{"id":"AIBD-0706","title":"Brazil's Federal Court of Accounts finds that Galileu's drafts cover 93.3% of claims in labor cases, with hallucinations in 7.3% of topics","summary":"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.","date_observed":"2026-06-10","date_basis":"event","date_precision":"day","benefit_status":"reported","source_collection":null,"source_entry":null,"account":"The Regional Labor Court of the 4th Region (TRT-4) in Rio Grande do Sul built Galileu, a generative AI tool on Google Gemini Flash models. It reads case filings and drafts the report section of judgments. By 10 June 2026, 4,904 users had produced 173,041 drafts, and about 40% of drafts of first-instance judgments at TRT-4 used it. The Federal Court of Accounts (TCU) audited the system and ruled on 14 October 2026. Auditors ran the code and the production models on 50 real, non-confidential cases with 383 topics. Two AIs built and checked a reference of claims and facts, and auditors confirmed each suspected hallucination. Drafts covered 93.3% of claims, 81.3% of facts and 86.1% of themes. Hallucinations appeared in 7.3% of topics, none of them of high severity. Because each case has many topics, 42% of cases had at least one. In paired tests, the auditors found no statistically significant bias by the party's gender or occupation. The TCU concluded that Galileu is reliable in its central dimensions. It did not measure time saved.","category":"aggregate result","unexpected":false,"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.","system":"Galileu (generative AI drafting assistant on Google Gemini Flash models)","model_version":null,"developer":"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)","deployer":"TRT-4 (Rio Grande do Sul), with a national rollout to the Labor Justice coordinated by the CSJT","location":"Rio Grande do Sul, Brazil","domain":"Justice administration: labor court drafting","segment":null,"benefit_type":"Repetitive Task Automation","work_activities":["4.A.2.a.2","4.A.3.b.6"],"taxonomy_note":null,"registry_ids":{},"ratings":[{"group":"organizational","depth":1,"reason":"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."}],"ai_role":{"status":"unassessed","statement":"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."},"evidence":[{"kind":"audit report","citation":"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.","title":"Acórdão 2836/2026 – TCU – Plenário: auditoria operacional no Tribunal Regional do Trabalho da 4ª Região, sistema Galileu","url":"https://contas.tcu.gov.br/sagas/SvlVisualizarRelVotoAc?codFiltro=SAGAS-SESSAO-ENCERRADA&seOcultaPagina=S&item0=919995","date":"2026-10-14","written_by":"independent","quote":"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)"}],"replication":{"produced_by":["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."],"must_hold":["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."],"echoes":[]},"paired_incident":null,"headline":"Brazil's audit court finds *Galileu* drafts reliable in labor cases","metric":"93.3% of the parties' claims","comparator":null,"later":null,"value":null,"developer_country":"Brazil","deployer_country":"Brazil","unrated_because":null,"brief_notes":null,"brief":null,"revisions":[{"date":"2026-10-31","change":"Added the record from the finding of the Tribunal de Contas da União.","source":"https://contas.tcu.gov.br/sagas/SvlVisualizarRelVotoAc?codFiltro=SAGAS-SESSAO-ENCERRADA&seOcultaPagina=S&item0=919995","editor":"AIBD"}],"confidence":"found","shade":"high","claim_status":"Found by an auditor","claim_basis":"Sources: one independent audit report. The cited evidence includes no independent study or report.","case_jurisdictions":["Brazil"],"profile":{"case_id":"AIBD-0706","catalog_version":"1.0.0","source_fingerprint":"925e9e1c48db73fa61ad6f83b184571b2e962f32bddb5388a448fd651f590f76","collection_refs":[{"collection_id":"aibd","version":"2026-10-31"},{"collection_id":"querist","version":"2026-10-25"}],"source_ids":["AIBD-0706-source-0001"],"families":{"strategic assessments":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"Editorial assignment.","review_state":"not_assessed","coverage":"partial","record_ids":[]},"evaluations":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"Editorial assignment.","review_state":"not_assessed","coverage":"partial","record_ids":[]},"monetary observations":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"Editorial assignment.","review_state":"not_assessed","coverage":"partial","record_ids":[]},"institutional decisions":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"Editorial assignment.","review_state":"not_assessed","coverage":"partial","record_ids":[]}},"fields":{"actors[]":{"state":"value_present","value":{"kind":"record_fields","fields":{"developer":"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)","deployer":"TRT-4 (Rio Grande do Sul), with a national rollout to the Labor Justice coordinated by the CSJT"}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["developer","deployer"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"actors[].ownership_control":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"ai_contribution":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"ai_relationship":{"state":"value_present","value":{"kind":"record_fields","fields":{"category":"aggregate result","ai_role":{"status":"unassessed","statement":"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."}}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["ai_role","category"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"assessments[].alternatives":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"assessments[].assessor_date":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"assessments[].depth":{"state":"value_present","value":{"kind":"record_fields","fields":{"ratings":[{"group":"organizational","depth":1,"reason":"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."}]}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["ratings"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"assessments[].display_confidence":{"state":"value_present","value":{"kind":"structured","data":{"confidence":"high","evidence_word":"found","claim_status":"Found by an auditor","basis":"Sources: one independent audit report. The cited evidence includes no independent study or report.","interpretation":"Existing display rule preserved; the richer outcome-specific confidence question still needs review."}},"provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"assessments[].rationale":{"state":"value_present","value":{"kind":"record_fields","fields":{"ratings":[{"group":"organizational","depth":1,"reason":"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."}]}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["ratings"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"assessments[].rubric_version":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"assessments[].status_basis":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"assessments[].target":{"state":"value_present","value":{"kind":"record_fields","fields":{"ratings":[{"group":"organizational","depth":1,"reason":"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."}]}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["ratings"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"availability":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"brief":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"case_kind":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"case_links[]":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"claims[].evidence_links":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"claims[].evidence_strength":{"state":"value_present","value":{"kind":"structured","data":{"answer":"Found by an auditor. Sources: one independent audit report. The cited evidence includes no independent study or report.","evidence_word":"found","claim_status":"Found by an auditor"}},"provenance":{"kind":"source_review","basis":"Editorial assignment.","record_fields":["evidence"],"source_refs":["AIBD-0706-source-0001"],"reviewed_on":"2026-10-29"},"review_state":"reviewed","coverage":"complete","display":{"answer":"Found by an auditor. Sources: one independent audit report. The cited evidence includes no independent study or report.","inferred":false}},"claims[].missingness":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"claims[].proposition":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"cross_border_flows[]":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"decisions[].action_outcome":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"institutional decisions"},"decisions[].grounds":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"institutional decisions"},"decisions[].institution_task":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"institutional decisions"},"decisions[].success_criterion":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"institutional decisions"},"deployment_stage":{"state":"value_present","value":{"kind":"structured","data":{"classification":"routine use","answer":"The system was in routine use. About 40% of drafts of first-instance judgments at TRT-4 use Galileu, and the Labor Justice is rolling it out nationally.","scope":"The period and activity described in the case."}},"provenance":{"kind":"source_review","basis":"Editorial assignment.","record_fields":["account","evidence"],"source_refs":["AIBD-0706-source-0001"],"reviewed_on":"2026-10-29"},"review_state":"reviewed","coverage":"complete","display":{"answer":"The system was in routine use. About 40% of drafts of first-instance judgments at TRT-4 use Galileu, and the Labor Justice is rolling it out nationally.","inferred":false}},"economics[].amount":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"monetary observations"},"economics[].costs[]":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"monetary observations"},"economics[].counterfactual":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"monetary observations"},"economics[].currency_price_basis":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"monetary observations"},"economics[].gross_net":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"monetary observations"},"economics[].measure_type":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"monetary observations"},"economics[].period_unit":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"monetary observations"},"economics[].realization":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"monetary observations"},"economics[].recipient_payer":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"monetary observations"},"economics[].return_payback":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"monetary observations"},"economics[].valuation_method":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"monetary observations"},"evaluations[].ai_uplift":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"evaluations"},"evaluations[].comparator":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"evaluations"},"evaluations[].design":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"evaluations"},"evaluations[].effect":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"evaluations"},"evaluations[].exposure":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"evaluations"},"evaluations[].id":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"evaluations"},"evaluations[].measurement_method":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"evaluations"},"evaluations[].metric":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"evaluations"},"evaluations[].reproduction_materials":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"evaluations"},"evaluations[].setting_sample":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"evaluations"},"evaluations[].uncertainty_limitations":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"evaluations"},"external_references[]":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"follow_up[]":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"human_oversight":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"id":{"state":"value_present","value":{"kind":"record_fields","fields":{"id":"AIBD-0706"}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["id"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"infrastructure_dependencies[]":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"outcomes[].beneficiaries":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"outcomes[].beneficiary_count":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"outcomes[].benefit_type":{"state":"value_present","value":{"kind":"record_fields","fields":{"benefit_type":"Repetitive Task Automation"}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["benefit_type"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"outcomes[].claim":{"state":"value_present","value":{"kind":"record_fields","fields":{"account":"The Regional Labor Court of the 4th Region (TRT-4) in Rio Grande do Sul built Galileu, a generative AI tool on Google Gemini Flash models. It reads case filings and drafts the report section of judgments. By 10 June 2026, 4,904 users had produced 173,041 drafts, and about 40% of drafts of first-instance judgments at TRT-4 used it. The Federal Court of Accounts (TCU) audited the system and ruled on 14 October 2026. Auditors ran the code and the production models on 50 real, non-confidential cases with 383 topics. Two AIs built and checked a reference of claims and facts, and auditors confirmed each suspected hallucination. Drafts covered 93.3% of claims, 81.3% of facts and 86.1% of themes. Hallucinations appeared in 7.3% of topics, none of them of high severity. Because each case has many topics, 42% of cases had at least one. In paired tests, the auditors found no statistically significant bias by the party's gender or occupation. The TCU concluded that Galileu is reliable in its central dimensions. It did not measure time saved.","benefit_type":"Repetitive Task Automation"}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["account","benefit_type"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"outcomes[].distribution":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"outcomes[].duration":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"outcomes[].id":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"outcomes[].stakeholder_group":{"state":"value_present","value":{"kind":"record_fields","fields":{"ratings":[{"group":"organizational","depth":1,"reason":"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."}]}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["ratings"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"outcomes[].status":{"state":"value_present","value":{"kind":"record_fields","fields":{"benefit_status":"reported"}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["benefit_status"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"outcomes[].tradeoffs":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"outcomes[].unexpected":{"state":"value_present","value":{"kind":"record_fields","fields":{"unexpected":false,"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."}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["unexpected","purpose"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"places[]":{"state":"value_present","value":{"kind":"record_fields","fields":{"location":"Rio Grande do Sul, Brazil","developer_country":"Brazil","deployer_country":"Brazil"}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["location","developer_country","deployer_country"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"purpose":{"state":"value_present","value":{"kind":"record_fields","fields":{"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."}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["purpose"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"record_dates":{"state":"value_present","value":{"kind":"record_fields","fields":{"date_observed":"2026-06-10","date_basis":"event","date_precision":"day","revisions":[{"date":"2026-10-31","change":"Added the record from the finding of the Tribunal de Contas da União.","source":"https://contas.tcu.gov.br/sagas/SvlVisualizarRelVotoAc?codFiltro=SAGAS-SESSAO-ENCERRADA&seOcultaPagina=S&item0=919995","editor":"AIBD"}]}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["date_observed","date_basis","date_precision","revisions"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"replication.attempts[]":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"replication.conditions":{"state":"value_present","value":{"kind":"record_fields","fields":{"replication":{"produced_by":["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."],"must_hold":["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."],"echoes":[]}}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["replication"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"revision_history[]":{"state":"value_present","value":{"kind":"record_fields","fields":{"revisions":[{"date":"2026-10-31","change":"Added the record from the finding of the Tribunal de Contas da União.","source":"https://contas.tcu.gov.br/sagas/SvlVisualizarRelVotoAc?codFiltro=SAGAS-SESSAO-ENCERRADA&seOcultaPagina=S&item0=919995","editor":"AIBD"}]}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["revisions"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"sector":{"state":"value_present","value":{"kind":"record_fields","fields":{"domain":"Justice administration: labor court drafting"}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["domain"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"sources[].access_version":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"sources[].citation_locator":{"state":"value_present","value":{"kind":"record_fields","fields":{"evidence":[{"kind":"audit report","citation":"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.","title":"Acórdão 2836/2026 – TCU – Plenário: auditoria operacional no Tribunal Regional do Trabalho da 4ª Região, sistema Galileu","url":"https://contas.tcu.gov.br/sagas/SvlVisualizarRelVotoAc?codFiltro=SAGAS-SESSAO-ENCERRADA&seOcultaPagina=S&item0=919995","date":"2026-10-14","written_by":"independent","quote":"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)"}]}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["evidence"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"sources[].id":{"state":"value_present","value":{"kind":"identifiers","identifiers":["AIBD-0706-source-0001"],"role":"Existing evidence items, in their original order"},"provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"review_state":"record_mapped","coverage":"complete"},"sources[].language":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"sources[].relationship":{"state":"value_present","value":{"kind":"record_fields","fields":{"evidence":[{"kind":"audit report","citation":"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.","title":"Acórdão 2836/2026 – TCU – Plenário: auditoria operacional no Tribunal Regional do Trabalho da 4ª Região, sistema Galileu","url":"https://contas.tcu.gov.br/sagas/SvlVisualizarRelVotoAc?codFiltro=SAGAS-SESSAO-ENCERRADA&seOcultaPagina=S&item0=919995","date":"2026-10-14","written_by":"independent","quote":"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)"}]}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["evidence"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"sources[].upstream_sources":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"strategic_assessments[]":{"state":"unresolved","review_state":"not_assessed","coverage":"partial","family":"strategic assessments"},"summary":{"state":"value_present","value":{"kind":"record_fields","fields":{"summary":"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.","account":"The Regional Labor Court of the 4th Region (TRT-4) in Rio Grande do Sul built Galileu, a generative AI tool on Google Gemini Flash models. It reads case filings and drafts the report section of judgments. By 10 June 2026, 4,904 users had produced 173,041 drafts, and about 40% of drafts of first-instance judgments at TRT-4 used it. The Federal Court of Accounts (TCU) audited the system and ruled on 14 October 2026. Auditors ran the code and the production models on 50 real, non-confidential cases with 383 topics. Two AIs built and checked a reference of claims and facts, and auditors confirmed each suspected hallucination. Drafts covered 93.3% of claims, 81.3% of facts and 86.1% of themes. Hallucinations appeared in 7.3% of topics, none of them of high severity. Because each case has many topics, 42% of cases had at least one. In paired tests, the auditors found no statistically significant bias by the party's gender or occupation. The TCU concluded that Galileu is reliable in its central dimensions. It did not measure time saved."}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["summary","account"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"systems[].autonomy_permissions":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"systems[].component_role":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"systems[].deployment_configuration":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"systems[].inputs_outputs":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"systems[].model_name":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"systems[].model_version":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"systems[].product_name":{"state":"value_present","value":{"kind":"record_fields","fields":{"system":"Galileu (generative AI drafting assistant on Google Gemini Flash models)"}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["system"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"systems[].provider":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"taxonomy_crosswalks[]":{"state":"unresolved","provenance":{"kind":"curatorial","basis":"Editorial assignment.","record_fields":[],"source_refs":[]},"reason":"This is an explicit unresolved assignment: the sources on record do not settle this question.","review_state":"not_assessed","coverage":"partial"},"title":{"state":"value_present","value":{"kind":"record_fields","fields":{"title":"Brazil's Federal Court of Accounts finds that Galileu's drafts cover 93.3% of claims in labor cases, with hallucinations in 7.3% of topics","headline":"Brazil's audit court finds *Galileu* drafts reliable in labor cases"}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["title","headline"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"},"version":{"state":"value_present","value":{"kind":"identifiers","identifiers":["23d083a0491f1132a39493f1987f04bb7a8735a5d986331994788b05483a24ba"],"role":"SHA-256 of the complete public record and its assessments"},"provenance":{"kind":"curatorial","basis":"SHA-256 of the public record, its assessments, and its displayed answers and notes, excluding its own identifier and the derived coverage summary.","record_fields":[],"source_refs":[]},"review_state":"record_mapped","coverage":"complete"},"workflow":{"state":"value_present","value":{"kind":"record_fields","fields":{"account":"The Regional Labor Court of the 4th Region (TRT-4) in Rio Grande do Sul built Galileu, a generative AI tool on Google Gemini Flash models. It reads case filings and drafts the report section of judgments. By 10 June 2026, 4,904 users had produced 173,041 drafts, and about 40% of drafts of first-instance judgments at TRT-4 used it. The Federal Court of Accounts (TCU) audited the system and ruled on 14 October 2026. Auditors ran the code and the production models on 50 real, non-confidential cases with 383 topics. Two AIs built and checked a reference of claims and facts, and auditors confirmed each suspected hallucination. Drafts covered 93.3% of claims, 81.3% of facts and 86.1% of themes. Hallucinations appeared in 7.3% of topics, none of them of high severity. Because each case has many topics, 42% of cases had at least one. In paired tests, the auditors found no statistically significant bias by the party's gender or occupation. The TCU concluded that Galileu is reliable in its central dimensions. It did not measure time saved.","replication":{"produced_by":["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."],"must_hold":["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."],"echoes":[]}}},"provenance":{"kind":"legacy_record","basis":"Editorial assignment.","record_fields":["account","replication"],"source_refs":[]},"review_state":"record_mapped","coverage":"partial"}},"summary":{"total":91,"addressed":91,"known":28,"reviewed":2,"reviewed_complete":2,"reviewed_partial":0,"record_mapped":26,"review_not_assessed":63,"not_assessed":0,"complete":false,"by_state":{"value_present":28,"not_reported":0,"unresolved":63,"not_assessed":0,"not_applicable":0,"withheld":0,"source_unavailable":0}}},"presentation_version":"cc940f51e801e23e279ec5d957781076757ee2cad59ecdb29d0549eeef8135fc"}