Who Pays When Agent Earns? Autonomous AI and Attribution Gap in Indian Tax Law

Summary: Artificial intelligence agents are becoming increasingly capable of operating with limited human supervision, raising difficult questions about how existing Indian tax laws attribute income, supplies and digital assets. The experience of Truth Terminal and the GOAT memecoin illustrates the problem: developers, funders, token issuers, AI-operated accounts and traders may participate in a single economic chain without the AI itself having legal personality. Under the Income-tax Act, 2025, an AI agent does not presently fall within the recognised categories of a “person” merely because it operates autonomously. Similar attribution difficulties can arise under GST because the law requires an identifiable supplier. Virtual digital assets add another dimension where tokens may accumulate in wallets associated with AI personas even though legal ownership and control ultimately rest elsewhere. Existing provisions governing virtual digital assets determine the applicable tax treatment but do not independently resolve who should be regarded as the holder of an AI-linked wallet. The article argues against treating AI systems themselves as taxpayers and instead proposes explicit operator-attribution rules for income tax, clarification of supplier attribution under GST and a wallet-control principle for VDAs. These measures could preserve the existing human-centred tax framework while reducing uncertainty as autonomous systems assume greater economic functions.

Introduction

In 2024, an experimental AI account called Truth Terminal became an unlikely financial story. Its creator, researcher Andy Ayrey, saw it inspire the Goatseus Maximus (GOAT) memecoin, which turned a crude joke into roughly US$1 billion in wealth. The details matter, though. Ayrey has acknowledged that the AI operates semi-autonomously and often relies on human discretion, and that neither he nor Truth Terminal created GOAT. The coin was launched independently by a follower, and Truth Terminal endorsed it.

It is worth a tax practitioner’s attention because of that messiness, not despite it. A builder, a funder, an anonymous token issuer, an AI account with a wallet, and thousands of traders all took part in one chain of value. This article does not claim that AI agents are taxable entities in India. It makes a narrower claim: Indian tax law answers the question “who is taxed?” by pointing to a human operator, and that answer becomes harder to apply as the operator’s role shrinks.

I look at three pressure points: the definition of “person” under income tax, supplier identification under GST, and the treatment of tokens held in wallets linked to AI agents. I then propose three reforms.

1. The Default Rule, and Why It Is Mostly Right Today

Most AI-generated output today is taxed as the output of a tool. A company that uses a coding assistant to produce software is taxed on the revenue it earns. The assistant is not an assessee, just as a spreadsheet is not.

This works because of a basic feature of Indian law. A law that does not recognize AI as a legal person cannot hold AI liable, so liability flows to a human. The Information Technology Act, 2000 reflects the same logic in another context: its attribution rule treats an electronic record as that of the originator where it is sent by an information system programmed by or on behalf of the originator to operate automatically.

The tool model is a sound default but the question is how far it can stretch.

2. Income Tax: The Person Problem

Definition of “Person” Under the Income-tax Act, 2025

The Income-tax Act, 2025 came into force on 1 April 2026. It defines “person” in Section 2(77), and it retains the same seven categories as Section 2(31) of the 1961 Act: individual, HUF, company, firm, AOP/BOI, local authority and artificial juridical person.

The residual category, “artificial juridical person”, is where an AI agent might be expected to fit. It does not as artificial juridical persons are entities that are not natural persons but are separate entities in the eyes of law, such as deities and idols. Public corporations established under special Acts are another example. What they share is formal legal recognition, whether by custom, courts or statute. Till now, no Indian statute or reported ruling confers anything similar on an AI system. An autonomous agent is therefore not a “person”, and it cannot be an assessee.

Why Human Attribution Becomes Difficult

The tool model handles this by taxing the human behind the agent, but the allocation gets harder in three situations.

First, the operator is not one person. In the Truth Terminal story, one party built the system, another funded it, a third issued the token, and a community traded it. “Whoever deployed it” does not identify a single taxpayer.

Second, the agent holds assets. Reports differ on the details as one account says the chatbot cannot trade or create tokens itself, and that Ayrey controls its wallet. Another reports that the account’s holdings reached about $20 million. Either way, value accumulates at an address associated with an AI persona. Income tax law needs to know whose income that is.

Third, the operator’s decisions shrink. A deployer who gives one instruction and then watches an agent make thousands of operational choices is taxed on outcomes they did not individually decide. Tax law has no principle for this, because it has never needed one. Income is taxed in the hands of whoever earns it, and the law has always assumed that person made the decisions.

None of this requires treating AI as a taxpayer. In fact, doing so would be unhelpful: an AI “assessee” has no assets, cannot file returns, and cannot be prosecuted. What is missing is a clear statutory rule on attribution.

3. GST: Who Is the Supplier?

Supplier Identification for AI-Generated Supplies

GST is built around the supplier. A supplier is the person supplying goods or services, and a taxable supply must be made for consideration by a person in the course or furtherance of business. GST’s definition of “person” also includes artificial juridical persons, so it has the same gap as income tax.

Take an AI agent that builds and sells a software product, or a service, to Indian customers. Software services generally attract GST at 18%, so the tax rate is not in doubt. The identity of the supplier is.

• The model provider supplied general capability, not the specific output, so treating it as the supplier would be like taxing a tool manufacturer for the work done with the tool.
• The operator is the natural candidate. A fair objection, though, is that the operator is simply the supplier, as with any software tool, and that nothing needs reform. For ordinary deployments I agree. The difficulty arises when no identifiable operator is making the supply, or when several parties share the operator role, as in the Truth Terminal chain.
• The customer is the recipient, not the supplier.

Cross-Border AI Services and Digital Taxation

The cross-border dimension adds to this. India’s digital taxation has moved away from the Equalisation Levy. India abolished the 2% e-commerce levy in August 2024 and the 6% advertising levy in April 2025. The remaining tools, such as significant economic presence for income tax and the OIDAR provisions of the IGST Act, all assume a foreign supplier that can be identified. If an AI agent offers services to Indian users with no identifiable human behind the supply, those tools have nothing to attach to.

4. Crypto: The Wallet Problem

Taxation of Virtual Digital Assets

India has a developed regime for virtual digital assets. Income from the transfer of any virtual digital asset is taxed at a flat 30% under Section 115BBH of the 1961 Act, continued under the Income-tax Act, 2025, and a 1% TDS applies on transfers. The charging provision now sits in Section 194(1) of the 2025 Act, and the TDS in Section 393(1).

AI-Linked Wallets and Attribution of Income

For investors who bought and sold GOAT, the answer is clear: Indian VDA tax applies to them regardless of how the token came into being. The open question is the wallet associated with the AI persona. Tokens received by an AI-linked wallet, whether as gifts, endorsement-linked allocations or trading gains, raise the same question as Part 2: whose income is it? Existing VDA rules assume a human holder.

5. Three Reforms

Reform One: Operator-Attribution Rule Under Income Tax

Reform one: an operator-attribution rule in the Income-tax Act, 2025. Rather than recognising AI systems as persons, Parliament should add a deeming provision. Income generated by an autonomous system should be treated as the income of its operator, defined by a hierarchy: the person who controls the system’s keys, wallet or deployment, and failing that, the person who commissioned it or receives its economic benefit. The aim is to prevent the allocation question from being resolved by default in the taxpayer’s favour.

Reform Two: GST Clarification for AI-Made Supplies

Reform two: a GST clarification on AI-made supplies. The GST Council or CBIC should issue a clarification that a supply made through an autonomous system is treated as made by the registered person on whose behalf, or for whose benefit, it operates, with the “course or furtherance of business” test applied to that person. A circular can clarify the existing law but cannot extend it. If the operator hierarchy needs statutory force, a short explanation to the supplier definition would be the cleaner route.

Reform Three: Wallet-Control Rule for VDAs

Reform three: a wallet-control rule for VDAs. CBDT guidance, backed by the crypto-asset reporting framework in the 2025 Act, should provide that tokens held at an address controlled by or operated for an AI system are treated as held by the person who controls the keys. This closes the gap that the Truth Terminal story exposes, and it requires no change to the VDA tax rate.

Each reform has the same design: do not treat AI as a taxpayer, but make the human attribution explicit rather than leaving it to inference.

Conclusion

It would be easy to frame this as a problem of AI outrunning the law. That would overstate it. The tool model works for nearly all current deployments, and the Truth Terminal story is a case of experimental, semi-supervised AI, not a rogue economic actor. Research simulations point the same way. In Altera’s Project Sid, an arXiv preprint, 25 constituent agents, 3 influencers and an election manager operated under a constitution that set a 20% tax, but the legal system was supplied to the agents in advance. These are illustrations of agent behaviour under rules, not evidence that AI can design fiscal systems unaided.

The real point is more modest and more practical. As agents act with less supervision, the gap between “who decided” and “who is taxed” will widen. India’s tax statutes were written on the assumption that those are the same person. Clear attribution rules, adopted before a dispute forces the issue, would cost little and spare taxpayers and the department a great deal of uncertainty.