Russia's AI-driven tax tools have reduced routine audits while raising the yield and effectiveness of enforcement, sharpening the state's capacity to detect VAT fraud; but the shift concentrates power, erodes taxpayer trust and increases dependence on foreign technology, posing new risks to fiscal legitimacy.

Main Finding

The article finds that deploying AI-based tax systems in Russia (notably ASK VAT-2) has substantially increased the productivity of tax administration between 2019–2024: fewer audits are conducted, but those audits are more effective and yield larger volumes of additional assessments. At the same time, algorithmic enforcement introduces new risks to sustainability, fairness, and legitimacy—centralized control, reduced taxpayer trust, and dependence on foreign technology—that complicate the state's economic-security objectives.

Key Points

• Implementation effect: ASK VAT-2 and related systems have reduced the total number of tax audits while raising effectiveness metrics (higher additional assessments per audit and more violations detected per case).
• Duality of impacts:
  • Positive: stronger detection of evasion and contraction of shadow economic activity; more targeted enforcement lowers administrative burden and may free resources.
  • Negative: heightened centralization of monitoring and decision-making power in the tax authority; potential erosion of taxpayer trust if algorithmic decisions seem opaque or arbitrary; strategic risk from reliance on non‑domestic software and hardware.
• Legitimacy and fairness concerns: automated targeting and profiling can produce perceived or real biases (e.g., sectoral or regional concentration), complicate appeal and due‑process, and disproportionately affect small or informal firms.
• Economic-security framing: tax administration is reframed as a technological institution; dependence on AI affects state capacity, resilience, and sovereignty.
• Institutional remedies emphasized: transparency of algorithms, governance mechanisms, certification and auditing of systems, investments in domestic solutions, and strategic oversight to balance efficiency with rights and trust.

Data & Methods

• Data: Official statistics from the Federal Tax Service of Russia covering 2019–2024 (audit counts, outcomes, volumes of additional assessments, possibly sectoral/regional breakdowns).
• Quantitative analysis: Time-series comparison before and after ASK VAT-2 deployment showing declines in audit counts alongside increases in per-audit yields and total additional assessments; descriptive metrics demonstrate the shift toward fewer, higher‑yield interventions.
• Qualitative/institutional analysis: Examination of policy documents, governance arrangements, and strategic risk factors (centralization, foreign dependence, legitimacy). The authors discuss institutional design and strategic implications for economic security under algorithmic governance.
• Methodological limits noted (implicit): reliance on administrative data that reflect enforcement outputs rather than full social welfare impacts; potential confounders and causal attribution challenges in observational administrative settings.

Implications for AI Economics

• Efficiency vs. distributional/trust trade-off: AI raises enforcement productivity (a static efficiency gain) but may create dynamic costs through reduced taxpayer cooperation, compliance legitimacy, and increased social friction—factors that economists must incorporate when evaluating AI in public finance.
• Market structure and sovereign risk: Reliance on foreign AI solutions constitutes a form of technological dependence with macroeconomic and geopolitical dimensions; promoting domestic AI capacity becomes an economic-security and industrial-policy priority.
• Political economy of enforcement: Algorithmic targeting reshapes incentives for firms and officials; concentrated, high‑precision enforcement can alter bargaining, lobbying, and tax avoidance strategies—models of tax evasion and compliance should incorporate algorithmic detection probabilities and transparency.
• Welfare and distributional analysis: Changes in enforcement intensity across firm sizes and sectors can have heterogeneous effects on growth, informality, and inequality; cost‑benefit assessments should include legitimacy and long-term compliance feedback loops.
• Policy design recommendations for AI-integrated tax systems:
  • Build auditability and procedural safeguards into algorithms (explainability, human‑in‑the‑loop, appeal mechanisms).
  • Invest in domestic development and open standards to reduce external dependence.
  • Monitor distributional impacts and incorporate randomized or equity‑aware targeting to avoid biased enforcement.
  • Establish independent oversight, transparency obligations, and periodic fairness/security audits to sustain trust and legitimacy.
• Research directions: quantify long-run effects of algorithmic enforcement on voluntary compliance, firm behavior, and shadow-economy size; study optimal governance mixes balancing automation and human discretion; model strategic interactions between taxpayers and algorithmic auditors.