ATLAS
← Back to News and Insights

Brazil’s ANPD chairs approval of Ibero-American AI data standards

3 min

The guidance says public availability is not a licence to train models, and generic AI notices are inadequate.

LinkedIn Post
Brazil’s ANPD chairs approval of Ibero-American AI data standards
Brazil · Photo: Renan / Unsplash

Brazil’s National Data Protection Authority (ANPD), chairing the Ibero-American Data Protection Network (RIPD) for 2025 to 2027, announced on 18 November 2026 that the network had unanimously approved the Ibero-American Standards on the Use of Personal Data in Artificial Intelligence Systems. The document, available in Spanish, gives guidance on applying data protection rules to processing that uses AI, including automated decisions, generative and agentic AI, biometrics, emotion recognition and neurodata.

The standards are described as guiding and complementary, rather than as a new binding regulation. For compliance teams, however, they establish a detailed regional reference point on training data, explainability, vulnerable people and AI governance. Atlas reads the approval as a soft-law convergence step: firms operating across Ibero-American markets now have a common set of expectations to test against.

Training data and meaningful notices

The most consequential data-use statement is that publicly available personal data cannot be used to develop or train an AI system merely because it is public. The standards say the use still requires an appropriate legal basis, respect for the data’s original purpose and consideration of people’s reasonable expectations.

That applies beyond the information used directly to train a model. The standards address inferred data, alongside purpose limitation, data minimisation, transparency and explainability. Firms should therefore be able to identify the sources and categories of personal data entering an AI pipeline, the purpose for which the data was first collected and the legal basis supporting the later use.

The transparency requirements are also more specific than a general statement that a business uses AI. People should receive understandable information about:

- the system’s purpose;
- the organisation’s role in the processing;
- the categories and sources of data;
- the degree of human intervention; and
- the consequences the technology may produce.

Where necessary, the standards call for a meaningful explanation of the factors that influenced an outcome. This is relevant to firms using AI to assess, select or prioritise people, as well as to organisations making or supporting automated decisions. A privacy notice that identifies AI in the abstract would not address all of these points.

Stronger controls for vulnerable people

Systems aimed at people or groups in vulnerable situations, or capable of significantly affecting them, should use protective configurations by default. The standards also call for age-appropriate design and stronger safeguards against profiling, behavioural advertising and improper collection of personal data.

The document gives separate attention to generative and agentic AI, biometrics, emotion recognition, neurodata and systems used to assess, select or prioritise people. It also covers algorithmic governance, impact assessments, human oversight, security, incident response and the allocation of responsibility between actors in the AI value chain.

The language is important but must be read with the instrument’s legal status in mind. The standards use directive language for safeguards, while the ANPD describes the overall document as guidance intended to promote harmonised application of data protection. Firms should not treat the approval as equivalent to a Brazilian resolution.

The emphasis on children and vulnerable users also fits the ANPD’s recent domestic work. On 6 November, the authority concluded its first AI regulatory sandbox and announced a second edition focused on the safe use of digital games by children and adolescents.

From regional guidance to domestic supervision

The standards were prepared with the participation of RIPD authorities, including the ANPD, and funded by the European Union with the Foundation for the Internationalisation of Public Administrations under the EU-LAC Digital Alliance. Their stated aim is regulatory convergence and innovation compatible with fundamental rights and freedoms.

The ANPD’s wider enforcement activity shows why the underlying data-governance requirements matter even where AI is not involved. On 26 October, the authority sanctioned Feira de Santana city hall with warning and publication sanctions and ordered corrective measures after the exposure of the names and diagnoses of 1,241 people in the municipal Official Gazette. The case concerned a basic disclosure failure, but the same controls on purpose, access, security and incident response underpin responsible AI deployment.

For firms, the immediate review should cover four areas: whether training and fine-tuning data has a defensible legal basis; whether original purposes and reasonable expectations have been assessed; whether notices explain the system’s role and effects; and whether high-impact or vulnerable-user applications have default safeguards, human oversight and an impact assessment.

With the ANPD chairing the RIPD through 2027, Atlas expects the standards to serve as a continuing reference for Brazil’s domestic AI policy and the network’s next phase of regulatory convergence.

Official source: gov.br
AIAI FrameworksAI Governance

Permanent link to this Atlas analysis

Continue with coverage connected by market, topic and operating context.

Continue with Atlas

Move from this development into the relevant research, comparison and workflow.