Brazil's audit court finds Galileu drafts reliable in labor cases

Brazil's Federal Court of Accounts tested the Galileu judgment-drafting tool on 50 real cases. Its drafts covered 93.3% of the parties' claims, with hallucinations in 7.3% of topics.

  • 1 an organization

    About 40% of drafts of first-instance judgments at TRT-4 come from Galileu, and the TCU found that they cover 93.3% of claims, though it did not measure time saved.

Depth runs from 1, a convenience or time saved, to 5, a life or a livelihood preserved.

Status
Reported benefit.
Evidence

foundstudiedreportedclaimed

An independent audit report supports the benefit.

The system’s part
Not isolated. The TCU tested the drafts against a reference built from the filings, which shows what the model produces, but it did not compare drafting with and without Galileu. Galileu runs on large language models, and the judge reviews and signs every draft.
Limits
Not yet checked.
Costs and harms
Not yet checked.
Later
Not yet checked.

Download data Cite Suggest a correction

Benefit type
Repetitive Task Automation
System
Galileu (generative AI drafting assistant on Google Gemini Flash models)
Made by
Tribunal Regional do Trabalho da 4ª Região (TRT-4), using Google Gemini models through the cloud contract of the Conselho Superior da Justiça do Trabalho (CSJT)
Used by
TRT-4 (Rio Grande do Sul), with a national rollout to the Labor Justice coordinated by the CSJT
Where
Rio Grande do Sul, Brazil
Stage
In routine use
Date
Kind
Aggregate result
Broad benefit
Processing Information, Documenting/Recording Information
Collections
Scrutiny and findings
Last revised
31 October 2026

Sources

  1. Audit report, by authors with no tie to the system.

    Tribunal de Contas da União. Acórdão 2836/2026 – TCU – Plenário (TC 001.458/2026-4), relator Ministro Odair Cunha, session of 14 October 2026. Operational audit of the Galileu system at TRT-4, report by the Unidade de Auditoria Especializada em Tecnologia da Informação (AudTI), items 2.2, 2.6 and 155.

    Os atributos de IA responsável estão presentes de forma satisfatória. Não há evidência estatística de viés de gênero ou de ocupação da parte, o sistema cobre quase todos os pedidos, com alucinação em 7,3% dos tópicos (nenhuma de severidade alta)

Details

AI system

Workflow

Galileu extracted the filings from the electronic case system, removed hidden text and screened for prompt injection. It replaced the parties' personal data with labels before sending text to the model. The judge reviewed, edited and signed each draft.

Follow-up

Conditions for replication

Judges and their staff must review drafts closely, since 42% of sampled cases had at least one hallucination. Defenses against prompt injection must be tested as the system changes.

Entry details

Sector

Justice administration: labor court drafting.

Purpose

TRT-4 built Galileu to read the filings in a labor case and draft the report section of a judgment for the judge to review. It also drafts appeal judgments and admissibility decisions on appeals to the Superior Labor Court.

Not yet checked: model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.

Cite this record

Peter H. Vartanian. (2026). Brazil's audit court finds Galileu drafts reliable in labor cases (AIBD-0706). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0706.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)

BibTeX Data Field definitions