(Co-authored with Lekha Chakraborty)
The rapid ascent of artificial intelligence (AI) promises substantial productivity gains and improved public services but poses significant challenges to labour markets and fiscal stance. AI is already reshaping core public-finance functions - revenue forecasting, tax and customs compliance1, social programme targeting, procurement, auditing, debt management and contingent liability monitoring and fiscal risk analysis2 (IMF, 2024; OECD, 2025). Although, India does use AI to identify fraudulent applications for input tax credits via false GST registrations besides Income Tax Department is using AI to identify falsified income tax deductions. It uses algorithms designed to identify unusual ratios between income and political or charitable donations. In India, a longstanding challenge in fiscal federalism has been the inefficiency in intergovernmental fiscal transfers. Funds released by the central government often remain idle in state accounts for extended periods due to implementation delays or other bottlenecks, yet the centre continues to bear the interest cost on its borrowings to finance these transfers.
Artificial intelligence offers a promising path forward by shifting from traditional demand-driven, lump-sum releases (viz., quarterly) to a more precise, just-in-time mechanism. Machine learning models could predict the exact timing of liquidity needs at the departmental level, based on hysteresis of public spending patterns, project milestones, and real-time data. This would enable the central treasury to hold onto cash longer, minimizing idle balances and reducing overall borrowing needs, and thus interest expenses, for the quarter. Such predictive approaches, already gaining traction in corporate treasury management, could enhance fiscal efficiency and strengthen public financial management in emerging federations like India.
Global experiences on AI in Public Finance
Globally, tax authorities are increasingly harnessing AI to strengthen revenue administration, with widespread adoption in fraud detection, compliance monitoring, and taxpayer services. Table 1 summarises country-specific experiences primarily drawn from global tax authority adoptions as of early 2025, focusing on fraud detection, compliance, and efficiencies in tax/VAT administration. Many countries also employ AI-driven virtual assistants for taxpayer services (e.g., Estonia, Finland, Ireland, Latvia, Mexico, Peru, Russia, Spain). Numerous countries are increasingly deploying artificial intelligence to combat tax evasion and enhance revenue collection. The UK's His Majesty's Revenue & Customs (HMRC) is developing a large language model to detect VAT evasion, while Malta raised an additional €400 million in tax revenue in 2024 through predictive AI, and also employs daily AI comparisons of declared wealth against public sources, bank data, and registers alongside the UK, Canada, the Netherlands, and Ireland.
Table 1: AI and Fiscal Policy: A Cross-country analysis
| Country | AI Applications | Key Benefits/Outcomes |
|---|---|---|
| Australia | Deep learning, natural language models, gradient-boosting for GST fraud detection | Identified >$530million unpaid taxes; prevented $2.5billion fraudulent claims |
| Austria | Predictive Analytics Competence Centre (PACC) with machine learning for risk management | Analysed 6.5million cases in 2023; recovered €185 million in fraud detection |
| Brazil | 'High Performance Inspection' (FAPE) for behavioural insights in tax communications | Optimises tone of tax letters for better compliance |
| Canada | Daily wealth comparison with public data; virtual assistants | Identifies undeclared assets; improves compliance and administrative efficiency |
| France | Satellite image scanning for conspicuous consumption signs (e.g., pools, cars) | Detects discrepancies for local taxes |
| Greece | Specialist AI unit using MyDATA VAT transactions | Real-time fraud detection and audit automation |
| India | Fraudulent GST input credit detection via false registrations (since May 2023) | Collaboration between Central government’s Business Intelligence and Fraud Analyst (BIFA), e-way portal, and Rajasthan government’s Business Intelligence Unit (BIU) |
| Italy | VeRa algorithm cross-referencing data; identifies high-risk cases | >1million high-risk cases detected; improves with data processing |
| Malta | Predictive AI for tax/VAT analysis | Raised €400million in taxes in 2024; combats evasion |
| Netherlands | Xenon tool for internet-based evasion investigation (shared with EU countries) | Enhances cross-border detection; past scandal with biased algorithms noted |
| Poland | STIR system for real-time bank transaction analysis | Detects carousel fraud in near-real-time (vs. previous 2 months) |
| Romania | AI and robots for VAT receipt boosting | Up to 1% increase in VAT collections |
| Singapore | Network visualizer with graph database; chatbot with LLM | Uncovers deep entity relationships; saves taxpayer hours |
| Sweden | AI for risk flagging in new business incorporations (since 2021) | Speeds applications; reduces manual reviews |
| United Kingdom | Wealth comparison system; developing LLM for VAT evasion; virtual assistants | Identifies undeclared assets; enhances evasion detection |
| United States | IRS AI plan for high-income taxpayers; Modernized e-File system | Focus on $1million income; processes 76% returns without human intervention |
| Vietnam | Flagging unusual invoicing patterns (adopted by end-2023) | Identifies fraud attempts to reduce taxable revenue |
Source: Authors’ collations3
Greece is establishing a specialist AI unit leveraging MyDATA VAT transactions for fraud detection, Austria's AI efforts have yielded impressive results in targeting missing trader VAT fraud, and Romania reports a boost in VAT receipts by up to 1% over the past year via AI and robotics. Italy stands out as one of the most advanced users, identifying over 1 million high-risk cases last year through AI-driven analysis, including its VeRa algorithm that cross-references tax filings, earnings, property records, bank accounts, and electronic payments to flag discrepancies, prompting explanatory letters to taxpayers while improving through machine learning. Vietnam plans to adopt AI for flagging suspicious invoicing patterns indicative of revenue manipulation, and Australia has uncovered over $530 million in unpaid taxes and prevented $2.5 billion in fraudulent claims using deep learning, natural language models, and gradient-boosting techniques for GST fraud.
In the US, the IRS is rolling out AI tools focused initially on high-income taxpayers and modernizing e-filing processes. Sweden uses AI to screen new business incorporations for tax avoidance risks since 2021, accelerating approvals, while France employs satellite imagery AI to spot signs of unreported wealth like swimming pools or multiple vehicles for local property taxes -a technique that could extend to nightlights data for mapping economic activity against collections. Six European countries, including the Netherlands which developed it, use the Xenon tool for internet-based surveillance in evasion probes, and Brazil applies AI behavioural insights in its 'High Performance Inspection' program to optimize tax communication tones based on taxpayer profiles, potentially reducing litigation.
Empirical Evidence so far on AI and Fiscal Policy
Recent studies highlight how AI-driven automation is weakening the employment effects of traditional fiscal stimulus. For instance, research shows that rising robot density has halved the job-creating impact of government spending in Europe, with low-skilled workers bearing the brunt4. Ebeke and Eklou (2023) examined how industrial robot adoption reduces the effectiveness of fiscal stimulus in creating jobs. In highly automating economies, extra demand is met more by robots than hiring. Evidence from 18 European countries shows automation halves employment response to stimulus, especially in manufacturing, low-skill jobs, and women's employment.
Positive innovation shocks from AI can boost total factor productivity, wages, and output while reducing unemployment, as evidenced by UK patent-based analyses. Evgenidis and Fasianos (2025) analyses AI innovations' macroeconomic impacts in the UK (1982–2022) using machine learning on 550,000 patents to measure AI news shocks. They inferred that AI boosts output, investment, wages, hours worked, and reduce unemployment, with immediate TFP gains outweighing displacement effects due to rapid AI diffusion5. In the US, simulations suggest optimistic scenarios where AI slashes healthcare costs could substantially lower deficits, though pessimistic outcomes involving unchecked spending hikes point to fiscal pressures amid aging populations6. A Brookings paper projects AI's long-term effects on US fiscal outlook, focusing on old-age entitlements. Estimates show AI-driven shocks could increase annual budget deficits by up to 1.6% of GDP or decrease them by 0.8% of GDP by 2044, depending on impacts to health costs and longevity.
To navigate these shifts, fiscal policy must adapt boldly. As labour’s share of income potentially declines under transformative AI, traditional tax bases erode, necessitating pivots toward taxing capital, data, or AI-generated rents7. A NBER paper by Anton Korinek explores transformative AI's potential to make intelligence abundant and reproducible, accelerating economic growth while devaluing human labour. It discusses shifting bottlenecks to scarce resources, changing prices and income distribution, varied scenarios from prosperity to inequality, labour disruptions, global divides, and AI's role in enhancing economic research amid alignment risks.
Strategies like global minimum taxes on multinationals and levies on AI infrastructure emerge as resilient options across displacement scenarios8. Transformative artificial intelligence (TAI) threatens public finance through widespread labour displacement eroding wage-based revenues (over 50% of OECD taxes), borderless economic activity evading geographic levies, and extreme concentration of gains bypassing traditional mechanisms. In "Funding Government in the Age of AI," by Huynh, Mittal, and Frank (2025) evaluate nine revenue strategies using the “FIRI Framework” (Feasibility, Incidence, Resilience, Incentives) across scenarios varying in displacement, productivity, and inequality, as follows: (i) Land value taxes exploit scarce land's value, offering high feasibility via existing systems, progressive incidence, resilience as AI wealth flows into property, and strong productive incentives. (ii) Consumption taxes like VAT prove broad and stable, minimally distortive, resilient to job loss since spending persists, and feasible with offsets for regressivity. (iii) Global corporate income taxes target mobile rents through coordination, addressing borderless profits and concentration but depending on international cooperation. (iv) Sovereign wealth funds capture and redistribute AI-generated wealth, requiring governance for feasibility and equity amid productivity surges. (v) Windfall clauses share unexpected gains via mechanisms, countering concentration with proactive design. (vi) Net wealth taxes tackle capital accumulation progressively but face evasion and valuation challenges in borderless worlds. (vii) Automation taxes penalize labour substitution yet struggle with definition, enforcement, and observability. (viii) Digital services taxes serve as interim borderless tools but foster friction and low resilience. (ix) Pigouvian taxes internalize externalities like displacement harms, appealing in theory but hampered by measurement and consensus. No single strategy suffices; the authors recommend a hybrid portfolio prioritizing land value and consumption taxes for superior FIRI performance, complemented by global corporate taxes, wealth funds, and others to ensure fiscal stability in TAI futures. For emerging economies, progressive shifts from labour to corporate taxes, combined with digital levies to fund retraining and social protections, are key, alongside connectivity subsidies9.
AI offers significant potential for public financial management (PFM) through task automation and predictive analytics, enhancing efficiency in fiscal forecasting, spending, budgeting, and stakeholder engagement10. However, AI adoption remains cautious due to challenges like data quality, legacy systems, biases, and unquantified impacts. Fiscal policymakers should prioritise IT upgrades, skill training, ethical standards, and robust evaluation frameworks for safe, effective integration of AI in fiscal policy.
As global tax authorities increasingly leverage AI for enhanced fraud detection, compliance, and revenue mobilization, as evidenced by significant gains in countries like Malta and the UK, India is poised to accelerate this trend, with the upcoming Union Budget 2026-27 expects to prioritise AI-driven initiatives in tax administration and fiscal policy imperatives to boost efficiency.
[1] OECD.2025, Using Artificial Intelligence in Public Financial Management. https://one.oecd.org/document/GOV/SBO(2025)4/en/pdf
[2] IMF.2024, Understanding Artificial Intelligence in Tax and Customs Administration. https://www.elibrary.imf.org/view/journals/005/2024/006/article-A001-en.xml
[3] Based on 2025 New models of Artificial Intelligence (AI) and Machine Learning (ML) are being evaluated by global tax authorities to tackle fraud. https://www.vatcalc.com/artificial-intelligence/tax-authorities-adopt-ai-for-tax-fraud-and-efficiencies/
[4] Ebeke, Christian H., and Kodjovi M. Eklou. “Automation and the Employment Elasticity of Fiscal Policy.” Journal of Macroeconomics 75 (2023): 103502. https://www.sciencedirect.com/science/article/pii/S0164070423000022
[5] Evgenidis, Anastasios, and Apostolos Fasianos. “AI News Shocks and the Macroeconomy: Evidence from UK Patent Data.” Institute for Fiscal Studies Working Paper WP2025/48 (2025). https://ifs.org.uk/publications/ai-news-shocks-and-macroeconomy-evidence-uk-patent-data
[6] Harris, Benjamin H., Neil Mehrotra, and Eric So. “The Fiscal Frontier: Projecting AI’s Long-Term Impact on the US Fiscal Outlook.” Brookings Center on Regulation and Markets Working Paper (2024). https://www.brookings.edu/articles/the-fiscal-frontier/
[7] Korinek, Anton. “The Economics of Transformative AI.” NBER Working Paper 32980 (2025). https://www.nber.org/reporter/2024number4/economics-transformative-ai?page=1&perPage=50
[8] Huynh, Ky-Cuong, Suchet Mittal, and Noah Frank. “Funding Government in the Age of AI.” 2025 https://www.convergenceanalysis.org/fellowships/spar-economics/funding-government-in-the-age-of-ai?ref=kycuong.com
[9[ Liu, Yan, Antonio Martins Neto, Sharada Srinivasan, Saloni Khurana, He Wang, and Juan Porras. “Digital Progress and Trends Report 2025: Strengthening AI Foundations.” Washington, DC: World Bank, 2025. https://www.worldbank.org/en/publication/dptr2025-ai-foundations
[10] OECD. Using Artificial Intelligence in Public Financial Management.” OECD (2025) https://one.oecd.org/document/GOV/SBO(2025)4/en/pdf