Brazil approves school AI rules barring automated grades

Brazil's reported guidelines draw a boundary between AI assistance and authority over students. UOL says the framework preserves human responsibility for consequential educational decisions, restricts unsupervised use by younger children and introduces risk categories. The report also distinguishes approval from publication: the resolution was expected in the official gazette on November 11. That distinction matters when assessing what institutions must do immediately. ([bol.uol.com.br](https://www.bol.uol.com.br/noticias/2026/11/10/mec-homologa-regras-para-uso-de-ia-na-educacao-e-veta-nota-automatica.amp.htm?utm_source=openai))

My assessment is that the competitive effect will fall more on product design than foundation-model research. Education suppliers may need to make supervision, decision records and role boundaries central to their offerings rather than optional controls. For Race to AGI readers, this is an example of how increasingly capable systems can encounter limits on delegated authority even when their outputs improve. Better performance alone does not settle whether a machine should assign a grade or determine a student's progression. The policy could shape which forms of autonomy are commercially acceptable in a large education market, but it provides no direct evidence that general intelligence is closer or further away. Implementation and the final legal text remain the next checkpoints.