Brazil's Social Security AI Tool Is Allegedly Rejecting Complex Claims Improperly

Brazil’s AI-powered Meu INSS app allegedly misclassified a retirement claim filed by Josélia de Brito, a former sugarcane worker, by purportedly recording her as male. De Brito reportedly suffers from multiple chronic illnesses and faced repeated denials of social benefits due to minor filing errors. Analysts cited in the reporting suggest the system's design may disadvantage rural and digitally excluded users.

Risk classification

• Primary risk domain: 1 Discrimination & Toxicity
• Primary risk subdomain: 1.3 Unequal performance across groups

The AI system exhibits unequal performance and effectiveness, disproportionately failing to accurately process complex claims from rural and agricultural workers compared to simpler urban claims.

Additional risk subdomains

• 7.3 Lack of capability or robustness: The system fails to reliably process complex documents and makes basic errors like misidentifying a female claimant's gender.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The automated rejections and demographic misclassifications were caused by the deployed Meu INSS AI system operating in production, which was an unintentional outcome of the government's automation efforts.

EU AI Act risk tier

• Risk tier: 2 High Risk

High Risk: The AI system is used by a public authority to automate decisions regarding eligibility for social security and retirement benefits, which has significant implications for fundamental rights and public interests.

AI system and alleged parties

• AI system: Meu INSS AI (Dataprev)
• AI purpose: Application Evaluation; Identification
• Behaviour type: Autonomous
• Alleged developer: Dataprev
• Alleged deployer: Instituto Nacional do Seguro Social (INSS), Government of Brazil
• Alleged harmed parties: Rural workers in Brazil, Digitally excluded populations

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Substantial. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Substantial, indirect Minor
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Minor, indirect Minor
• Civil rights: direct Minor, indirect Minor
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Minor
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Financial

Reported: The report explicitly describes financial losses and delays in receiving vital social security benefits caused directly by the incident.

Directly caused: Vulnerable individuals like Joselia de Brito had their monthly social benefits (such as sick pay of approximately $260/month) denied or delayed.

Indirectly caused: Claimants had to pay for travel, such as four-hour bus rides, to resolve issues or undergo delayed medical examinations.

Inferred additional harm: With 87,700 denied claims across 2023 and 2024, the total inferred financial loss to rural workers is in the millions of dollars, severely impacting their ability to afford basic necessities.

Differential treatment

Reported: The report explicitly describes systematic differential treatment of rural and agricultural workers caused directly by the incident.

Directly caused: The AI system disproportionately rejected claims from rural and agricultural workers due to its inability to handle complex agricultural documentation and their lower digital literacy.

Indirectly caused: This created a systemic barrier where rural workers faced digital exclusion and higher rates of benefit denials compared to urban claimants.

Inferred additional harm: Systemic differential treatment likely affected tens of thousands of rural claimants who were excluded from social safety nets due to the app's design.

Civil rights

Reported: The report explicitly describes violations of social and economic rights caused directly and indirectly by the incident.

Directly caused: The automated denial of legally entitled social security and retirement benefits violates the basic social and economic rights of vulnerable citizens.

Indirectly caused: The lack of accessible human review and long appeal times (averaging 278 days) effectively denied citizens their right to timely administrative recourse.

Inferred additional harm: Tens of thousands of citizens likely had their constitutional rights to social security temporarily or permanently violated due to automated errors.

Psychological

Reported: The report explicitly describes psychological distress and feelings of humiliation caused directly by the automated denials.

Directly caused: The report notes that Joselia de Brito described the repeated denial of her benefits and being misidentified as a man as a 'humiliation'.

Indirectly caused: The stress and anxiety of losing vital social security benefits and navigating a complex digital system likely caused indirect psychological distress to thousands of vulnerable rural workers.

Inferred additional harm: It is likely that thousands of the 87,700 rural workers who had their benefits denied experienced significant stress, anxiety, and feelings of humiliation, though specific numbers are not detailed.

People affected

• Occurrences reported: 1
• People reportedly harmed: 87700
• People reportedly exposed: 87700

Potential causes

Management

• Prioritizing Automation Over Welfare: Governments push for rapid automation without preparing the target population.
• Aggressive Workforce Reductions: Cutting staff to force AI adoption causes system chaos and errors.

Technology

• Algorithmic Identification Errors: AI system misidentified female claimant as male, denying benefits.
• Inability to Process Complex Cases: NLP and CV algorithms cannot interpret nuanced social security claims.
• Inaccurate Generative AI Responses: Chatbot provided vague or inaccurate information to agency staff.

Data Inputs

• Incomplete Government Databases: Rural workers lack sufficient registration data, causing AI rejections.
• Strict Sensitivity to Minor Errors: Algorithms automatically reject claims due to minor document discrepancies.
• Sensitive Data Upload Risk: Lack of guidance led to risks of uploading sensitive personal data to AI.

Human Factors

• Low Digital Literacy in Rural Areas: Vulnerable claimants struggle to navigate complex digital interfaces.
• Inadequate Staff Training: Training video failed to instruct staff on secure AI chatbot usage.

Process and Methods

• Lack of Effective Recourse: Automated rejections force users into lengthy, 278-day appeal processes.
• Insufficient Human Oversight: Complex claims are processed by AI without necessary manual review.
• Rushed Codebase Rewrites: Fast-paced AI-assisted coding risks introducing system-wide software errors.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide consistent, detailed accounts of the Meu INSS app's deployment, its automated decision-making process, and the specific negative impacts on rural workers, supported by statistical data on denied claims and a detailed personal case study.
• Ambiguities identified: The exact technical architecture of the AI model is not fully specified beyond using computer vision and natural language processing.
• Alternative interpretations: None. The reports clearly attribute the wrongful denials to the automated system's limitations.

The Brazilian government's integration of the 'Meu INSS' AI system led to wrongful denials of social security benefits for tens of thousands of rural workers due to document processing errors. While causing significant financial hardship and digital exclusion for vulnerable citizens, the incident remains a domestic administrative and human rights issue with negligible to minor national security implications.

• Overall national security impact: Minor
• Scope: Single nation
• Primary target: Brazil
• Alleged perpetrator: Brazil's National Social Security Institute

Threat characteristics

• Imminence: Long-term. Represents an ongoing systemic administrative issue rather than an active national security crisis.
• Autonomy: Full autonomy. The AI system autonomously analyzes documents and executes binding decisions to deny or approve claims without direct human oversight.
• Novelty: Established threat. Automated government decision-making errors and systemic bias against vulnerable populations are well-documented globally.

Impact by dimension

• Physical security: Negligible. No kinetic threat, physical damage, or critical infrastructure compromise reported.
• Information security: Negligible. No intelligence compromise, classified model theft, or information warfare operations indicated.
• Sovereignty: Minor. Minor impact on government service delivery due to automated welfare processing errors, but does not threaten state authority or constitutional processes.
• Economic security: Negligible. No threat to strategic industries, financial systems, or national technological competitive advantage.
• Societal stability: Minor. Systematic exclusion and wrongful denial of benefits for tens of thousands of rural workers represents a minor societal stability issue, currently managed through administrative appeal processes.