Explainable AI and Transparent Social Security Decisions

This page is a translation of the original Portuguese text, provided for convenience. It deals with Brazilian law. If there is any discrepancy, the Portuguese version prevails. Read the original in Portuguese

Artificial intelligence already influences decisions on social security benefits in Brazil, but a lack of algorithmic transparency can undermine the fundamental rights of insured persons.

The Rise of Artificial Intelligence in Brazilian Social Security

In recent years, we have seen a significant transformation in how the INSS (Brazil’s National Social Security Institute) processes and reviews benefit applications. Automated systems now perform tasks that once depended entirely on public employees, from the initial screening of documents to the automatic granting of certain benefits. This technological shift has brought undeniable efficiency gains, shortening queues and speeding up cases that used to take months to resolve.

However, the speed at which these systems were rolled out raises fundamental questions about the quality and fairness of automated decisions. When an algorithm denies a social security benefit, the insured person often receives a generic response and has no way of knowing the specific criteria behind the outcome. This algorithmic opacity is a legal challenge of the first order, because it collides with core constitutional principles such as due process of law, the duty to state reasons for administrative acts, and the right to a full defense.

We have observed that the digitization of social security services, which accelerated especially from 2020 onward, has created a paradoxical situation. On one hand, technology has democratized access, allowing insured persons in remote areas to file applications without having to travel. On the other, it has put up invisible barriers for those who cannot understand the logic behind automated decisions, creating a new form of exclusion that hits the most vulnerable hardest.

What Explainable AI Is and Why It Matters for Social Security Law

Explainable artificial intelligence (known by the acronym XAI) refers to a set of techniques and methods that make the workings of AI systems understandable to human beings. In practical terms, an explainable AI system does not merely deliver a result (approving or denying a benefit, for example). It also provides the detailed reasons behind that specific decision.

In the social security context, explainable AI would mean that the insured person has access to clear information about which criteria were evaluated, how much weight each factor received, and why certain information led to the conclusion reached. Consider a retirement application based on contribution time: an explainable system would indicate exactly which periods were recognized, which were discarded, and the specific reasons for each partial decision that made up the final result.

We distinguish three levels of explainability that we consider relevant to social security law. The first is procedural transparency, which shows the steps the system followed to reach its decision. The second is causal explainability, which reveals which factors actually influenced the result. The third is counterfactual explanation, which tells you what would have had to be different for the result to change. This last level is particularly valuable to lawyers and insured persons, because it pinpoints what can be corrected or supplemented in an administrative appeal.

Algorithmic transparency in social security decisions is not a technological luxury. It is a constitutional requirement that guarantees insured persons the right to understand, question and appeal any decision that affects their benefits.

The importance of explainable AI in social security law goes beyond technical curiosity. It is a democratic requirement. The Federal Constitution establishes that all administrative acts must state their reasons, and that requirement cannot be watered down simply because an algorithm, rather than a human employee, made the decision. Algorithmic reasoning must be just as robust and accessible as human reasoning.

Brazil’s Legal Framework for AI and Its Effects on Social Security

The legislative debate over regulating artificial intelligence in Brazil has advanced considerably in recent years. The National Congress has been considering proposals that set basic principles for the government’s use of AI, including requirements for transparency, explainability and human oversight. These discussions reflect growing concern about the social impact of automating decisions that affect fundamental rights.

Even without a specific AI statute, Brazilian law already contains provisions that support a requirement of transparency in automated decisions. The General Data Protection Law (Law 13.709/2018) provides, in Article 20, that data subjects may request review of decisions made solely on the basis of automated processing of personal data. This provision applies directly to automated social security decisions, since processing a benefit application necessarily involves handling the insured person’s personal and sensitive data.

Beyond the LGPD (General Data Protection Law), the Federal Administrative Procedure Law (Law 9.784/1999) already requires the INSS to state reasons for its decisions, identifying the facts and legal grounds that support them. When an automated system makes a decision, that duty remains fully in force. We argue that automation, far from excusing the duty to state reasons, makes it even more necessary, because the insured person loses the chance to speak directly with the decision-maker to get informal clarification.

Applicable Constitutional Principles

Several constitutional principles converge to support a requirement of explainability in automated social security decisions. The principle of publicity (Article 37 of the Federal Constitution) requires that acts of public administration be transparent. The principles of adversarial proceedings and full defense (Article 5, item LV) presuppose that citizens know the grounds of a decision so that they can challenge it properly. The principle of efficiency (Article 37) cannot be read in a way that sacrifices fundamental guarantees for the sake of speed.

We understand that, taken together, these principles create a constitutional duty of algorithmic explainability that binds the INSS and any public body that uses AI systems to make decisions affecting citizens’ rights. The point is not to prevent the use of technology, but to require that it be implemented in a way that is consistent with the democratic rule of law.

Practical Challenges and the Role of the Social Security Lawyer

In day-to-day practice, we find that the opacity of automated decisions creates real obstacles for social security lawyers. When a benefit is denied by an automated system and the stated grounds amount to codes or boilerplate phrases, the lawyer has trouble identifying the exact point in dispute and building an effective appeal.

This situation demands a dual competence from the social security lawyer. On one hand, technical command of social security rules remains indispensable. On the other, it is increasingly necessary to understand, at least in broad terms, how automated systems work, which databases they query and which criteria they apply. That understanding makes it possible to draft more precise information requests and to build stronger arguments in both administrative and judicial proceedings.

We have identified several strategies that practitioners can use when faced with automated decisions they consider unfair. First, it is possible to file an administrative request for a detailed account of the criteria the system used, based on the LGPD and the Access to Information Law (Law 12.527/2011). Second, where that detail is not provided or is insufficient, the omission can serve as an additional argument in an administrative appeal or lawsuit, as a violation of the duty to state reasons.

Human Oversight as an Essential Safeguard

We maintain that no social security decision that denies or reduces a benefit should be made exclusively by an automated system, without any human oversight. The so-called “human in the loop” is not merely a good AI governance practice. It is a requirement of Brazilian law for decisions that affect fundamental rights.

Qualified human oversight works as an additional layer of protection against algorithmic errors, bias in training data, and unusual situations that the automated system was not designed to handle properly. Cases involving special working conditions, rural work periods with no formal records, or informal employment are examples of situations in which human analysis attuned to the specific case remains irreplaceable.

Prospects for a More Transparent Social Security System

We believe that the convergence of technology and social security law can produce positive results for everyone involved, provided that implementation respects fundamental rights. A social security system that uses explainable AI can be both more efficient and fairer, shortening review times without sacrificing the quality of decisions or the transparency of the process.

For that to happen, we think several structural measures are needed. The first is the adoption of technical explainability standards for every AI system the INSS uses, with periodic independent audits. The second is training for the public employees who operate and supervise these systems, so that they can actually exercise human control. The third is the creation of accessible channels through which insured persons and their lawyers can question automated decisions and obtain detailed explanations in plain language.

Mechanisms for algorithmic auditing are also important, so that it can be verified whether social security AI systems show discriminatory bias. International studies show that algorithms can reproduce and amplify inequalities present in the historical data used to train them. In the social security context, this could mean perpetuating disparities in how the rights of certain groups of insured persons are recognized.

Building a digitally transparent social security system is an ongoing process that requires the active participation of civil society, legal professionals and insured persons themselves. Critically monitoring how these technologies are implemented is not only a right but a civic duty, one that helps improve democratic institutions and protects the social security rights of all Brazilians.

This topic is directly related to digital medical examinations, which we cover in a dedicated article.

This topic is directly related to generative AI and the, which we cover in a dedicated article.

As informações deste artigo são de caráter informativo e não substituem consulta jurídica individualizada. Cada caso possui particularidades que exigem análise específica.