This paper deals with the application of artificial intelligence algorithms in processing legal language to identify a complete set of rules applicable to a given legal theme. In this study, we sought to delimit the regulatory framework that involves the Third Sector, based on the data set on the Brazilian regulation flow (RegBR). From the bibliographic research, machine learning techniques were applied to automate the classification of each sentence within the analyzed normative acts, allowing us to identify to what extent a norm applies to the selected topic. The BERT model with fine-tuning by a Brazilian legal dataset was highly effective, reaching 94% of precision (F1-Score and AUC). The results include a total found of 2,359 rules spread in 611 normative acts on the 1,330,190 sentences distributed in 51 thousand regulations contained in the dataset, demonstrating how the applied techniques can contribute to the improvement of the themes involved.