Artificial Intelligence (AI) is entering one of the most consequential areas of government: public financial management (PFM). The Open Budget Survey (OBS) 2025 offers a first baseline of how AI is entering this space, asking a focused question: where is AI being used in public financial management and for what purposes?
AI applications are emerging in countries in different regions, income levels and administrative contexts, making accountability around its use an increasingly important issue. Across the 82 countries assessed, OBS 2025 found AI use at varying stages of development—from systems already performing fiscal tasks to early strategies and pilots.
18
Countries with AI already performing public finance tasks during the survey period.
21
Countries with pilots, strategies, announced projects or post-cutoff develoments.
43
No publicly available evidence or confirmed use of AI.
The survey looked not just at whether AI is entering PFM, but whether governments disclose enough for citizens, civil society, legislatures and oversight institutions to understand how it’s used. So far, mostly not: AI is taking hold in public finance faster than accountability around it.
PFM areas where AI adoption is gaining ground
1. The most common use case of AI in PFM is tax compliance and fraud detection.
Revenue authorities have structured data and clear incentives to flag suspicious refunds, underreporting or high-risk sectors. The results are easy to count: a refund blocked, a filing flagged, revenue recovered.
Number of countries with confirmed operational AI use by application
South Africa’s revenue authority reported using AI risk engines to detect impermissible refunds by cross-checking information such as bank statements, VAT returns, and company registration data. Türkiye’s Revenue Administration utilizes an AI-powered Geographic Information System (GIS) called MEVA (Mekânsal Veri Analiz Sistemi) to automatically cross-reference real estate holdings with financial and tax records. The spatial AI maps data from land registries, bank loan amounts, tax declarations and residency records to flag massive disparities between declared income and apparent wealth. These use cases show how AI can help revenue authorities detect discrepancies across complex and fragmented datasets, strengthen compliance enforcement and protect the integrity of revenue systems.
2. A second area is budget execution monitoring: tracking whether money is spent as planned, correctly classified and traceable in real time after a budget is approved.
In the Philippines, Project DIME (Digital Information for Monitoring and Evaluation) uses citizen feedback, geotagged data, drone and satellite-based monitoring and AI image analysis to validate the physical progress of infrastructure projects, flag delays and combat corruption— assessing whether released funds are translating into results on the ground.
A related application is AI-supported expenditure classification and reporting. Brazil’s National Treasury (STN) uses machine learning to classify subnational spending according to the international Classification of the Functions of Government (COFOG), which organizes expenditure into areas such as health, education, social protection and environmental protection. In an article published on the International Monetary Fund’s Public Financial Management Blog, Treasury officials reported that the process reduced roughly 1,000 hours of manual work to eight hours while achieving an accuracy rate of over 97 percent.
3. A third area is predictive analytics for fiscal planning. This includes revenue forecasting, macroeconomic modelling, scenario simulation, debt projections and fiscal risk analysis. These applications help shape what governments expect to collect, what they believe they can spend and how they understand future risks. Egypt’s Ministry of Finance, for example, reported using the IMF’s Financial Programming and Fiscal Policies AI model to project revenues and expenditures over a three-year period.
4. A fourth use is citizen-facing assistance through chatbots and digital assistants. In the examples confirmed through our research, these applications were concentrated in tax administration. Chile’s tax authority uses SOFÍA, a virtual assistant that logged over 75,000 taxpayer conversations; Türkiye’s Revenue Administration runs GİBİ, a chatbot answering tax questions; Peru’s tax administration reports automated taxpayer services; and Morocco’s tax authority has a bilingual chatbot to guide users through tax procedures.
Is AI use in public finance transparent and accountable?
The transparency and accountability principles we should apply to AI are the same principles IBP has applied to public budgets for two decades: governments should explain what they are doing, publish enough information for others to understand it, and allow the public and oversight institutions to question decisions affecting public resources.
AI is evolving faster than many public budgeting systems and accountability processes. The core principles remain the same, but rapidly changing AI tools require deliberate and innovative efforts to apply them in new ways. As AI begins to shape the information, analysis and risk signals used to determine how governments raise and spend public resources, accountability needs to be built in from the design of an AI tool through its implementation and review.
This matters across the main areas where the OBS found AI being used in public finance:
- Revenue administration: AI can help protect public resources and improve enforcement, but it can also shape who is flagged, reviewed, audited or cleared— making transparency and safeguards essential.
- Classification and reporting: Efficiency gains aren’t the whole story. Public reporting should explain how errors are caught, whether uncertain outputs receive human review and who is responsible for the final data.
- Predictive analytics: Errors here don’t stay technical. They travel through the budget process and affect real services. The closer AI gets to decisions about who pays and who benefits, the stronger the case for scrutiny and public consultation in the design of these tools.
- Citizen-facing assistance: These tools ease navigation of government information, but they also raise questions about what happens when people rely on guidance that is incomplete or wrong.
OBS 2025 asked directly: when AI-driven fiscal models, forecasts, or decisions are used, are they subject to transparency and oversight? The results suggest accountability lags far behind adoption.
Countries by level of disclosure and oversight for AI-driven fiscal models, forecasts, and decisions
Brazil was the only country to meet the strongest standard: the National Treasury has publicly described how its classification tool works — purpose, methodology, reported accuracy gains — and published the resulting data. That doesn’t answer every accountability question, and it is not the same as full public oversight or participation. But it shows that meaningful disclosure is possible without exposing sensitive systems.
Elsewhere, this kind of disclosure remains the exception. In many countries, the issue was not only that AI accountability practices were limited; it was that AI use itself was difficult to identify. In 16 countries, government reviewers supplied AI-related information that researchers had not identified through public sources. In 11 of those cases, the information pointed to confirmed operational AI use during the survey period. In a small but revealing way, the research process reproduced part of the accountability problem the survey was trying to assess: information was not necessarily secret, but it was hard to find without knowing whom to ask or where to look.
This does not mean AI is being misused. It means the public record is thin. And that is itself an accountability concern. It also reflects a broader pattern in digital public finance reform: governments may invest in back-end systems — financial management platforms, digital tax administration, or AI tools — without making complementary investments in publication, interoperability, public access and independent review. The result may be a system that works better inside government but remains difficult for the public to understand, use or question.
Publication is not a bureaucratic formality—if information is not accessible, it cannot support oversight and accountability. Beyond disclosure, accountable AI in public finance also requires external scrutiny and participation: someone outside the system must be able to review whether it is working as intended, and the public must have a way to raise concerns about how AI is being used in decisions that affect them.
What should governments disclose?
The OBS findings do not settle what an accountable AI ecosystem in public finance should look like. But they show why the debate is urgent: transparency, oversight, safeguards and public participation need to be defined before systems become entrenched and accountability questions become harder to ask.
The public doesn’t need every technical detail, but they need enough information to see how a system is used and what happens when it fails.
Just as the public should have opportunities to engage on budget and revenue decisions, citizens and civil society should also have channels to question and provide feedback on how governments are designing and applying AI in public financial management.
Good disclosure and consultation practices should cover:
- Purpose and scope: what the system does and what data it uses
- Human oversight: whether humans review outputs before decisions are finalized
- Safeguards: what happens when the system makes an error and who catches it
- Institutional ownership: which government body is accountable
- External review: whether an audit, independent body, legislature or other oversight actor can review the system beyond internal checks
- Public feedback: how citizens and civil society can raise concerns, contest errors or provide input on AI tools that affect fiscal decisions or public services
Governments may have legitimate reasons to protect sensitive enforcement methods, confidential data or internal risk models. But public trust requires more than internal assurances. Systems that work in controlled settings can still produce unexpected consequences when deployed at scale. That is why AI in public finance should be treated not only as a technical reform but as an accountability reform.
