evidoria

← Back to browse

Good practice Imported

Meu INSS's Automated Benefit Engine — Brazil's AI Denies Over Half of Fully Electronic Retirement Claims, TCU Finds

Brazil · Brasília · See the Brazil profile · See the Brasília profile

Evidence: Observational / pre–post Top 75% 47/100 · Ask Evidence Copilot about this practice

Brazil's INSS uses an automated engine on the Meu INSS platform to grant or deny retirement and benefit claims from database records alone. A 2026 federal audit found it auto-rejected 280,231 of 543,419 fully electronic H1-2025 claims (51.6%) and ordered a 180-day fix.

543419 claims
Fully electronic claims processed, H1 2025 (H1 2025)
280231 claims
Fully electronic claims automatically denied, H1 2025 (H1 2025)
51.6 %
Automatic denial rate for fully electronic claims (H1 2025)
3.1 million claims
INSS pending claims caseload, February 2025 (Feb 2025)
2.2 million claims
INSS pending claims caseload, May 2025 (May 2025)
180 days
Deadline given by TCU to redesign the automatic-denial flow
Meu INSS's Automated Benefit Engine — Brazil's AI Denies Over Half of Fully Electronic Retirement Claims, TCU Finds

Details

Maturity
Established
Promoter
Instituto Nacional do Seguro Social (INSS), Dataprev, Ministério da Previdência Social
Period
2024-2026
Keywords
social security, benefits administration, automated decision-making, algorithmic accountability

Context

Since the mid-2020s, Brazil's national social security institute (INSS) has used an automated concession engine on its Meu INSS portal to grant or deny retirement, disability and other benefit claims. The engine cross-checks the claimant's data against the Cadastro Nacional de Informacoes Sociais (CNIS) database alone, without requesting or weighing any supplementary evidence a claimant may have attached.

Objectives

To speed up processing of retirement, disability and other social-security claims submitted through the Meu INSS platform by granting or denying them automatically from database records, without manual case review.

Activities

The system cross-checks each fully electronic claim against the CNIS database and returns an automatic grant or denial. A Tribunal de Contas da Uniao (TCU) operational audit -- Acordao 1.498/2026-Plenario, process TC 007.094/2025-6 -- reviewed this flow for claims processed entirely electronically in the first half of 2025. Press investigation documented individual claims rejected within minutes, often over a single unchecked box or a rural-work period the system could not interpret from attached documents.

Results

The TCU audit found that of 543,419 claims processed entirely electronically in H1 2025, 280,231 (51.6%) were automatically denied. Separately, INSS's overall pending caseload fell from roughly 3.1 million claims in February 2025 to 2.2 million in May 2025 as automation scaled.

Conclusions

The TCU ruled that the automatic-denial flow creates a 'concrete risk of harm' to claimants, since it gives them no chance to correct database inconsistencies or submit missing evidence before a denial is issued. It ordered INSS, Dataprev and the Ministry of Social Security to redesign the flow within 180 days -- including paying out any uncontested portion of a claim immediately and notifying claimants of unresolved gaps -- and federal courts have separately ordered INSS to reopen specific cases for manual review. The throughput gains are real, but the audit frames them as a direct trade-off against claimants' due-process rights.

Implementation

Indicative cost
Medium (€50k–€500k)
Time to results
Long (> 3 years) — TCU gave INSS, Dataprev and the Ministry of Social Security 180 days to redesign the automatic-denial flow following the 2026 audit.
Staffing & skills
INSS caseworkers/reviewers handling claims and appeals, Dataprev technical team operating the automated engine, Ministerio da Previdencia Social policy oversight

Conditions for success

  • Independent audit oversight (TCU) able to obtain administrative processing data and compel remediation
  • Redesigning the flow to notify claimants of unresolved gaps and pay out uncontested portions immediately, per the TCU order

Common failure modes

  • Automatic denial based solely on a CNIS database cross-check, without requesting supplementary evidence
  • No opportunity for claimants to correct database inconsistencies before an automatic denial is issued
  • TCU found a 'concrete risk of harm' to claimants
  • Individual claims rejected within minutes over a single unchecked box or a misinterpreted rural-work period

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful