What happened
Brazil's national film agency, Ancine, had a backlog of more than 4,000 unanalyzed accountability reports for audiovisual funding. In 2024 it launched Projeto Malha Fina with the Controladoria-Geral da União (CGU). The CGU trained a LightGBM model on 1,300 judged reports to flag signs of loss to the Treasury, with accuracy of 0.82. The triage also uses value bands and five automated checks. Projects above R$2.5 million still wait for manual review. Projects of R$1.5 to 2.5 million pass the checks and then the model. Projects below R$1.5 million, half of all projects, are approved on the checks alone. In a ruling of 25 March 2026, the Federal Court of Accounts (TCU) reported that Ancine analyzed 3,634 reports in 2024, against 2,400 from 2008 to 2023. In the TCU's sample, 200 of 349 cases went through the method. The TCU judged the project a good practice. It relays Ancine's own calculation of a R$665 million saving. It also found that 761 of the 1,752 remaining cases may be time-barred.
Sources
-
Audit report, by authors with no tie to the system.
Tribunal de Contas da União. Acórdão 726/2026 – TCU – Plenário (TC 014.873/2025-7), relator Ministro Augusto Nardes, session of 25 March 2026. Follow-up report on the management of accountability reports and special audits by the transferring bodies (Ancine and Ministério da Cultura), paragraphs 262-285 and 358-361.
Para os processos da amostra, constatou-se que 200 dos 349 processos selecionados foram analisados com o uso da metodologia, representando 57,3% dos processos. Sendo assim, observa-se que a Ancine conseguiu reduzir mais da metade do seu passivo aplicando o modelo
Details
AI system
- Workflow
-
The CGU trained the model on 16 fields from 1,300 accountability reports that had already been judged. Value bands set which projects went to manual review, to the model or to the automated checks alone. Five automated checks sent any project with signs of irregularity to a person.
Money
- Kind of value
-
Saving from analyzing accountability reports through Projeto Malha Fina, by Ancine's own calculation
- Amount
-
Estimate: R$ 665.000.000,00
- Recipients and payers
-
Ancine
Follow-up
- Conditions for replication
-
Projects above R$2.5 million still need manual review. Human review must be prioritized by the risk of time-barring, the amount at stake and the model's flags.
Entry details
- Sector
-
Public administration: grant accountability review.
- Purpose
-
The Controladoria-Geral da União built the Malha Fina model with Ancine data to flag accountability reports on film funding that show signs of loss to the Treasury.
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, currency and prices, period and unit, realization, gross or net, included costs, comparison, method, return and payback, 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). Malha Fina helps Ancine clear a backlog of grant reviews (AIBD-0707). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0707.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)