The Brazilian Federal Court of Accounts (TCU) was found by the Organization for Economic Co-operation and Development (OECD) to be a leading institution in the use of artificial intelligence (AI). After surveying 59 organizations from 39 countries, the OECD found the Court to be the sole institution to have an advanced-level use of generative AI, through the development of its ChatTCU. The work assessed the opportunities and challenges related to the use of AI by government entities, including anticorruption agencies and Supreme Audit Institutions (SAIs).

According to the OECD, the development of ChatTCU illustrates how this type of technology can be used to enhance auditors' efficiency in collecting and reviewing documentation. The LLMs (Large Language Models) are machine learning models trained to learn and generate information from large databases.

The publication outlines ChatTCU's model and presents the main lessons learned by the Court of Accounts with the development of the platform. The OECD concluded that the Brazilian tool serves as an example of how generative AI can be used in different types of organizations, including supreme audit institutions, and in several countries.

Generative artificial intelligence refers to systems that can create information, rather than just analyzing or reproducing existing data. The use of technology revolutionizes internal practices and strengthens external control work. According to the OECD's report, ChatTCU "allows auditors to request a summary of a case document, ask technical questions related to the TCU's work and court rulings and seek help for administrative services".

The organizations interviewed by the OECD assessed the value of generative AI to investigative and auditing processes. Gathering evidence and reviewing documents were identified as the most valuable functions. In total, 37% of interviewees (22) ranked these activities as their first choice. In the second place, the tool’s primary purpose for 15 organizations would be selecting audits and investigations. Other mentioned functions included elaborating reports and designing graphs, planning audits and investigations, generating content for public relations; and documenting processes.

Explore the OECD's evaluation of ChatTCU step by step:

Internalization of technology: The proactive development of ChatTCU, adapted to the TCU's needs, suggests that integrity actors could consider building their own AI solutions rather than relying solely on external tools. This also helps build digital literacy and capacity within the organization, which will be valuable in other areas.

Integration with existing systems: ChatTCU's integration with existing TCU systems allowed auditors to access administrative information and obtain insights on audits more efficiently, highlighting the importance of perfect integration with flowswork and existing systems.

Scalability and preparing for the future: The TCU's plans to expand ChatTCU’s features demonstrate the need for scalability and adaptability in today's AI solutions, encouraging stakeholders to envision future with integrity.

Potential for standardization: The TCU's consideration to incorporate audit standards into ChatTCU indicates the potential of AI tools to assist in maintaining standards, suggesting that other integrity actors may explore similar possibilities to improve their processes.

Feedback-oriented development: The TCU highlighted the importance of collecting user feedback to continuously improve AI solutions and emphasized the need for integrity actors to create mechanisms enabling staff to provide suggestions for improvement.

Multidisciplinary approach: The TCU estabilished a multidisciplinary working group to assess the risks and opportunities of generative AI use, involving representatives of several areas and promoting debates to inform decisions about the implementation of AI.

Investment in training and awareness: The TCU's emphasis on raising awareness among employees about the potential and risks of AI highlights the need to train and equip team members on how to effectively use AI technologies. Involving the team in developing AI solutions internally will also help them learn how to face challenges.

What is ChatTCU?

Launched in February 2023, ChatTCU is a virtual assistant designed to be internally used by the Court’s staff, developed with Microsoft Azure OpenAI Service. Currently, it has more than 1,400 users. The Chat can be used as a supporting tool for several tasks, such as document analysis, legal research, translation, and administrative inquiries.

In its third version, the tool is integrated with different TCU systems and uses data, precedents, and information about the Court’s administrative systems. The platform was developed following the creation of a multidisciplinary working group and intense discussions. Based on natural language processing, ChatTCU operates in Microsoft's cloud, under a contract that guarantees security and confidentiality. The information is kept under the TCU's domain and goes through layers of protection and privacy.

In 2023, the TCU has made significant progress in innovative initiatives and has become  one of the first public entities in Brazil to provide public agents and collaborators with generative AI  connected to various internal systems.