Brazil's data regulator wraps AI sandbox, plans one for kids' games

Brazil’s National Data Protection Authority (ANPD) closed its first AI regulatory sandbox, publishing its seventh and final report on Thursday after testing three AI products under controlled, supervised conditions with scientific support from the University of São Paulo’s AI and machine-learning center. The same day, the agency opened a second edition focused on the safe, balanced use of digital games by children and teenagers — built with the UN Development Programme and meant to align game studios with the child-protection requirements of Brazil’s Digital Statute for Children and Adolescents and its implementing decree.

ANPD’s technology-innovation superintendent, Lucas Anjos, announced the new sandbox’s focus at an event marking the agency’s sixth anniversary. The process will first pick an academic or scientific institution to help verify design and default-protection practices in electronic games, then let ANPD’s technology-innovation unit select participating game companies for controlled testing of ways to curb excessive, problematic or compulsive play — including excessive notifications and rewards tied to screen time — without undermining the player experience or children’s decision-making autonomy.

The agency also published a regulatory-sandbox manual distilling the pilot’s lessons for any government team running a similar experiment, covering everything from scoping and legal review to candidate selection, testing and final reporting. ANPD says the manual isn’t a binding rule — each sandbox edition’s specific terms are still set by its own public call.

Among the pilot’s three tested products: STAiDOC, a clinical decision-support tool from Prevvine Tecnologia, showed that human doctors must keep final say even over pseudonymized health data; Brazil’s health regulator, Anvisa, flagged it as high-risk medical-device software needing further clearance. GuardionAI, from Guardion (formerly Metatext), won a favorable technical review for shielding financial-sector AI-agent data flows and providing audit trails. And Trajetto, a pedestrian-routing tool from Synapse AI built with Insper’s data and AI center, scored 86 of 100 on a UNESCO ethics benchmark but must get its in-app privacy and explainability screens approved before any commercial launch, after regulators ruled that combining continuous location tracking with accessibility data counts as high-risk processing.

Why it matters

ANPD’s conclusion — that AI oversight needs rigorous empirical testing, not just compliance paperwork — will carry into the next sandbox, extending Brazil’s hands-on approach to regulation into a sector it hasn’t yet tackled: digital games aimed at children.