Shaleen ShahThis Article was originally published in Chartered Accountants Association Ahmedabad Journal (Dec-2025)
As India prepares to welcome 2026 and a new Income Tax law, the country’s tax administration stands at an inflection point shaped not merely by legislative reform but by the exponential advance of artificial intelligence and algorithmic governance. In the last couple of years, AI has revolutionised tax compliance—automating data matching, risk profiling and return processing at a scale and speed unimaginable to earlier generations of officers and taxpayers. That revolution shows no sign of slowing; if anything, the trajectory suggests that within a few years, most routine tax determinations will be made not by human Assessing Officers deliberating over files, but by algorithms executing encoded rules. This shift brings undeniable efficiency gains, but it also introduces a structural tension: tax law, with its labyrinthine provisos and context‑dependent exceptions, does not translate neatly into the binary logic that machines require. Taxpayers navigating this new landscape increasingly find themselves not arguing with an officer who can be persuaded or corrected, but contesting a demand generated by code—code they cannot see, cannot question, and often cannot understand. Until tax governance frameworks evolve to keep pace with AI’s capabilities and limitations, taxpayers must brace for a reality in which their legal positions are adjudicated at the processing stage by systems that lack the contextual judgment the statute still presumes. The Centralized Processing Centre at Bengaluru exemplifies this shift, and examining its role and limitations offers a window into the broader question of what “assessment” will mean in an algorithmically governed tax system.
Statutory status of CPC and CIT(CPC)
The Centralized Processing Centre at Bengaluru operates as a core component of India’s income tax administration, functioning as a processing mechanism created under the Centralised Processing of Returns Scheme, 2011, read with the enabling provisions in section 143(1A) of the Income Tax Act, 1961, and the Board’s powers under sections 119 and 120. Despite its central role in processing millions of returns annually, CPC is not itself named anywhere in the Act as an “income‑tax authority,” nor is “CPC” included in the definition of “Assessing Officer” in section 2(7A). Section 2(7A) is exhaustive: an “Assessing Officer” means the Assistant Commissioner or Deputy Commissioner or Assistant Director or Deputy Director or the Income‑tax Officer vested with jurisdiction by an order under section 120, and in some situations Additional/Joint Commissioners when specifically directed. A processing centre or an algorithmic engine does not fall within this list.
The Commissioner of Income‑tax (Centralized Processing Centre), Bengaluru – usually styled “CIT (CPC), Bengaluru” in notifications – is an administrative designation at the Commissioner level within the departmental hierarchy. Commissioners are higher supervisory authorities appointed under section 117; they are distinct from the cadre of “Assessing Officers” enumerated in section 2(7A) and do not become Assessing Officers merely by virtue of their rank.
CBDT’s Notification No. 155/2025 authorises the Commissioner of Income‑tax, CPC Bengaluru to exercise concurrent powers to: (a) rectify mistakes apparent from the record under section 154 in cases processed through the CPC–AO interface, and (b) issue notices of demand under section 156 for such rectified cases. The notification further permits the Commissioner to delegate these operations to Additional or Joint Commissioners, who in turn may authorise Assessing Officers to perform the actual rectifications and issue related orders. This structure is important: it presupposes that the Commissioner (CPC) sits above, and may empower, Assessing Officers – it does not purport to re‑define the Commissioner as an Assessing Officer under section 2(7A). The general assessment powers of scrutiny under section 143(3), the power to issue notices under section 143(2), or to reopen under sections 147/148, remain tied to the statutory “Assessing Officer,” not to CIT(CPC).
Within that frame, CPC is best characterised as a centralised, largely automated processing set‑up operating under schemes and notifications, while the “proper officer” for assessment in the sense of the Act remains the jurisdictional Assessing Officer defined in section 2(7A) and notified under section 120.
Processing vs assessment: where CPC should stop
The Act itself draws a sharp line between processing of returns under section 143(1) and assessment under section 143(3), and that line is central to evaluating CPC’s role. Processing under section 143(1) is designed as a summary, largely mechanical step. The return is subjected to automated checks for arithmetical errors, internal inconsistencies, incorrect claims apparent from information on the face of the return, disallowance of losses or specified deductions in late‑filing situations, and mismatches with third‑party data such as Form 26AS, AIS or TIS. The output is an “intimation” determining tax or refund, with a notice of demand deemed under section 156 where tax, interest or fee is payable. No enquiry is undertaken, no evidence is called, and no opinion is expressed on the substantive correctness of the returned figures.
Assessment under section 143(3) is of a different character. It is preceded by a statutory notice under section 143(2) and may be supported by further requisitions under sections 142(1), 131 or 133(6). Here, the Assessing Officer examines books, hears explanations, evaluates documents and third‑party information, and then determines total income and tax by a reasoned order. In ACIT v. Rajesh Jhaveri Stock Brokers (P) Ltd., the Supreme Court emphasised that an intimation under section 143(1) “could not be treated to be an order of assessment,” and that at this stage “no opinion is formed” by the department on the returned income. Processing is thus not assessment in the strict sense.
A live example shows how critical this boundary is. Consider a discretionary family trust created under a Will, which is the only trust under that Will. On those facts, the scheme of section 164, read with its provisos, generally contemplates that such a testamentary trust enjoys an exception from the maximum marginal rate and may be taxed at appropriate slab rates in the representative capacity, subject to the specific conditions being satisfied. In the case at hand, the return of income correctly reflected the trust’s income and claimed taxation at the appropriate rate. CPC, while processing the return, retained the income figure but recomputed tax at the maximum marginal rate, issued a demand, and thereby treated the trust as if it fell outside the testamentary exception and squarely within the discretionary trust regime.
That adjustment is not an arithmetic correction or a removal of an obvious internal inconsistency. It involves a series of legal and factual judgments: whether the trust is indeed created by a Will, whether it is the only trust so created, whether beneficiaries and their shares are determinate, and how the provisos to section 164(1) apply to those facts. These are questions that the Tribunal in similar cases has remanded to the Assessing Officer for verification of the Will and trust deed before deciding the applicable rate, recognising that they require an evaluative, quasi‑judicial determination. When CPC, through an automated rule, unilaterally applies the maximum marginal rate to such a trust at the processing stage, it is no longer “processing” in the sense of section 143(1); it is, in substance, attempting an assessment‑level decision on how section 164 should operate in that fact pattern, without the procedural safeguards or the jurisdiction of an Assessing Officer.
The same pattern appears in other areas. In capital gains cases, CPC has made additions under section 50C by substituting stamp duty valuation for declared sale consideration while processing under section 143(1), without examining whether the assessee has grounds to invoke the statutory mechanism for reference to a valuation officer, or whether the consideration actually received is fully supported by banking records—issues that the Mumbai ITAT in Prabha Anil Gandhi v. ADIT held are beyond CPC’s scope at the processing stage and require assessment‑level inquiry. In cases involving carried‑forward losses, CPC has disallowed set‑off purely on the ground that the return was filed after the due date, or without distinguishing whether the loss relates to speculation business or to business where different timing rules may apply, and without evaluating explanations that an Assessing Officer could have considered in regular assessment; tribunals have treated such mechanical disallowance as inappropriate where factual verification and legal interpretation are required. Similarly, in TDS credit mismatch situations, CPC has denied credit where Form 26AS shows a different PAN or figure, without considering cases where tax was deducted but deposited with an incorrect identifier, or where the deductor’s statement contains errors—scenarios that call for verification and correction at the deductor’s and Assessing Officer’s level, rather than permanent denial of credit through an automated 143(1) adjustment.
In each of these scenarios, the common feature is that CPC’s algorithms have been used to make determinations that involve interpretation of statutory provisions, evaluation of supporting evidence, or resolution of factual disputes—functions that section 143(1) does not contemplate and that the Act reserves for assessment under section 143(3). That is where the statutory distinction between processing and assessment becomes outcome‑determinative: the more CPC steps into these interpretative and evidential decisions, the more it assumes a role that belongs to the Assessing Officer.
Algorithms, illogical tax law, and the need for real intelligence
The examples above sit at the intersection of two themes: the legal limits of section 143(1), and the practical reality that CPC’s decisions are primarily driven by software rules. Tax professionals often joke that tax law is illogical—or at least, that it is full of provisos, exceptions, deeming fictions and context‑dependent carve‑outs that do not follow linear, computer‑science logic. Coding, by contrast, is uncompromisingly logical: an algorithm lives on “if–then–else” paths and binary flags. Bridging a body of law that frequently says “X is taxable, unless A, except where B, subject to C, deemed as D” with a machine that needs crisp decision trees inevitably requires real human intelligence to interpret, sequence and sometimes temper those instructions.
CPC, by design, embodies pure logic. Its workflows are coded to say, in effect, “if trust return shows status X and answer Y in this schedule, apply rate Z,” or “if deduction exceeds statutory cap, disallow excess,” or “if Form 26AS shows higher receipts than ITR, raise a mismatch.” Where the law itself is clear and mechanical—such as a hard monetary cap or a missed due date—this kind of coding maps well onto the statute. The difficulty arises when the legal rule being encoded is not a clean “if–then,” but an “if–then–unless–subject to–deemed as” chain that presupposes an officer reading the deed, the Will, the banking records, or the surrounding facts and, sometimes, the history of the provision.
Judicial insistence on application of mind now has to be read against this technological and logical backdrop. In Mantra Industries Ltd. v. NFAC, the Bombay High Court treated a faceless assessment order that simply reproduced the draft order, without grappling with the assessee’s detailed reply, as non‑est, emphasising that such mechanical reproduction evidenced absence of application of mind and violated the mandatory faceless procedure. In other cases, High Courts have set aside orders that blandly asserted “no reply has been received” when replies were in fact on record, treating the contradiction as proof that the decision‑maker had not actually examined the file. These outcomes do not merely criticise poor drafting; they reflect a deeper expectation that when the law vests power in an officer, some genuine, case‑specific reasoning should sit behind its exercise.
Seen in this light, the tension is not between humans and machines in the abstract, but between an illogically structured legal text and a logically structured codebase. Tax law tolerates, and often requires, exceptions based on purpose, history and equitable considerations; code does not. A testamentary family trust that squarely fits a proviso to section 164 but has an innocuous field left unticked may look identical, to an algorithm, to a fully discretionary private trust. A capital gain that reflects genuine consideration but triggers a section 50C flag because of a higher stamp valuation may be indistinguishable, to a rule engine, from an under‑reported sale value warranting addition. A human Assessing Officer, reading the Will, the sale deed, the bank statements and the provisos, would at least recognise the need to resolve such ambiguities before making an adjustment; a rules engine cannot.
That mismatch has governance consequences. As more of the system becomes automated, there is a real risk that the centre of gravity of decision‑making shifts from officers to algorithms, while the law and the case law continue to speak in terms of what “the Assessing Officer” has believed, formed, considered or satisfied himself about. The more CPC‑style engines are entrusted not just with computation but with interpretative moves – whether on trust rates, capital gains valuations, loss carry‑forwards or TDS credits – the more courts are likely to be confronted, in concrete disputes, with the question whether an intimation or demand backed only by code, and not by any discernible human application of mind, meets the standards the Act and the Constitution still assume.
Conclusion / My view
On the face of the statute, the answer to whether CPC is a “proper officer” for anything beyond apparent, mechanical matters lies in two anchors: who the Act defines as an Assessing Officer in section 2(7A), and what the Act allows to be done under section 143(1) as opposed to 143(3). CPC and CIT(CPC) operate within the framework of schemes and notifications tied primarily to section 143(1) and rectification provisions; they are not recast as Assessing Officers by those instruments, nor are they given a free hand to resolve debatable issues or rate disputes at the processing stage. The examples discussed—whether trust taxation, capital gains valuations, loss carry‑forwards, or TDS credit disputes—illustrate how easily a rules‑driven system can move from acceptable automation into decisions that, in substance, belong in an assessment order after some examination of the relevant documents and the applicable statutory exceptions.
As automation deepens, these boundary cases are likely to become more frequent rather than less. The law at present still assumes that when a taxpayer’s legal position is accepted or rejected, it is because an Assessing Officer has thought about it, however briefly, and that such thinking can later be scrutinised in appeal or judicial review. The growing role of CPC and similar engines introduces an additional layer between the text of the Act and the taxpayer’s experience: the layer of encoded logic. How far that layer can be allowed to carry decisions that go beyond arithmetic without some visible imprint of real intelligence is an issue the statute does not yet answer explicitly, but which live disputes are already beginning to pose in concrete terms.
