at all times, including by third parties with whom they have a legal relationship. Article 29 further requires that every controller designate a person or department responsible for processing data subjects’ requests, the closest domestic equivalent to a data protection officer obligation.

Regarding anonymisation, the LFPDPPP uses the concept of disociación (ie, a process by which personal data cannot be associated with the data subject, or be used, by virtue of its structure, content or degree of disaggregation, to identify that person). Once data has been processed so that it can no longer be associated with an identifiable person, it falls outside the LFPDPPP regime. The law does not recognise pseudonymisation as a distinct legal category; partially anonymised data that still permits re-identification remains subject to the personal data regime. Training or testing AI systems on fully disassociated or truly synthetic datasets therefore falls outside the LFPDPPP regime.

The lawful basis and transparency obligations applicable to AI-driven processing are discussed in 3.6 Data, Information or Content Laws and 17.1 AI Training and Data Protection.

Where personal data is subject to automated processing without human intervention, in a way that produces adverse legal effects or significantly affects a data subject’s interests, rights or freedoms, and where such processing is designed to evaluate, analyse or predict aspects such as professional performance, economic situation, health, sexual preferences, reliability or behaviour, the data subjects can object to, or request cessation of, processing, except where processing is necessary to comply with a legal obligation. This is relevant for AI‑based HR tools, scoring, KYC and risk models that operate on a fully automated basis (see 3.1 General Approach to AI-Specific Legislation).

Data subjects may exercise the following rights before the controller (known as ARCO rights):

• Access personal data processed and information on the conditions and generalities of its processing;
• rectify inaccuracies or incomplete or outdated personal data;
• cancel their personal data, which entails a prior blocking period and subsequent deletion, subject to the statutory exceptions where retention is permitted; and
• oppose certain forms of personal data processing (eg, automated processing that produces adverse legal effects or significantly affects their interests, rights or freedoms).

The LFPDPPP does not impose data protection impact assessment obligations for AI systems, nor does it include an express privacy by design or by default requirement. Nonetheless, the SABG can conduct privacy impact studies prior to the implementation of new processing modalities or material modifications to existing ones, which is relevant where AI systems introduce a new or materially different form of processing. Given the relative novelty of this authority, there are limited precedents on how these powers will be exercised.

The LFPDPPP imposes security obligations that apply to AI-related processing in the same way as to any other data processing action. Controllers must implement administrative, technical and physical safeguards proportionate to the inherent risk, including the scale of processing involved, and must notify data subjects without undue delay when a security incident significantly affects their economic or moral rights.

Where AI models are hosted in the cloud or rely on third‑party providers, those providers qualify as processors (a party that processes personal data on behalf of the controller) and must be bound by contracts or other legal instruments that reflect these requirements. Under the current regulations, such instruments must allow the controller to demonstrate the existence, scope and content of the processing arrangement, and the processor’s obligations – including treating data solely per the controller’s instructions, maintaining confidentiality, implementing applicable security measures and suppressing or returning data upon termination of the relationship – must be expressly covered.

For cross-border transfers, the LFPDPPP does not establish an adequacy framework or a list of approved recipient countries. Instead, it requires that recipients – whether domestic or foreign – be informed of the privacy notice and assume the same obligations as the transferring controller when processing personal data. This applies equally to transfers made for AI training or deployment purposes, including to cloud providers or AI model developers located abroad.

The Federal Economic Competition Law (LFCE), enforced by the CNA following the July 2025 reform, applies to AI-related markets without sector-specific rules, prohibiting monopolies, monopolistic practices, unlawful concentrations and barriers to competition.

Absolute monopolistic practices – such as price-fixing and market allocation – are void and subject to significant fines, which can capture concerns around algorithmic collusion and AI-driven price coordination. Relative practices by firms with substantial market power – including exclusivity, tying and denial of access to essential inputs (insumos esenciales) – are prohibited where they harm competition. In the authors’ view, and depending on the factual context, data, cloud infrastructure and key AI services may be characterised as essential inputs under the LFCE’s criteria.

The 2025 reform reduced merger notification thresholds, making AI-driven acquisitions, talent-focused acquisitions (acqui-hires) and vertical integrations across the AI value chain (foundation models, cloud infrastructure and applications) more likely to trigger mandatory review and potential conditioning or blocking by the CNA.

Mexico lacks AI-specific cybersecurity legislation. Several bills have been introduced in the Federal Congress to amend Article 73 of the Mexican Constitution to grant Congress explicit authority to legislate on AI and cybersecurity. These includes proposals filed in September 2023, November 2025, and most recently in April 2026, the latter of which looked to expressly combine AI and cybersecurity in a single proposal. None of these bills have been enacted.

In the meantime, the General Cybersecurity Policy for the Federal Public Administration, published in December 2025, establishes a general cybersecurity framework applicable to all activities, processes, technologies, information systems and digital assets of federal government entities, as well as introducing incident reporting obligations and supply chain security requirements. No specific rules address adversarial attacks, data poisoning or secure AI development lifecycles.

According to the Latin American AI Index 2025 (pgs. 189-191), Mexico ranks among the regional leaders in sustainable data centre practices and has developed a sustainable finance taxonomy, in line with regional trends. This taxonomy operates as a voluntary reference framework to classify environmentally and socially sustainable activities and to support access to green or sustainable financing for sectors such as data infrastructure.

The General Regulations applicable to Capital Markets Issuers (Circular Única de Emisoras) require issuers to prepare a sustainability report in line with the IFRS Sustainability Disclosure Standards. To facilitate compliance, the CNBV has presented a set of support tools:

• a supporting document for the preparation of the sustainability report;
• an ESG self‑assessment tool, allowing entities to evaluate the integration of sustainability and climate risks and opportunities in their business; and
• the “Ruta de la Sostenibilidad” platform. These tools aim to strengthen transparency, comparability and alignment of the Mexican securities market with international best practices.

On the social and governance aspects, see 12.1 Algorithmic Bias and Fairness, 14.1 Hiring and Termination Practices, and 21.1 AI Governance Frameworks and Implementation.

In Mexico, no binding AI governance framework exists for the private sector. The non-binding Chapultepec Principles (see 3.3 Jurisdictional Directives) provide the closest domestic reference, addressing AI lifecycle governance, prior impact assessments and identifiable human responsibility in AI-assisted decisions. ISO/IEC 42001 remains the primary voluntary standard for AI risk management in Mexico (see 6.1 National Standard-Setting Bodies).

Key issues for implementing AI governance include:

• periodically training staff at all levels who interact with AI to identify potential biases and inaccurate outputs, given the fast-paced evolution of the technology;
• homogenisation of the functioning of ethics committees, as has occurred with other similar corporate structures in the past;
• identification of key obligations that developers and/or deployers must meet for AI interfaces already in use or to be used in the future;
• governance of third-party AI tools and vendors, including due diligence on AI systems integrated into business processes, in the absence of specific contractual or regulatory standards in Mexico; and
• determination of decisions that cannot be made by AI without human intervention (eg, pertaining to employment issues).