A deep-dive into the new tax system

Pakistan’s tax system has been caught in a vicious cycle for many years: a low tax base, high levels of informality, and an administration with wide discretion that has traditionally been used by those who have worked in taxation as a tool for intimidation and extortion rather than for revenue generation.

Pakistan currently has one of the lowest tax-to-GDP ratios in the region, and policy and enforcement remain weak. In this context, the National Faceless Centre (NFC), a centralised online tool intended to carry out all aspects of tax auditing and assessment without direct or indirect taxpayer interaction with government agents, was enacted under the Finance Act 2026, which its authors characterise as a clear departure from prior practices. This reform’s justification is simple: if corruption results from human interaction, then removing human interaction will also eradicate corruption.

Beginning with the arithmetic, there are 27 officers in addition to a pilot covering third-party data for 300,000 to 400,000 individuals. This means that the officers will have over 11,000 potential cases each to go through, each involving at least three officers: an auditor, an assessor, and a reviewer.

This audit regime appears “faceless”; however, an analysis of it provides a more concerning view. The way this system has been constructed; the legal structure on which it will operate; and the institutions within which it is now being implemented; all demonstrate how the facelessness of the audits could provide an opportunity for a new level of opaqueness, whereby algorithmic decisions become non-contestable; due process is consistently denied; and where the most vulnerable taxpayers lack both the technological resources and/or the legal standing to challenge or defend their positions.

The present configuration of this system poses the risk of replacing one form of arbitrary authority (the discretion of the tax officer) with another, namely, the discretion conferred by algorithms and the institutions that control them.

The stated objectives and their limits

The non-personalised audit process, created by the National Faceless Centre (NFC) in Islamabad, will be organised into a three-wing structure: a National Faceless Audit Wing, a National Assessment Wing, and a Field Operations Wing.

This new structure was designed to move functionally related work currently being completed by a single official, thus removing the ability of the one official to have both power over the taxpayer and their money, a structural reform that can potentially “cut down corruption up to 75%” according to the FBR.

All audits, assessments and decisions are to occur electronically. Once the case has been assigned via an electronic algorithm, all officer identities will remain anonymous to taxpayers. Additionally, electronic hearings will replace in-person contact, and all aspects of the process will be documented in a digital audit trail.

This reform is based on a diagnosis that has some validity. There is well-documented evidence of misuse of taxpayer information by tax officers. There is also evidence of “rent seeking” (i.e., using official positions for private gain) and collusion between tax officials and taxpayers.

Similarly, there are differences in how cases are treated across different regions. In fact, even the current FBR chairman testified before a parliamentary committee, admitting that his officers were misusing taxpayer information. Taxpayer harassment complaints have consistently been an issue within the tax system, and with all auditing, assessment, and enforcement powers concentrated in the hands of individual tax officers, the potential for abuse was clear. The faceless model correctly identifies these problems and attempts to address them structurally rather than through exhortation.

Algorithms without accountability – the ‘black box’ problem

However, the central component of this new government is the Algorithmic Settlement Mechanism (ASM). This mechanism will allow the FBR to provide taxpayers with an opportunity to settle outstanding issues through an automated system. If a taxpayer accepts the system-generated proposal, they have 10 days to deposit the proposed settlement and submit a revised return. Once the taxpayer submits the revised return, the audit on the case is terminated. Many believe that ASM can help alleviate some of the pressure on Pakistan’s massive backlog of tax disputes, but its design raises profound concerns.

Firstly, and most importantly, the parameters for the algorithm used in settlement offers are provided by the FBR. The law has an open-ended provision for taking account of “any other basis the Board may consider relevant”. There is no independent validation, no statutory cap on demands and no formal right of appeal against the offer. This implies that taxpayers, particularly smaller ones that lack legal capacity, have 10 days to accept the offer before they can challenge the validity of their defence. Furthermore, there is no statutory Right of Appeal against the Offer, so the result of the algorithmic process is final and binding unless the taxpayer challenges the assessment, which the ASM was supposed to avoid.

This is specifically seen by Ianushevych and Hubanova, who conducted a systematic analysis of 53 studies on AI in tax administrations (Multidisciplinary Reviews, June) and found that “risk-scoring” models can both misidentify taxpayers and indirectly discriminate against certain groups of taxpayers if the data they were trained on contain bias or the model’s decision-making processes are opaque. As one might expect in developing countries like Pakistan, there are biases evident in how different populations build an online footprint. For example, salaried workers and bank account holders create a trail of financial transactions that is easily monitored. Additionally, individuals who have already filed taxes will likely continue to file. This, then, leaves out the cash economy completely.

Tax administrators around the world are implementing risk scoring models using internal guidelines/technical standards, and then, if at all, establishing legal safeguards afterwards. International standards currently exist, i.e., OECD Principles on Artificial Intelligence (non-binding); therefore, unless a particular jurisdiction has its own legally binding safeguards in place (Pakistan does not), these international standards do not limit the actions of tax authorities.

Although Pakistan has a draft version of the Personal Data Protection Bill, it remains uncodified. There is no current legislation granting citizens the right to inspect and correct any information a government agency may hold about an individual. This “transparency” issue arises from an opaque auditing environment itself. The use of algorithms to monitor taxpayer behaviour, with each taxpayer subject to AI-driven review before any human official, will likely compound these issues. So, who specifically audits the algorithm?

The problem of “false positives” compounds this opacity further. An algorithm can only analyse recorded transactions; it cannot capture economic activity conducted entirely in cash, which remains a substantial portion of Pakistan’s economy. The system will therefore scrutinise the documented (those already within the tax net) while leaving the undocumented untouched.

A small distributor moving Rs150 million in goods annually at thin margins may face the same algorithmic scrutiny as a genuine tax evader, while the cash-based trader operating entirely outside the formal system remains invisible. This is precisely why these systems are described as “black boxes” whose logic is closed to taxpayers and often to the officers applying them, making an assessment hard to verify and even harder to contest.

This holds even in the United States, where the right to appeal, access to information and judicial review are well established and algorithms only support human decisions. Another example can be seen in the Dutch tax authority’s risk profiling in the 2021 childcare benefits scandal, which wrongly branded thousands of families as fraudsters due to its algorithms.

Institutional capacity and the lessons of the faceless customs assessment

The Finance Act has essentially created a system in which no one can question the authority to make a decision on jurisdictional grounds. This removes all forms of procedural accountability from the quasi-judicial body acting in this capacity. The claim that closing the “historical loophole” used by taxpayers to delay enforcement of their liabilities actually removed the right to challenge the authority of the quasi-judicial body imposing such liability. Anonymity and accountability are not equivalent; the risk here is quite high.

The faceless model may reduce visible harassment while creating a system in which the grounds for decisions are opaque, the decision-makers are anonymous, and the avenues for challenge are systematically narrowed. The taxpayer who previously faced a corrupt officer could at least identify the source of their grievance; the taxpayer facing an algorithmic determination has no such recourse.

It is also not as if Pakistan lacks its own precedent. The faceless customs assessment system introduced in Karachi in December 2024 was also meant to end collusion. An internal review committee found it had not achieved that objective. Average clearance time rose by 57%, and the committee advised against further rollout unless the design was reviewed. None of this is an argument for the old arrangement. Officer discretion has been a source of harassment for decades, and splitting audit, assessment and review between three officers is a sound idea. The pilot also targets conspicuous spending by individuals, which is a fair place to start. But the objection here is much narrower. The anonymity by which this system operates protects the taxpayer from a corrupt officer, but it equally protects a careless officer from the taxpayer.

This indicates that technology, institutions and law move at different speeds. The FBR has moved the first two. Until the third catches up, taxpayers are dealing with a system they can neither see nor question.