ANPD Closes Its First AI Sandbox and Plans a Second One on Children’s Digital Games
Brazil’s National Data Protection Authority (ANPD) has finished testing in its pilot regulatory sandbox on artificial intelligence and data protection and published the seventh and final monitoring report of the first edition, the agency said in a statement on November 5. The programme tested three AI solutions in a controlled, supervised environment with scientific support from the University of São Paulo’s Centre for Artificial Intelligence and Machine Learning (CIAAM/USP). An institutional evaluation report from the sandbox commission, due in December, will formally close the pilot.
The three participants work in health, finance and urban mobility. Prevvine Tecnologia’s STAiDOC, a clinical decision-support system, received a favourable opinion with conditions; the tests underlined that the final decision and clinical responsibility stay with the doctor, and Anvisa representatives said tools of this kind would likely qualify as high-risk software as a medical device. Metatext’s GuardionAI, which protects data flows for companies running autonomous agents, received a favourable technical recommendation. Synapse AI’s Trajetto pedestrian-routing system scored 86 out of 100 points on the UNESCO ethics standard but must show its explainability and privacy screens before any commercial launch.
The second edition will focus on balanced and safe use of digital games by children and teenagers. Announced by Lucas Anjos, the agency’s superintendent of technological innovation, at an event marking six years of the ANPD, it will be run with the United Nations Development Programme (UNDP) to align game developers with the Digital Statute of Children and Adolescents (ECA Digital) and its regulatory decree, No. 12,880/2026. Controlled tests will look at ways to curb excessive or compulsive use, including excessive notifications and rewards tied to screen time.
What it means in practice
The ANPD also published a Regulatory Sandbox Manual that collects the pilot’s lessons and gives practical guidance to teams in other bodies, including for technologies other than AI. The agency said the first edition showed that AI oversight cannot rely on documentary declarations of compliance alone and requires empirical tests that compare a system’s design with its observed behaviour. Its findings will feed new guidance and cooperation with sector regulators such as Anvisa, ANS and ANTT.
The results follow earlier sandbox reports on Metatext, Synapse AI and Prevvine and the agency’s preparations for a second round. What to watch is the December evaluation report, which academic institution and game studios the ANPD selects for the new edition, and how the conclusions shape rules while Congress still debates the AI legal framework.