A Risk Framework for Artificial Intelligence Use in Tax Administration and Preparation
Risks related to the use of artificial intelligence in tax administration are examined, and dedicated risk registers are proposed for tax authorities and preparers.
The integration of artificial intelligence technologies into tax administration and preparation processes offers significant performance opportunities while simultaneously bringing legal and financial risks. Accordingly, field-focused risk frameworks and registers have been developed for tax authorities and preparers.
The Role of Artificial Intelligence in the Tax Field
While artificial intelligence offers revolutionary opportunities in areas such as efficiency and customer satisfaction, it also brings risks that are not yet fully understood. Dedicated risk frameworks are needed in fields such as law, medicine, insurance, accounting, and finance.
The tax sector stands out as one of the most compatible professional fields with artificial intelligence due to its operation on deterministic input-output logic and its massive scale of data. However, the consequences of errors in this domain can be highly costly.
Risk Register for Tax Authorities
The risk register prepared for tax authorities covers twenty risk areas across four main categories: information integrity, fairness and legitimacy, security and data, and institutional capacity. These risks aim to enable agencies to securely manage their own artificial intelligence applications.
Information Integrity and Accuracy Risks
In the information integrity category, critical risks include artificial intelligence producing factual but convincing outputs—known as hallucination—drift occurring in outputs over time, unexplainability issues, and the uncritical acceptance of artificial intelligence outputs.
Fairness, Legitimacy, and Security
Under the heading of fairness and legitimacy, biased approaches toward different taxpayer groups and audit targeting issues are addressed. In the security and data category, situations such as data leaks, prompt injections, and the compromising of sensitive taxpayer information are examined.
Institutional Capacity and Governance
The use of artificial intelligence can impact institutional capacity, leading to expertise erosion, scope creep, and audit vulnerabilities. This carries the risk of weakening organizations' ability to oversee artificial intelligence.
Implementation and Prioritization Framework
Because risks can be simultaneous and contradictory, a multidimensional approach based on the COSO enterprise risk management framework has been adopted. Risks are evaluated using a three-tier prioritization system.