The Ibero-American Data Protection Network approves the first international standards on AI and personal data
The Ibero-American Data Protection Network unanimously approves standards detailing how to apply privacy rules to artificial intelligence systems.

The Ibero-American Data Protection Network (RIPD), the forum in which the Spanish Data Protection Agency (AEPD) holds the permanent secretariat, has unanimously approved the Ibero-American Standards on the use of personal data in artificial intelligence systems. The approval took place on 12 November 2026 and, according to the AEPD, it is the first international instrument to address, in a comprehensive and systematic way and through an articulated body of provisions, the application of personal data protection to AI processing.
The document has international scope and a normative format, and is addressed to authorities, legislators, companies and administrations. Its stated purpose is not to create a parallel legal regime or new general obligations, but to develop the specific issues raised by applying data protection law to artificial intelligence, with an approach based on risk, proportionality and technological neutrality. The Agency stresses that the text supports innovation compatible with the rights of more than 600 million people across the Ibero-American region.
Inferred data: AI deductions are personal data too
One of the cores of the Standards is the protection of the data that AI systems infer or deduce about people. The text starts from a significant premise: inferences, predictions and profiles can constitute personal data when they relate to identified or identifiable persons, and their use must take into account the associated degree of uncertainty, without attributing to them more certainty than is justified.
The practical consequence is significant: when a system deduces aspects of health, ideology or sexual orientation from apparently neutral data, the reinforced protection proper to sensitive data applies. For firms advising companies deploying predictive analytics, this turns the classification of inferences into a first-order compliance task, not a theoretical nuance.
Public is not free: limits on training with internet data
The Standards set a principle that settles one of the sector's most frequent debates: the mere fact that personal data is publicly accessible is not enough, on its own, to use it in training AI systems. An adequate legal basis is required, along with regard for what people could reasonably expect when they published that data.
The formulation connects with the European doctrine on reasonable expectations, but expressly transposes it to model training. For anyone compiling datasets from open sources, the standard requires documenting the legal basis for the processing and analysing compatibility with the purpose for which the data was published, instead of presuming lawfulness from mere accessibility.
Transparency: saying "we use AI" is not informing
On transparency, the Network establishes that merely indicating that AI is used does not necessarily satisfy the duty to inform. Organizations must explain in understandable language what they use the system for, what role it plays and what consequences it may have for people, and trade secrecy does not justify evading that right.
The standard also strengthens rights vis-à-vis already trained models. People must be able to know whether their data has been used in an AI system and, in the case of automated decisions, to challenge their conclusions, demand real human intervention and exercise their corresponding rights. The text is explicit on a delicate point: if the person reviewing a decision merely rubber-stamps what the AI proposes, there is no effective human intervention. And when a model does not allow personal data to be fully erased, the organization must justify it, inform the person and adopt measures such as excluding that data from future training.
The Standards also draw a red line on uses: they declare incompatible with data protection those AI systems intended to manipulate deceptively, to exploit people's vulnerability or to score them socially in a discriminatory way.
Allocation of responsibilities along the AI chain
The document determines the responsibilities of the various actors involved in developing and using the systems. Two rules stand out. A provider's use of clients' personal data to train or improve its own models requires an adequate legal authorization and compliance with the corresponding obligations, so the service contract does not by itself authorize that training. And whoever uses third-party AI systems remains responsible for the processing whose essential purposes and means they determine, even if they neither developed nor trained the model.
This second rule most directly affects firms and advisers that integrate commercial tools into their practice: outsourcing the technology does not outsource responsibility for the processing the firm itself decides to carry out with it.
Why these Standards differ from a declaration of principles
The AEPD insists on the instrument's architecture: the Standards avoid reproducing existing general obligations, distinguish processing according to its risks and favor proportionate measures. Unlike general declarations of principle or guides devoted to specific issues, they offer a specialized and cross-cutting framework, built as an articulated body of provisions. The starting premise is that data protection and technological innovation are compatible objectives: AI can deduce traits about people, predict their behavior and make automatic decisions affecting their lives (when they apply for credit, apply for a job or access a service), and those capabilities do not alter the validity of data protection principles, but require detailing how they apply.
Specific guidelines for generative AI, biometrics and neurodata
The document includes specific guidelines for generative and agentic AI, biometrics, emotion recognition and neurodata, and stresses the importance of data protection by design, governance, data quality, human oversight and risk assessment and management, with measures adapted to the characteristics of each processing operation.
The Standards are the result of a process of drafting, consultation and technical deliberation among the RIPD member authorities, starting from an initial version by expert consultant Marcos Sánchez Acevedo. The work was funded by the European Union in collaboration with the Fundación para la Internacionalización de las Administraciones Públicas (FIAP), within the EU-LAC Digital Alliance, and complements the already existing Data Protection Standards for Ibero-American States, which incorporated advances on artificial intelligence, automated decisions, bias prevention and algorithmic governance.
What it means for practice in Spain
Although the Standards are not in themselves binding rules, the AEPD holds the Network's permanent secretariat and instruments of this kind guide how authorities interpret the GDPR and national law. In practice, they set the reasonable level of demand on matters that courts and agencies were already addressing case by case: inferred data, training on public sources, effective human intervention and the responsibility of the system user.
For firms advising on AI projects, the document offers something valuable: an articulated framework, unanimously approved, to which risk analyses and impact assessments can refer. Related coverage in our pieces on the AEPD decalogue for AI assistants and agents, on agentic AI and data protection and on generative AI policy in law firms, plus the guide on data protection for the self-employed lawyer.
FAQ
What have the Ibero-American data protection authorities approved? The Ibero-American Data Protection Network unanimously approved, on 12 November 2026, the Ibero-American Standards on the use of personal data in artificial intelligence systems, the first articulated international instrument on the subject.
Can public internet data be used to train AI? Not merely because it is public. The Standards require an adequate legal basis and regard for what people could reasonably expect when publishing the data.
What is effective human intervention in an automated decision? A genuine review of the decision. The Standards specify that if the reviewer merely rubber-stamps what the AI proposes, there is no effective human intervention.
Who is responsible when a third-party AI system is used? The user remains responsible for the processing whose essential purposes and means they determine, even if they neither developed nor trained the model.
Sources
- Spanish Data Protection Agency (AEPD), "La Red Iberoamericana de Protección de Datos aprueba unos Estándares internacionales pioneros sobre inteligencia artificial y protección de datos personales" (press release, 12 November 2026), consulted 13 November 2026.
Fuentes
- Source. Accessed 13 November 2026.
The specific data in this article come from the official sources cited.


