ANPD sandbox: Trajetto reaches 86 points
On November 3, 2026, ANPD published the partial reports for cycles 4, 5 and 6 of its regulatory sandbox on artificial intelligence and data protection. In the cycle 6 report, section 4.1, Trajetto, Synapse AI's mobility platform, reached 86 points in the overall ethics and governance assessment. The technical opinion, however, still requires proof before any regular operation.
Tests in three sectors
The pilot project, run by the Agency with the Center for Artificial Intelligence and Machine Learning at the University of São Paulo (CIAAM/USP), followed three solutions: Trajetto, from Synapse AI; STAiDOC, from Prevvine; and GuardionAI, from Metatext. The reports record progress in location anonymization, warnings about the limits of clinical recommendations and explanations of anti-fraud blocks.
The score went up; the opinion kept conditions
In Trajetto's overall ethics and governance assessment, the score went from 78 to 86 points between cycles 5 and 6. The report classifies the experimental closure as “sufficient with conditions” (cycle 6, section 4.1).
The report rates the inherent risk (before mitigation measures) as high and the aggregate residual risk as medium. It also keeps high individual risks tied to coordinate manipulation and to the possibility of re-identification through accessibility attributes (cycle 6, section 4.1). In ANPD's recommendation, approving the closure of the experimental cycle “does not represent authorization for regular operation” (section 4.1).
The three solutions had different outcomes
In cycle 6, STAiDOC finished experimentation with corrections and retests still needed to verify the effectiveness of its safeguards. For GuardionAI, ANPD recommended a satisfactory closure of its participation after the five conditions in the opinion are met, with recommendations for continued improvement. The summary and details of the three outcomes are in sections 4.1, 4.2 and 4.3 of the cycle 6 report. The documents assess the pilot's systems and tests; they don't compare every product available on the market.
What to check before expanding a pilot
For managers, the case shows that a high score, on its own, doesn't replace evidence of controls in the experimental environment. For Trajetto, the report points to two concrete gaps: demonstrating warnings and explanations in the interface and proving resistance to coordinate manipulation under real conditions (cycle 6, section 4.1).
Before expanding a pilot of your own, confirm which version was tested, in what environment, which controls were demonstrated, which risks remain and who can pause the system if a safeguard fails.
The analysis considers processes, data maturity, feasibility, risks and return.
Sources
ANPD, 3 Nov 2026. https://www.gov.br/anpd/pt-br/assuntos/noticias/anpd-publica-os-tres-ultimos-relatorios-parciais-de-monitoramento-do-sandbox-em-ia-e-protecao-de-dados
ANPD/CIAAM-USP, 2026. https://www.gov.br/anpd/pt-br/assuntos/projetos-acoes-iniciativas/sandbox/sandbox-relatorio-parcial-de-monitoramento-ciclo-4-v1-0.pdf/@@display-file/file
ANPD/CIAAM-USP, 2026. https://www.gov.br/anpd/pt-br/assuntos/projetos-acoes-iniciativas/sandbox/sandbox-relatorio-parcial-de-monitoramento-ciclo-5-v1-0.pdf/@@display-file/file
ANPD/CIAAM-USP, 2026. https://www.gov.br/anpd/pt-br/assuntos/projetos-acoes-iniciativas/sandbox/sandbox-relatorio-parcial-de-monitoramento-ciclo-6-v1-0.pdf/@@display-file/file