India’s tax department needs an AI rulebook
The Income Tax Department has officially started using artificial intelligence (AI) for various purposes. Under Project Insight and the Income Tax Transaction Analysis Centre, AI and big-data tools analyse large datasets from banks, property registrars, stock exchanges and financial institutions to build a 360-degree taxpayer profile and identify possible tax evasion and hidden assets.
Data-driven scrutiny
For data integration and analysis, information from the GST Network (GSTN), Securities and Exchange Board of India (Sebi), Ministry of Corporate Affairs (MCA), real estate regulatory authorities (RERA), transport authorities and the Financial Intelligence Unit-India (FIU-IND) is cross-referenced to detect mismatches between reported income and actual financial or business activity.
Machine-learning algorithms are also used to identify cases for risk-based scrutiny. They scan tax returns, the Annual Information Statement (AIS), Tax Information Statement and Annual Tax Statement for anomalies, helping select high-risk profiles for scrutiny. In faceless assessments, scrutiny cases are allocated, notices generated and draft orders processed through automated, system-driven workflows.
Predictive analytics, large language models (LLMs) and automated communications are also used to generate compliance alerts and nudges for non-filers or under-reporters. Taxpayers’ digital footprints, including high-value transactions, digital wallets and broader financial behaviour, are analysed to track potentially unreported wealth, foreign assets and virtual digital assets such as cryptocurrencies.
The impact of these tools is evident from the fact that, during the current year, scrutiny notices have become more focused and direct in addressing relevant risk points than in earlier years. The use of AI by the tax administration has undoubtedly helped detect tax evasion and improve taxpayer compliance.
In recent months, however, one has also noticed certain tax officials using AI for other purposes. Replies to notices are sometimes scrutinised using AI to generate subsequent notices. Some orders, including assessment orders and appeals decided by Commissioners of Income Tax (Appeals), appear to have been drafted using AI.
This raises concerns about the accuracy of AI-generated content. In one order, reliance was placed on a Tribunal ruling to dismiss an appeal, even though the Tribunal had decided the case in precisely the opposite manner to that stated as the basis for dismissal.
Who is accountable?
Several important questions arise about the use of AI for these purposes. Is such usage authorized by the tax department? How are privacy and confidentiality protected when officers use AI in these cases? What controls do senior authorities exercise over their juniors’ use of AI? Can decision-making powers and judicial discretion vested in tax authorities be delegated to AI? What recourse does a taxpayer have if an officer uses AI without authorization, or uses it incorrectly while processing an assessment or appeal?
The Central Board of Direct Taxes (CBDT) needs to formulate a policy governing the use of AI by the tax department. This policy should be made public and incorporated into tax law. AI use should be transparent wherever it affects taxpayer interactions, particularly assessments and appeals, and appropriate safeguards must be built in.
Taxpayers need to know when AI use is permissible and when it is not. There must also be an effective remedy when unauthorized use of AI by tax officials prejudices a taxpayer, including a formal reporting or grievance-redressal mechanism.
The CBDT must also educate its officers that AI is, at best, a tool to support decision-making. It is not a substitute for, and cannot replace, the judicial decision-making of tax officers.
Tax officers build their knowledge and experience over years through the continuous application of their minds to different cases. If they abdicate this responsibility to AI, the long-term consequence could be a generation of officers without the knowledge and judgement required to discharge their duties, potentially spelling disaster for the tax department.
Lessons from abroad
Several jurisdictions, including the European Union, the UK, Australia and Singapore, have introduced policies governing AI use, supported by legal frameworks. The OECD has also published a policy paper emphasising that AI systems should be transparent, reliable, secure and subject to meaningful human oversight. It calls for accountability for AI-assisted decisions and governance mechanisms capable of identifying and correcting algorithmic errors or unintended bias.
The underlying principle is clear: AI should support decisions, not replace human judgement.
Given the swift and common usage of AI, it is imperative that the government and the CBDT introduce a clear and transparent policy at the earliest. Such a framework is necessary to protect taxpayers’ interests, privacy and confidentiality before affected taxpayers are forced to seek legal recourse through the courts.
Gautam Nayak is a partner at CNK & Associates LLP. Views are personal.