Victor — Brazil's Supreme Federal Court AI for extraordinary appeal triage

Brazil's Supreme Federal Court (STF) receives more than 50,000 extraordinary appeals a year, each of which must be screened for a 'general repercussion' (repercussao geral) theme that would bind lower courts nationwide -- previously a roughly 44-minute manual task per case for clerks.

Victor, launched in 2018 in partnership with the University of Brasilia (UnB) using neural networks trained on past STF decisions, converts scanned appeal PDFs to text, segments the documents, and suggests a general-repercussion theme in about five seconds; it was fully integrated into the court's case-management platform (STF Digital) by 2020, with an upgraded version (VictorIA) introduced in May 2023. Clerks review and validate every suggestion before a theme is formally applied, keeping a human in the loop.

The documented result is a triage-speed improvement (about 5 seconds versus roughly 44 minutes manually) applied at the scale of the court's full annual caseload (~50,000 appeals/year).

A 2023 review by the Access to Justice Lab found no published statistics on downstream outcomes such as case resolution time, consistency or appeal rates, and enforcement of CNJ Resolution 332/2020's periodic-evaluation requirement for judicial AI systems has remained limited, so Victor's effect on Brazil's roughly 102 million pending cases and on access to justice more broadly remains unquantified.