Seinc and Setid Promote International Debate on the Use of AI in Auditing

On April 28, representatives from the Department of External Control for Strategic Information and Innovation (Seinc) and the Department for Information Technology and Digital Evolution (Setid) shared the Brazilian Federal Court of Accounts' expertise in leveraging advanced data analytics and artificial intelligence (AI) technologies to improve government auditing. Supreme Audit Institutions (SAIs) from Germany, the Netherlands, the United Kingdom, Norway, and Finland took part in the event. 

As part of the monthly deep dive meeting series — organized on a rotating basis among the signatories of the Memorandum of Understanding on Data Analytics (MoU Data Analytics) — the session was co-hosted by the Brazilian Federal Court of Accounts (TCU) and the United Kingdom's National Audit Office (NAO). The goal was to exchange experiences and explore the use of large language models (LLMs) and data infrastructures in support of audit processes. The initiative reflects the ongoing commitment of participating SAIs to leveraging emerging technologies to enhance public sector auditing. 

LLMs in Practice

Cláudio Pimentel, auditor from the Audit Department for the Integration of Data, Methods and Technologies for Oversight (AudDigital/Seinc), opened the session by introducing the Collaborative Artificial Intelligence Network for Oversight (Riac). He then presented preliminary results from the internal AI mapping conducted by the General Secretariat for Government Audit (Segecex).

Aloísio Dourado Neto, Deputy Chief Auditor of AudDigital, presented the progress made with LabContas Assist, the virtual assistant launched by the TCU to facilitate auditors’ use of the Information Laboratory for Oversight (LabContas). Neto highlighted the importance of innovation in the public sector: “We are reshaping the way we audit, using cutting-edge technology to ensure greater efficiency and transparency.” 

Eric Hans, Head of the Artificial Intelligence Unit (NIA/Setid), discussed recent developments with ChatTCU, a tool powered by large language models (LLMs) designed to answer questions and facilitate access to information for TCU staff, contractors, and interns. “The solutions developed by the Court not only optimize the auditors' work, but also strengthen the citizens' trust in oversight processes,” Hans noted. 

Marcio Stern da Fonseca, auditor at the Audit Department for Special Rendering of Accounts (AudTCE), spoke about AI assistants integrated into the audit process. André Feitoza, from the Division of Digital Strategy Management for Government Auditing (Diged/AudDigital), and Rogerio Boaventura, from the Division for Information Foresight and Management for Oversight (DGI/AudDigital), showcased the ongoing development of Alice — a TCU-built robot for analyzing bidding documents. “We are building a future where technology and auditing go hand in hand, driving greater efficiency and social impact,” concluded Boaventura. 

Data Infrastructure in Foreign Institutions

The presentation from SAI Germany, delivered by Robert Schewski, highlighted the role of the Information Technology Department in supporting auditors. Schewski explained that his team develops tailored data products and provides training to empower auditors in using technological tools. He also addressed some of the challenges faced in Germany, such as the need to comply with local regulations and the complexity of managing large volumes of data. “Our focus is on empowering auditors through technology, ensuring they have the tools they need for efficient data analysis,” Schewski stated. 

SAI Netherlands shared its experimental initiatives involving LLMs, discussing the current status of the projects and the challenges encountered, such as hardware constraints, implementation hurdles, and model selection. The presentation offered insights into how these technologies can be integrated into audit processes and underscored the value of an experimental approach in identifying best practices. 

SAI United Kingdom presented two recently developed tools. The first is designed to generate press releases, incorporating AI to streamline textual output without compromising information quality or relevance. The second uses artificial intelligence to search through a data system, yielding results that match the semantic approach — the intended meaning — of the query. 

Next Steps

The next group meeting is already being planned, and participants look forward to continuing the exchange of experiences and lessons learned. The TCU reaffirms its commitment to innovation and international collaboration, strengthening its position as a global reference in public auditing.