The state of artificial intelligence in public audit
A · Strong evidence An OECD report based on structured consultations with 15 audit institutions, with documented use cases and adoption figures, though the sample is small and self-selected.
Summary based on the full text·Original: about 90 minutes to read
In short
An OECD report (May 2026) on how supreme audit institutions and other oversight bodies use AI. Between March and July 2025 the OECD consulted 15 institutions in 14 countries and the EU, including one multinational accounting firm and one public supervisory authority. Adoption is broad but shallow: most institutions have strategies, guidelines, training and at least one tool in production, yet many tools stay in pilot. Data quality and governance decide whether projects scale, and almost nobody can yet measure impact.
Key numbers
- 87%consulted audit institutions with at least one AI tool in production
- BRL 803 millionoperating-cost savings attributed to Brazil CGU's Malha Fina classification tool (EUR 129 million)
- 60 to 10-15 daysaudit turnaround at one private audit firm after AI tools plus process standardisation
- ≤3 peopletypical size of the dedicated AI teams inside audit institutions
As reported by the source.
English·PDF·opens oecd.org
What it reports
- 67% have a formal AI strategy, 80% have internal AI guidelines, 87% offer AI training to general staff, 87% have at least one AI tool integrated into a workflow.
- 67% have a dedicated AI team, usually no more than three people.
- Brazil's Comptroller General (CGU) used a classification model on 3,627 accountability reports from the National Cinema Agency. The report credits it with a 150% increase in analytical capacity, BRL 803 million (EUR 129 million) in operating-cost savings and "a 29-year reduction in analysis time".
- The French Court of Audit built ChatJF, an in-house assistant on open-source models (including LLaMA 3 and a Mistral model) with GPUs rented from a French cloud provider. As a prototype it logged about 300 unique users in 30 days and about 100 regular users, around 20% of staff.
- One institution set a target to cut its audit workforce by 50 positions through AI and automation.
- A private audit firm reported turnaround falling from 60 days to 10-15 days, attributed to tools combined with data standardisation and workflow redesign.
- The Netherlands Court of Audit is waiting for GPT-NL, a Dutch language model backed by about EUR 13.5 million in government funding.
- The use cases covered are anomaly detection, classification, predictive audit, search and retrieval, document processing, generative drafting and translation, and satellite or spatial analysis.
How they got it
A survey plus consultations with the participating institutions (Brazil, Costa Rica, France, Germany, Italy, Korea, the Netherlands, Norway, Singapore, the UK, three anonymous countries, and the EU). Case boxes describe specific tools. These are combined with the OECD's wider work on AI in government.
Limits
- Fifteen self-selected, mostly advanced institutions; the percentages describe this group, not the audit profession.
- Savings figures such as Malha Fina's come from the institutions themselves and are not independently checked.
- The report says no institution yet has a full framework for measuring AI impact; the indicators tracked are usage, training counts and time-to-value.
Why it matters for companies
Audit bodies face the same constraints as corporate finance, risk and internal audit functions: fragmented data, small expert teams and pressure for explainable outputs. The pattern of many pilots, few scaled tools and success tied to data standardisation applies to any control function. The private-firm example shows the gain came from redesigning the process, with the tool as one part. Companies should also expect that public auditors now use the same methods on the filings and records companies send them.
Summary based on the full text. Read the original for the full argument and data.Open PDF on oecd.org