Algorithmic Governance at the INSS

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

Algorithms already decide some Social Security benefits in Brazil, and the lack of proper governance puts millions of people insured by the INSS (Brazil’s National Social Security Institute) at risk.

The Rise of Algorithms in Social Security Decisions

In recent years, the Instituto Nacional do Seguro Social has gone through a significant digital transformation. Processes that once relied entirely on human review have begun to incorporate automated decision systems, from the initial screening of applications to the approval or denial of benefits. This shift has brought promises of greater efficiency and shorter waiting times for insured persons, but it has also raised fundamental questions about transparency, fairness, and accountability.

Looking at the current landscape, automation at the INSS is not limited to simple operational tasks. Computer systems cross-check registration data, verify employment relationships, calculate contribution periods, and, in many cases, issue decisions without a human employee reviewing the case individually. This model of algorithmic decision-making can process enormous volumes of applications, but it carries inherent risks that must be understood and mitigated through robust governance.

In this context, algorithmic governance refers to the set of principles, rules, processes, and oversight mechanisms that guide the development, implementation, and supervision of the automated systems used by the INSS. Without this governance structure, algorithms operate as black boxes. They make decisions that directly affect the livelihoods of retirees, pensioners, and vulnerable workers, with no real way to question those decisions or understand the criteria behind them.

Algorithmic Transparency and the Rights of the Insured Person

One of the fundamental pillars of algorithmic governance is transparency. In the Social Security setting, many insured persons receive denial decisions without understanding the real reasons behind the refusal. When the denial results from automated processing, this opacity gets worse, because in many cases not even the INSS employee in charge of the review can say precisely which criteria the system applied to reach its conclusion.

The 1988 Federal Constitution guarantees the rights to a full defense and to an adversarial process (contraditório, the right to know the grounds of a decision and respond to them) in administrative proceedings. For those rights to be exercised effectively, the insured person needs access not only to the outcome of the decision but also to the logic behind it. When an algorithm denies a benefit, citizens have the right to know what data was considered, what rules were applied, and how the system concluded that the requirements had not been met.

Algorithmic transparency at the INSS therefore involves clear documentation of the models in use, accessible explanations of the decision criteria, and effective channels through which insured persons can challenge results they believe are wrong. Without these elements, automation becomes an obstacle to exercising fundamental rights. It turns upside down the logic that should guide public administration: technology exists to serve the citizen, not to create additional barriers.

Technology at the INSS should expand access to Social Security rights, never restrict them through automated decisions without adequate human oversight.

We also note that the General Data Protection Law (Law 13.709/2018) establishes the right to have decisions reviewed when they are made solely on the basis of automated processing of personal data. This provision applies directly in the Social Security context. It gives the insured person the right to ask that a human being reassess an algorithmic decision, ensuring that automation does not override the qualified judgment of a public employee.

Risks of Bias and Discrimination in Automated Systems

A critical aspect of algorithmic governance that we examine closely is the risk of discriminatory bias. Algorithms are built from historical data, and that data can reflect structural inequalities that already exist in Brazilian society. If an automated decision system at the INSS was trained or configured on patterns that reproduce discrimination (based on gender, race, region, or social class), it tends to perpetuate and even amplify those injustices.

Consider, for example, rural workers and artisanal fishers, who have historically had difficulty proving employment relationships and contributions. An algorithm that gives excessive weight to the regularity of formal records can systematically harm these groups, disregarding the particulars of informal work and the documentation problems these insured persons face. Likewise, workers in regions with weaker digital infrastructure can be disproportionately affected by systems that require interactions exclusively online.

Impacts on Vulnerable Populations

When we look at the impacts on vulnerable populations, we find that automation without adequate governance can create a phenomenon we call Social Security digital exclusion. Older adults who struggle to access technology, people with disabilities who depend on accessibility features that are not always available on INSS platforms, and communities in remote areas without adequate connectivity are left at a significant disadvantage when processes move to exclusively digital and automated environments.

Algorithmic governance must build in specific safeguards for these populations, making sure that the INSS’s technological modernization does not widen inequalities in access to Social Security benefits. This includes keeping in-person service channels, implementing accessible interfaces, and creating assistance mechanisms for insured persons who cannot adequately interact with digital systems.

Regulatory Framework and Outlook for the Future

On the regulatory front, Brazil has made progress in building a legal framework for artificial intelligence, although significant gaps remain with respect to the use of algorithms in Social Security decisions in particular. The discussion on regulating artificial intelligence in the National Congress covers, among other points, the need for algorithmic impact assessments for high-risk systems, a category that includes INSS decision systems because they directly affect fundamental rights.

The Brazilian Artificial Intelligence Strategy includes principles such as human-centeredness, non-discrimination, and transparency. Although general, these principles provide the foundation for more specific rules governing the use of algorithms in Social Security public administration. We believe that sector-specific regulation aimed at the INSS is just as important as the general framework, because it can address the particularities of the Brazilian Social Security system.

Among the measures we consider essential for effective algorithmic governance at the INSS are a permanent algorithmic ethics committee, mandatory impact reports before new automated systems are deployed, periodic independent audits, and a guarantee that every algorithmic decision can be reviewed by a qualified human employee. It is also essential that insured persons have access to clear, understandable explanations of how algorithms influenced the decisions on their benefits.

The Role of Civil Society and Legal Professionals

Building robust algorithmic governance at the INSS is not a task for the government alone. Lawyers, public defenders, civil society organizations, and professional associations play a key role in monitoring automated systems and defending the rights of insured persons affected by algorithmic decisions. Their work helps identify flaws, propose improvements, and ensure that the voices of insured persons are heard as algorithms are developed and reviewed.

Practitioners of Social Security law increasingly need to understand the technology behind the systems the INSS uses in order to provide effective legal assistance. This does not mean that lawyers must become programmers. It means they must develop the ability to question the logic of automated decisions, request technical explanations, and, when necessary, show the courts that an algorithm produced an unfair or discriminatory result.

International Best Practices and Lessons for Brazil

Looking at international experience, we find relevant examples that can guide the implementation of algorithmic governance at the INSS. European countries, under the influence of the General Data Protection Regulation (GDPR) and the European Union’s AI Act, have developed specific frameworks for the use of artificial intelligence in public services. These initiatives stress the need for risk assessments proportionate to the impact of the automated decision on citizens’ rights.

In the Netherlands, for example, a scandal involving an algorithmic system for detecting fraud in social benefits showed how the lack of proper governance can have devastating consequences. The system flagged thousands of families as fraudsters based on biased criteria, leading to improper debt collection and situations of extreme social vulnerability. The case caused a significant political crisis and reinforced the need for effective human oversight of algorithmic decisions in social protection systems.

For the Brazilian INSS, these international experiences underscore the importance of taking a preventive approach. Investing in algorithmic governance before serious problems emerge is significantly more effective (and less costly, both financially and socially) than trying to repair the damage after millions of insured persons have already been affected by mistaken automated decisions. Public algorithm registries, in which the INSS would document the systems in use and their respective impacts, would be a concrete step in that direction.

This topic is directly related to how artificial intelligence, a subject we cover in a dedicated article.

To round out your reading, see also the firm’s article on Meu INSS and.

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