CRISP-LLM-Gov_TCs_survey_dataset_2025
Description
This dataset contains the institutional responses of Brazilian Courts of Accounts (Tribunais de Contas) to a structured survey on the adoption and governance of generative artificial intelligence (GenAI), specifically large language models (LLMs), in the drafting of procedural documents such as draft votes, audit reports, and ministerial opinions. The research universe comprises all 33 Courts of Accounts in Brazil. Data were collected through formal requests under the Brazilian Access to Information Law (nº 12.527/2011), via the electronic Citizen Information Service (e-SIC), between April and May 2025. Of the 33 institutions, 31 responded, a 93.9% response rate, characterizing a near-census survey. The instrument comprised 10 closed questions with optional open fields, organized into three dimensions: technological adoption, practical application, and governance and control. The dataset is composed exclusively of public, non-personal, institutional information about the practices of public bodies, obtained through a legal transparency channel. The original responses are in Portuguese; column descriptions are provided in English. This dataset is the empirical basis of the article "Governing generative AI in Brazilian Courts of Accounts: a near-census survey of LLM use and the CRISP-LLM Gov adaptive governance framework", currently under preparation.
Files
Steps to reproduce
Data were collected through formal requests submitted under the Brazilian Access to Information Law (Law No. 12,527/2011) via the electronic Citizen Information Service (e-SIC) to all 33 Brazilian Courts of Accounts between April and May 2025. Responses were manually coded and consolidated into a structured dataset. See the README file for the complete data dictionary and variable descriptions.