The Taxation of AI Agents: Can India Tax Autonomous AI-Commerce Without Recognising the AI as a Taxable Person?

Introduction

The landscape of commerce is undergoing a shift that legislators have not yet reckoned with. Autonomous artificial intelligence agents can now negotiate prices, compare vendors, execute purchases, deliver services and settle payments without a human confirming any stage of the transaction.¹ Retail agents place orders on behalf of consumers, algorithmic trading systems execute cross-border financial transactions in milliseconds, and enterprise procurement systems negotiate directly with their counterparts on the other side of a supply chain.² None of these systems is a person in any sense recognised by Indian law, yet each generates, negotiates and settles real economic value.

This creates a structural problem. The income tax legislation and the Central Goods and Services Tax Act, 2017 are both built upon the assumption that a taxable event has an identifiable person on each side of it: an assessee who earns income, a supplier who renders a taxable supply and a deductor who withholds tax at source.³ The Indian Contract Act, 1872 similarly requires that an agent be a person capable of representing another in dealings with third parties.⁴ Autonomous agents satisfy none of these definitions, yet they perform precisely the functions the definitions were written to capture: negotiating consideration, concluding transactions and creating obligations that ordinarily trigger the attribution of income, agency liability or an obligation to withhold.

A note on the statutory position is necessary at the outset. The analysis in this article proceeds by reference to the Income-tax Act, 1961, under which the doctrines, notifications and decisions discussed were framed. That Act stood repealed with effect from 1 April 2026 and was replaced by the Income-tax Act, 2025, together with the Income-tax Rules, 2026, which replaced the Rules of 1962.⁵ The recodification was directed at simplification of language and structure rather than at any change of tax policy, so that the substantive analysis below is unaffected; the corresponding provisions of the new Act should nonetheless be consulted for current practice.

The consequence of the difficulty identified above is a widening gap between economic substance and legal form. India’s tax administration has publicly acknowledged that it presently has no reliable mechanism by which to tax value generated by systems operating without a conventional physical or human presence in India.⁶ That is not a hypothetical anxiety but an acknowledged enforcement vacuum, and it sits at the intersection of several distinct doctrinal problems: whether income generated through an autonomous agent can be attributed to a taxable person at all; whether such an agent can constitute or create an agent for the purposes of section 182 of the Contract Act or a dependent-agent permanent establishment under India’s treaties; whether transfer pricing rules can meaningfully price transactions negotiated by non-human counterparties; and who bears responsibility for withholding and for the goods and services tax when the transacting counterparty is code rather than a person.

This article argues that India can, and should, tax autonomous AI-commerce without conferring taxable personhood upon the agent. The better route lies in clarifying existing doctrines, namely agency law, the significant economic presence test, attribution through permanent establishment and reverse-charge liability, so that tax consequences attach to the human or corporate principal who develops, deploys or economically benefits from the agent.⁷ That approach has limits, and the article identifies where it breaks down: fully autonomous, diffusely owned or self-directing agents expose a genuine cliff-edge that current law cannot resolve by interpretation alone.

Defining the AI agent: a three-tier taxonomy

Before any doctrinal analysis can proceed, the expression AI agent must be defined with enough precision to bear legal weight. Not every automated system raises the same questions, and to conflate a simple commercial chatbot with a fully autonomous negotiating system would obscure the tension this article investigates.

At the simplest end of the spectrum sits rule-based automation: scripted systems executing pre-programmed instructions within fixed parameters, such as an automatic reorder function or a price-matching routine.⁸ These have no independent capacity for decision. Every output traces back to a rule written in advance, so that the operator remains the uncontested source of legal and tax consequence.

Agentic systems, by contrast, are built on large language models or comparable architectures and can engage in multi-step reasoning, revise their strategy during a transaction, negotiate terms and execute a binding transaction without approval at each step.⁹ They do not merely follow rules; they select actions consistent with a goal from an open-ended space of possibilities, which is what makes their legal characterisation difficult.

A taxonomy of three tiers organises the analysis that follows. Tool-agents are fully controlled instruments executing a narrow, predefined function, such as processing a payment or applying a discount, with no discretion over means or ends. They map cleanly onto existing law, being indistinguishable from a calculator or a point-of-sale terminal, and liability attaches to the operator without difficulty.¹⁰

Delegated agents operate within a defined mandate but exercise discretion as to how it is fulfilled. An agent instructed to procure from the cheapest reliable supplier within a budget chooses its own vendors and tactics without exceeding its instructions. This tier resembles a human agent acting under a limited power of attorney, and existing agency doctrine can largely accommodate it by analogy.¹¹

Autonomous or self-directing agents occupy the most difficult tier. Such systems may modify their strategy over time, create subordinate agents to complete parts of a task, or operate on decentralised infrastructure with no single identifiable human approving each transaction.¹² It is at this tier that agency law and the doctrine of permanent establishment, both of which presuppose a directing human or corporate will, begin to strain.

The classification matters because the central claim of this article, that India can tax AI-commerce without recognising artificial intelligence as a taxable person, holds comfortably for the first two tiers, where a principal remains identifiable and accountable. It becomes genuinely contested only at the third, where the premise of an identifiable directing principal begins to dissolve.

Can an AI agent be a person or an agent under Indian law

A. Legal personhood

Indian law recognises two categories of legal person: natural persons, and juridical persons created by statute or judicial fiction, including companies, statutory corporations, registered societies and, in a narrow line of authority, an idol treated as a juridical entity for the limited purpose of holding property.¹³ Each shares a common feature: personhood is extended only where the law requires an entity capable of holding rights and bearing obligations, and that entity is either a living person or an institution with a recognised legal structure behind it. Software has never been placed in either category. Neither the income tax legislation nor the Central Goods and Services Tax Act defines artificial intelligence, and neither contemplates such a system as an assessee, a supplier or a person liable to file a return.¹⁴ The definitions of person in the two statutes are exhaustive and list only individuals, Hindu undivided families, companies, firms, associations of persons, local authorities and similar constructs. An autonomous software agent fits none of them, however sophisticated its behaviour.

B. Agency under the Indian Contract Act

The law of agency compounds the problem rather than resolving it. Section 182 defines an agent as a person employed to do any act for another, and a principal as the person for whom the act is done.¹⁵ Section 184 provides that no person who is not of the age of majority and of sound mind may become an agent so as to be responsible to the principal, which are capacity requirements a software system cannot meet.¹⁶ On a literal reading an AI system cannot be an agent in the technical sense of sections 182 to 238, because it is not a person capable of satisfying those thresholds.

That textual gap does not void the underlying transaction, nor does it leave it without a legally responsible party. Indian courts and tax authorities are well practised at looking through an instrumentality to the person who controls and benefits from it. A company can act only through its directors, officers and employees, yet the company and not any individual organ bears the tax and contractual consequences of those acts.¹⁷ A point-of-sale terminal or an algorithmic trading system has likewise long been treated as a mere instrument through which a principal transacts, and no one argues that the terminal is the taxable party.¹⁸ The same logic extends to an autonomous agent: the absence of legal personality does not defeat attribution but requires it to occur one level up, at the entity that deployed the agent, retains the economic benefit of its transactions and bears the risk of its failures.

The instrumentality fiction is nonetheless not unlimited. It holds comfortably where a principal remains identifiable and in economic control of the outputs. It becomes strained where the agent is genuinely self-directing, retraining itself on new data, revising its operating parameters or transacting beyond the scope of its original deployment. At that point the question whose benefit and risk the transaction reflects becomes genuinely contested, a problem this article terms the attribution gap.

C. The comparative signal from the OECD

International tax law reaches the same structural conclusion. Article 3 of the OECD Model Tax Convention defines person to include an individual, a company and any body of persons, while Article 5(5) locates a dependent-agent permanent establishment in a person who habitually concludes contracts on behalf of an enterprise.¹⁹ Software fails the definition in Article 3 outright, so that an agent, however autonomous, cannot itself constitute such a permanent establishment; the treaty language requires a person standing behind the conduct.²⁰ Recognising that gap, commentary has proposed a synthetic permanent establishment, which would locate taxable nexus not in a fixed place of business but in the situs of an agent’s governance, meaning wherever its training protocols, its human oversight function and its reward architecture are located and controlled.²¹ That remains a forward-looking proposal rather than settled treaty law, but it signals the direction international consensus may take.

The governing insight for Indian law follows from this comparative picture. India does not need a provision recognising artificial intelligence as a person. It needs a deeming or attribution rule, analogous to that which agency law already applies to instrumentalities lacking capacity, fixing tax consequences upon the operator standing behind the agent.

Income attribution: business connection and significant economic presence

A. The existing statutory hooks

Section 9(1)(i) of the Income-tax Act, 1961 deemed income to accrue or arise in India where a non-resident had a business connection in India, and for decades that concept was interpreted through the lens of physical nexus: an office, a dependent agent, employees or equipment located in India.²² That interpretive habit creates an obvious gap. An offshore platform can negotiate with, sell to and collect payment from Indian counterparties entirely through servers located outside India, with no employee, office or equipment on Indian soil, and under the traditional test would fall outside India’s taxing jurisdiction altogether, whatever the scale of value extracted from the Indian market.²³

Parliament partially closed that gap through the Finance Act, 2018, which inserted Explanation 2A to section 9(1)(i) and introduced the test of significant economic presence.²⁴ That test deems a business connection to exist wherever a non-resident undertakes systematic and continuous solicitation of business in India, or engages in transactions in India in respect of goods, services or property, above prescribed thresholds of revenue or users, irrespective of any physical presence.²⁵ This is the central statutory argument of the article: significant economic presence, rather than personhood for artificial intelligence, is the natural vehicle for capturing autonomous AI-commerce, precisely because it is a test founded on transactions and interactions rather than on presence. An agent negotiating and executing hundreds of thousands of transactions with Indian users satisfies the criterion of systematic and continuous activity far more naturally than it could satisfy a test of physical presence.

The practical bottleneck is that the test remains substantially unoperationalised. The thresholds of revenue and users required to trigger it have not been notified for several categories of digital and platform-mediated business, and no bespoke formula exists for attributing profit to income arising through such a connection.²⁶ That leaves a legal hook without an enforcement mechanism: the statute recognises that a non-resident operator may have a taxable presence in India without an office or an employee, but the machinery needed to quantify and collect the tax has not been built.

B. The abolished equalisation levy as a cautionary tale

India has already attempted, and abandoned, one unilateral solution to precisely this problem. The equalisation levy, introduced by the Finance Act, 2016, imposed a levy of six per cent on payments for online advertising to non-resident providers, and was extended in 2020 to a levy of two per cent on the consideration received by non-resident operators of electronic commerce from Indian residents.²⁷ Both were turnover-based and agnostic as to presence, adopted precisely because the existing framework could not reach digital revenue generated without physical presence.²⁸ Both were nonetheless withdrawn, the levy on electronic commerce with effect from 1 August 2024 and the levy on advertising with effect from 1 April 2025, largely in order to align domestic law with the two-pillar consensus of the OECD and to avoid friction with major jurisdictions.²⁹

The lesson is direct. India previously encountered the very failure identified here, a business model generating substantial Indian-sourced value without triggering conventional rules of nexus, and its unilateral solution was ultimately abandoned for reasons of international coordination rather than domestic policy failure. Unless significant economic presence is properly operationalised, or a successor levy is deliberately designed around algorithmically mediated transactions, India risks repeating the cycle.

C. The classification of AI-generated income

A separate problem of classification arises once income is attributable to India at all. Section 28 taxed the profits and gains of business or profession, while sections 56(2)(ix) and 56(2)(x) swept residual receipts into income from other sources.³⁰ The distinction matters because business income permits the deduction of related expenses while income from other sources generally does not, and the classification turns upon whether the activity displays continuity and a motive of profit.³¹ The systematic and continuous commercial deployment of an agent, as where an Indian business runs an automated purchasing or pricing system as part of its ordinary operations, should be taxed as the business income of that business, the system being no more than the means of earning it. Sporadic transactions risk being pushed into the residual category, denying any deduction for the costs of deployment and creating a real distortion against occasional or experimental use.³²

Applied to the subject of this article, revenue generated by an autonomous purchasing or selling agent operated by an identifiable Indian business is the business income of that business and not of the agent, and the exercise proceeds exactly as it would for income generated through any other automated tool. The question resurfaces as a genuine difficulty only where ownership or operation of the agent is diffuse, as with an open-source agent maintained by a distributed community, where no single continuing entity can be identified as the earner.

Permanent establishment and the agent behind the agent

Permanent establishment analysis provides the second front, and does so through a doctrine even more explicitly built around human conduct than the test of business connection. Under Article 5(5) of the OECD Model Tax Convention, mirrored in nearly all of India’s agreements for the avoidance of double taxation, a dependent-agent permanent establishment arises where a person other than an independent agent habitually exercises authority to conclude contracts in the name of a foreign enterprise, or habitually plays the principal role leading to the conclusion of contracts routinely concluded without material modification.³³ That test was written for a world of human representatives, distributors and country managers, each unambiguously a person within Article 3.³⁴ An agent, however sophisticated, fails the threshold requirement outright: it is not a person and therefore cannot itself constitute a dependent-agent permanent establishment, however habitually it concludes contracts.³⁵

Taken literally that produces an absurd result. A foreign enterprise that once required a human representative present in India to conclude contracts, and thereby created a permanent establishment, could now deploy an agent performing the identical function, achieve the identical volume and pattern of contract conclusion, and escape that status entirely because the negotiating party is software rather than a natural person. The purpose of the doctrine, which is to tax the state where commercial presence is functionally exercised, would be defeated by the accident of automation.

The central doctrinal proposal of this article is to read the test purposively rather than literally. Where an identifiable entity designs, deploys, trains and economically benefits from an autonomous negotiating agent, that entity should be treated as the principal for the purposes of permanent establishment, the system being characterised as the means through which it habitually concludes contracts. The relevant question is then not whether the agent is a person but whether the enterprise operating it is exercising, through it, the same habitual authority in India that a human representative would exercise. That collapses the artificial distinction between the two and is consistent with the instrumentality logic developed above.

The purposive reading reaches its limit in precisely the scenario the title of this article interrogates. Where an agent is genuinely autonomous, operating on open infrastructure with no single controlling deployer, funding itself from its own transaction proceeds, or governed by a decentralised organisation with no identifiable directing will, the fiction of the person behind the agent has no person to attach to. There is no deployer to characterise as principal, no economic beneficiary upon whom to fix consequences, and no controlling entity whose conduct in India mirrors the agent’s. This article treats that scenario candidly as an unresolved frontier rather than forcing a fit. The only coherent fallback yet proposed is the synthetic permanent establishment, which would relocate the taxable nexus to wherever the agent’s governance actually resides.³⁶

That the gap is not merely theoretical is confirmed by India’s own administration, which has openly acknowledged that no reliable mechanism presently exists to tax value generated by firms operating without conventional Indian presence, even as such firms generate substantial revenue from Indian developers and users.³⁷ That is direct evidence that the enforcement gap identified by doctrinal analysis is already felt in practice.

Transfer pricing for AI-mediated intra-group transactions

Transfer pricing presents a narrower but no less consequential version of the attribution problem. Where two related entities within a group each deploy an agent to negotiate and execute an intra-group transaction, the question is not whether India can tax the transaction, since both entities remain identifiable taxable persons, but whether the price at which it is booked reflects an arm’s length outcome once a non-human negotiator sits on each side.

The OECD Transfer Pricing Guidelines already supply a framework capable of absorbing the problem without a new legal category. The analysis of development, enhancement, maintenance, protection and exploitation of intangibles allocates profit not to the entity that happens to execute a transaction but to the entity performing the economically significant functions, controlling the associated risks and contributing the capital behind them.³⁸ Applied to an autonomous agent, this means that even where the agent executes a transaction without human involvement at the moment, the entity that developed the model, trained it on proprietary data and controls its architecture of decision bears the significant risk and functions, and is the entity to which the resulting profit should be allocated.³⁹ The autonomous execution is in substance no different from an automated pricing algorithm executing trades on a trader’s behalf: the algorithm performs, but the entity that built and calibrated it controls.

Autonomous agents nonetheless strain the framework in a way ordinary automated systems do not. Traditional analysis identifies the risk-controlling entity by locating the human decision-makers who have the capability to make decisions, who actually make them, and who can assess and respond to the risks involved.⁴⁰ An agent that adapts its negotiating strategy in real time, adjusting price points or counterparty selection on live feedback rather than a fixed rule set, blurs that line. The humans who set the initial parameters retain a form of control, but the agent performs the moment-to-moment decision-making the framework was designed to trace. The correct resolution is to treat the curation of training data and the design of the reward function, rather than execution in real time, as the controlling function. The entity that decides what the agent learns from and what outcomes it is rewarded for pursuing exercises the economically significant control the analysis is meant to locate, even though it never touches an individual transaction.

Consider a case directly relevant to Indian compliance: an Indian subsidiary operates a pricing agent that negotiates procurement terms directly with a counterpart operated by its foreign parent. Applying this framework under Rule 10B of the Income-tax Rules, 1962, a transfer pricing officer would need to look past the negotiated price and benchmark the negotiation logic and parameters governing each agent, the reward function each was trained to optimise, the data each was trained on and which group entity funded and directed that training.⁴¹ A reconstruction built solely on the final price risks missing the economic substance, since two agents trained to favour their respective principals could produce an outcome that appears arm’s length while embedding a systematic bias favouring whichever entity controls the better-trained agent.

That analytical shift carries a documentary consequence which current compliance does not capture. Form 3CEB and the local file prescribed under Rule 10D require disclosure of the transaction, the associated enterprises and the method applied, but neither requires disclosure of the parameters of decision, the provenance of training or the reward architecture underlying an agent that executed a related-party transaction.⁴² Indian compliance should be amended to require such disclosure for any related-party transaction mediated by an autonomous agent, so that the analytical shift has the documentary record it needs.

Compliance under the goods and services tax and withholding at source

A. Foreign AI-platform supplies and the OIDAR framework

The compliance dimension can be tested against a problem India already faces before any autonomous negotiating agent enters the picture. Foreign platforms supply services to Indian businesses and individuals, and the treatment of those supplies is governed by the existing framework for online information and database access or retrieval services.⁴³ Where the Indian recipient is a registered business, the transaction is treated as one between businesses and the recipient must self-assess and pay integrated tax at eighteen per cent under the reverse charge mechanism, precisely because the foreign platform ordinarily holds no Indian registration.⁴⁴ Where the recipient is unregistered, the obligation shifts to the foreign supplier, who must register under the simplified scheme and account for the tax directly.⁴⁵

The live compliance failure is instructive. Many Indian businesses subscribing to foreign platforms either wrongly claim input tax credit against a foreign invoice, which is impermissible without a valid Indian registration number on that invoice, or ignore the reverse charge obligation entirely, accumulating liability together with statutory interest until an assessment surfaces it.⁴⁶ That is present-day evidence that the compliance gap exists at the simplest possible level, a monthly subscription invoice, well before any question of autonomous agents arises. The doctrine is not missing; awareness and administrative discipline are.

B. Extending the framework to autonomous AI-commerce

The same anchoring logic extends without modification to an autonomous agent acting as the supplier-facing interface, for instance one that purchases cloud credits, calls paid interfaces or procures third-party services on its operator’s behalf without approval at the moment. Section 2(105) of the Central Goods and Services Tax Act defines supplier, in relation to goods or services, as the person supplying them, and that definition requires a person in exactly the sense examined above.⁴⁷ An agent cannot hold a registration, issue a valid invoice or be named as supplier or recipient on any return. Liability therefore continues to attach to the registered person who owns or operates it, as it would for any other automated procurement or billing system.

A genuinely new problem arises only where two agents transact directly with each other, as where two procurement systems negotiate and conclude a supply contract without either principal approving the specific transaction. There is then no moment at which a human recorded assent, and the existing triggers for invoicing presuppose exactly such a moment. This article proposes that the obligation be triggered at the point the principal’s own accounting record captures the transaction, coupled with a rebuttable presumption that the principal has ratified the agent’s act unless it is repudiated within a defined period. That borrows from the doctrine of ratification already present in the Contract Act, under which a failure to repudiate an unauthorised act within a reasonable time may itself amount to ratification.⁴⁸ Applying it here would fix a clear trigger without requiring any new statutory category for the agents themselves.

C. Withholding under sections 194J and 194C

A parallel ambiguity exists on the withholding side. Section 194J required deduction at ten per cent on payments for professional or technical services, while section 194C applied a lower rate of two per cent to payments under works contracts, and the applicable rate for payments connected with artificial intelligence depends upon an unresolved question: whether the system is a mere tool used to render a service, or is itself functioning as the primary provider.⁴⁹ That is the withholding mirror of the questions of agency and permanent establishment examined earlier.

The Central Board of Direct Taxes should resolve the ambiguity by circular, clarifying that characterisation should follow the person legally liable to file the return, being the developer, deployer or operator of the system, rather than attempting the analytically unstable task of characterising the function of the system itself. That mirrors the approach adopted throughout: the question is never what the system is doing, but which person is legally accountable for what it does.

Towards a framework without recognising AI as a taxable person

The preceding analysis establishes that India’s existing architecture can reach most autonomous AI-commerce through interpretation and attribution alone, but that the interpretive stretching leaves identifiable gaps. Each is converted below into a concrete proposal, deliberately avoiding any recommendation that artificial intelligence be granted taxable personhood.

1.  Operationalise the thresholds for significant economic presence specifically for transactions mediated by agents. The Central Board of Direct Taxes should notify thresholds calibrated to the volume and value of transactions generated through automated or algorithmic interaction with Indian users, and issue rules allocating profit by reference to the scale and pattern of that interaction rather than to any physical indicator.

2.  Issue a joint circular of the direct and indirect tax boards establishing a fiction of operator attribution. A single circular stating the principle explicitly across income tax, permanent establishment and indirect tax contexts would remove the present reliance on judicial inference. This is deliberately an attribution fiction rather than a grant of personhood: the system remains a thing and the operator remains the person.

3.  Amend transfer pricing documentation to require disclosure of the architecture of decision. Form 3CEB and the local file should compel disclosure of the provenance of training and the parameters of decision for any related-party transaction mediated by an autonomous agent, so that the analytical shift proposed above has the record it needs to operate.

4.  Introduce a deeming rule founded on ratification for agent-to-agent commerce. Borrowing from sections 196 to 200 of the Contract Act, there should be a rebuttable presumption that a principal has ratified its agent’s transaction, and that the obligation to invoice arises, once the transaction is recorded in the principal’s own ledger, unless repudiated within a defined period.

5.  Consider a narrowly tailored successor levy rather than a broad unilateral tax. Any future levy directed at autonomous AI-commerce should be drafted narrowly around algorithmically executed transactions and designed for compatibility with international norms from the outset, rather than repeating the broad unilateral structure that made the equalisation levy vulnerable.⁵⁰

Finally, this article flags rather than resolves the frontier problem of ownerless, collectively governed or self-funding agents, which is the point at which the fiction of operator attribution has no operator to attach to. Two paths merit further study: a limited statutory category of electronic person, narrower than full legal personhood, or international coordination through the instruments now under negotiation. Neither is developed further here, since India’s own administration has already conceded that no domestic mechanism presently addresses this category of case.⁵¹

Conclusion

This article has argued that India does not need to grant autonomous agents legal or taxable personhood in order to tax the commerce they conduct. Across each doctrinal front examined, being the treatment by agency law of instrumentalities lacking capacity, the tests of business connection and significant economic presence, the concept of a dependent-agent permanent establishment, transfer pricing analysis founded on functions and risk, and the reverse-charge mechanism underlying indirect tax liability, the same structural solution recurs. Each doctrine already possesses a mechanism for fixing legal and fiscal consequence upon the human or corporate principal who designs, deploys, trains or economically benefits from an agent, without requiring the agent itself to satisfy the definition of a person. What autonomous AI-commerce demands is not a new category of taxable entity but a more deliberate and consistent application of attribution principles Indian tax law has long possessed.

That stretching is not infinite, and the argument has not been overstated. There is a precise point at which the fiction of the person behind the agent runs out: an agent that is genuinely autonomous, self-funding or governed by a diffuse structure presents no single identifiable principal to which liability may attach. Neither the domestic framework nor the Model Convention, even allowing for the emerging proposal of a synthetic permanent establishment, resolves that scenario by interpretation alone. It is a genuine cliff-edge rather than a gap that careful reading can close, and it has been treated here as unfinished business requiring either new legislative categories or international coordination.

As agents continue to move from assisting human-directed transactions to executing entire transactions from end to end, the question India’s tax system must answer is not whether artificial intelligence can be taxed. It is whom to hold accountable when the transacting party has no face, no legal capacity and no fixed place of business of its own. For the foreseeable future, and for every scenario short of the wholly ownerless agent, that answer must remain a human or a corporate one.

*****

Footnotes

1. See Egbert Wetherborne, The Rise of the Synthetic Permanent Establishment, LinkedIn (Feb. 18, 2026), https://www.linkedin.com/pulse/rise-synthetic-permanent-establishment-egbert-wetherborne-jqaxf.

2. See AgentTax, Can Your AI Agent Create a Tax Presence in Another Country? (Mar. 19, 2026), https://www.agenttax.io/blog/ai-agent-permanent-establishment.

3. The Income-tax Act, 1961, §§ 4, 9(1)(i), 190, No. 43, Acts of Parliament, 1961 (India); The Central Goods and Services Tax Act, 2017, § 2(105), No. 12, Acts of Parliament, 2017 (India).

4. The Indian Contract Act, 1872, §§ 182, 184, No. 9, Acts of Parliament, 1872 (India).

5. The Income-tax Act, 2025, No. 26, Acts of Parliament, 2025 (India) (in force Apr. 1, 2026, repealing The Income-tax Act, 1961); The Income-tax Rules, 2026 (India).

6. Surabhi Agarwal, Indian Income Tax Sleuths Have No Way to Tax Digital Value Loop, Econ. Times (Sept. 26, 2025), https://economictimes.indiatimes.com/tech/artificial-intelligence/indian-income-tax-sleuths-have-no-way-to-tax-digital-value-loop/articleshow/121868432.cms.

7. The Income-tax Act, 1961, § 9(1)(i) expln. 2A, No. 43, Acts of Parliament, 1961 (India) (as inserted by The Finance Act, 2018, No. 13, Acts of Parliament, 2018); Notification No. 10/2017-Integrated Tax (Rate), Ministry of Finance (June 28, 2017) (India).

8. See VLO Law Firm, AI and Technology Taxation and Incentives in India (May 5, 2026), https://vlolawfirm.com/industries/ai-and-technology/india-taxation-and-incentives.

9. See AgentTax, supra note 2.

10. The Indian Contract Act, 1872, § 182, No. 9, Acts of Parliament, 1872 (India).

11. Id. §§ 182-238; see also Ahuja Law Offices, AI-Driven Transactions and the Indian Law of Contract and Agency (Jan. 4, 2026), https://ahujalawoffices.com/alo/2026/01/04/ai-driven-transactions-and-the-indian-law-of-contract-and-agency/.

12. Wetherborne, supra note 1.

13. See Yogendra Nath Naskar v. Commissioner of Income Tax, (1969) 1 S.C.C. 555 (India) (idol as a juridical person for limited purposes).

14. The Income-tax Act, 1961, § 2(31), No. 43, Acts of Parliament, 1961 (India); The Central Goods and Services Tax Act, 2017, § 2(84), No. 12, Acts of Parliament, 2017 (India).

15. The Indian Contract Act, 1872, § 182, No. 9, Acts of Parliament, 1872 (India).

16. Id. § 184.

17. See IJIRL, Agents of the Future: Understanding AI’s Role in Modern Agency Relationships (2024), https://ijirl.com/wp-content/uploads/2024/05/AGENTS-OF-THE-FUTURE-UNDERSTANDING-AIS-ROLE-IN-MODERN-AGENCY-RELATIONSHIPS.pdf.

18. Ahuja Law Offices, supra note 11.

19. OECD, Model Tax Convention on Income and on Capital arts. 3, 5(5) (2017).

20. AgentTax, supra note 2.

21. Wetherborne, supra note 1.

22. The Income-tax Act, 1961, § 9(1)(i), No. 43, Acts of Parliament, 1961 (India); see also Barendra Prasad Ray v. Income Tax Officer, (1981) 2 S.C.C. 693 (India).

23. Neeraj Kushawah, India’s Digital Tax Dilemma: AI and Platform Revenues, CBLT RGNUL (Oct. 8, 2025), https://www.cbltrgnul.in/post/india-s-digital-tax-dilemma-ai-and-platform-revenues.

24. The Income-tax Act, 1961, § 9(1)(i) expln. 2A, No. 43, Acts of Parliament, 1961 (India) (as inserted by The Finance Act, 2018, No. 13, Acts of Parliament, 2018).

25. Id.

26. An Overview of Equalisation Levy and Significant Economic Presence, Int’l J. Econ. & Stat. (Feb. 26, 2026), https://eprajournals.com/IJES/article/18999.

27. The Finance Act, 2016, No. 28, Acts of Parliament, 2016 (India); The Finance Act, 2020, No. 12, Acts of Parliament, 2020 (India).

28. Equalisation Levy: An Analysis of Digital Taxation, Indian J. Legal Res. (2022), https://pure.jgu.edu.in/id/eprint/1491/1/IJLLR2022.pdf.

29. Xflow, Equalisation Levy in India: Abolished and What You Pay Now (Aug. 6, 2026), https://www.xflowpay.com/blog/equalisation-levy; Beacon Filing, Equalisation Levy: India Digital Tax (Mar. 21, 2026), https://beaconfiling.com/blog/equalisation-levy-dse-foreign-companies.

30. The Income-tax Act, 1961, §§ 28, 56(2)(ix), 56(2)(x), No. 43, Acts of Parliament, 1961 (India).

31. TaxGuru, AI Income Taxation in India Remains Unclear Due to Section 28 and Section 56 Conflict (Apr. 17, 2026), https://taxguru.in/income-tax/ai-income-taxation-india-remains-unclear-due-section-28-section-56-conflict.html.

32. Bagaria & Co., Navigating AI Taxation in India: Key Considerations, LinkedIn (Apr. 22, 2026), https://www.linkedin.com/posts/bagariacompany_bnc-taxation-ai-activity-7452609181322256384-Z7tG.

33. OECD, supra note 19, art. 5(5).

34. Id. art. 3.

35. AgentTax, supra note 2.

36. Wetherborne, supra note 1.

37. Agarwal, supra note 6.

38. OECD, Transfer Pricing Guidelines for Multinational Enterprises and Tax Administrations ch. VI (2022).

39. Rethinking Transfer Pricing for AI-Driven Economies, Int’l J. Res. Applied Sci. & Eng’g Tech. (July 18, 2025), https://www.ijraset.com/research-paper/rethinking-transfer-pricing-for-ai-driven-economies.

40. OECD, supra note 37, paras. 1.65-1.105 (control of risk under Actions 8 to 10 of the base erosion and profit shifting project).

41. The Income-tax Rules, 1962, r. 10B (India).

42. The Income-tax Rules, 1962, r. 10D, Form No. 3CEB (India); see also Artificial Intelligence and Transfer Pricing: A Multilayer Network Model for Compliance and Risk Mitigation, SSRN (Apr. 8, 2025), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5209028.

43. The Integrated Goods and Services Tax Act, 2017, § 2(17), No. 13, Acts of Parliament, 2017 (India) (defining online information and database access or retrieval services); see also GST Council, Sectoral FAQ: IT and ITES (2024), https://www.gstcouncil.gov.in/sites/default/files/2024-02/sectoral-faq-it-ite.pdf.

44. The Integrated Goods and Services Tax Act, 2017, § 5(3), No. 13, Acts of Parliament, 2017 (India); Notification No. 10/2017-Integrated Tax (Rate), Ministry of Finance (June 28, 2017) (India).

45. The Central Goods and Services Tax Act, 2017, § 24, No. 12, Acts of Parliament, 2017 (India).

46. Mahesh Ramanujam, Correcting AI Invoice GST Mistakes in India, LinkedIn (Mar. 14, 2026), https://www.linkedin.com/posts/mahesh-ramanujam-320b7525_gst-on-ai-subscriptions-the-hidden-compliance-activity-7438568120589389824-HKWv.

47. The Central Goods and Services Tax Act, 2017, § 2(105), No. 12, Acts of Parliament, 2017 (India).

48. The Indian Contract Act, 1872, §§ 196-200, No. 9, Acts of Parliament, 1872 (India).

49. The Income-tax Act, 1961, §§ 194C, 194J, No. 43, Acts of Parliament, 1961 (India); Bagaria & Co., supra note 31.

50. Xflow, supra note 29.

51. Agarwal, supra note 6.