Author is Adity srivastava a 5th year law student and co author is Arindam sood a 4th year law student at Symbiosis Law school pune.
Introduction
Corporate transacting in recent years has come to rely upon autonomous AI Agents that negotiate, draft and in certain cases conclude and execute contracts on behalf of a company and this is done without any human being reviewing the specific terms at the point of formation. Software’s that finalise purchase orders, do due diligence in mergers and acquisitions and do negotiations that respond directly on price and delivery are no longer a dream of the future, but are quintessentially a part of ordinary corporate practice today. It is against this background that a question arises, that is, whether an AI agent, acting without human intervention at the point of transactions can be said to bind the company on whose behalf it operates, in the same way that a human agent binds a principal under apparent authority doctrine.
When an AI system communicates in a manner that suggests it has authority to act for a company, replying from a corporate email address, or negotiating terms in a way that an authorised representative would, a counterparty may reasonably believe a binding agreement has been reached regardless of the fact, whether any person within the company ever turned their mind to that specific deal. Agency law, however, has traditionally presumed that an agent is a legal person who is capable of intention and consent, a premise the artificial agent does not satisfy straightforwardly. It is aptly clear that the doctrine of apparent authority, though flexible enough to be stretched to cover such situations, does so only by way of analogy and therefore not because it was ever built with this problem in mind.
But, even where courts are willing to hold a company bound by its AI agent’s conduct, a separate problem still persists, that is, whether disclosure based regimes within corporate law itself, such as section 188 of the Companies Act, 2013, which governs related party transactions, are equipped to catch a transaction that was negotiated by an AI agent rather than a director or a key managerial personnel. The existing language of Section 188, and the “related party” definition under Section 2(76) of the Act is, drafted entirely with a human actor in mind, and a transaction routed through an AI system may accordingly slip past the disclosure regime that is meant to catch precisely this Agency Law Foundations & Doctrine of Apparent Authority
Agency law in its most basic form governs the relationship between a principal and an agent, wherein, the agent is authorised to act on behalf of the principal and the principal becomes bound by the acts of the agent that are done within the scope of that authority. This doctrine is based on the idea that a principal must be held liable for the acts of an agent, so that third parties dealing with the agent are not made to bear the burden of verifying the agent’s credentials before contract. It is this very rationale that protects the third party’s reliance and gives rise to the doctrine of apparent authority, which in itself is quintessentially different from actual authority.
Actual authority arises where the principal has expressly or impliedly, communicated to the agent that the agent may act on behalf of the principal. Apparent authority, on the other hand arises not from a communication between the principal and the agent, but from a representation made by the principal to a third party, whether by words or by conduct, that the agent possesses the authority to enter into contracts and consent thereof. The leading authority on this point remains Freeman & Lockyer v. Buckhurst Park Properties (Mangal) Ltd., wherein, Lord Justice Diplock (as he was then), held that it is the principal’s own representation, and not any actual grant of authority, that binds the principal to a third party who has relied on that representation. It can be clearly seen from this statement that the doctrine does not at any point require a third party to establish that the agent subjectively believed itself to be authorised, nor does the third party require proof of a written delegation of power. And therefore, it is the outward manifestation of authority and a third party’s reliance upon it, is protected by the doctrine.
This is where the problem beings to show in the context of an AI Agent. Both actual and apparent authority presupposes an agent capable of receiving instructions, and understanding the scope of such a mandate, and in the case of apparent authority in a manner comprehensible to that third party. An AI agent does not, in true terms understand a mandate in the way a human agent does, nor does it possess the legal personality that agency law has taken as its starting premise. It is precisely because the doctrine of apparent authority focuses on outward manifestation rather than the agent’s internal state of mind that courts have found it relatively easy to stretch the doctrine to cover AI agent mediated contracts even without resolving the deeper question of whether an AI system can be said to possess authority at all.
This shows that the doctrine of apparent authority, precisely because it was built around what a reasonable third party perceives rather than what an agent intends and therefore, survives the substitution of a human agent with an artificial one far more comfortably than actual authority does.Where an AI agent corresponds to an official company e-mail address, it uses a corporate signature and negotiates terms of contracts in a manner an employee authorise to that effect would, is in fact, indistinguishable from those sent by a human agent acting with apparent authority, which can termed as a functional equivalence, that has allowed the doctrine to be applied by analogy notwithstanding the origins or the doctrine. It is precisely this analogy, however, that this essay questions in the sections that follow, since the doctrine that survives by analogy alone, without any test of whether it was designed to answer the question now being asked of it, cannot be assumed to close the very gap that arises once the agent ceases to be human.
Indian Application: Section 188 & The Related-Party Disclosure Gap
Having already examined the law of agency in abroad, by way of analogy, the law has stretched itself to accommodate an AI agent, it becomes necessary to turn to Indian law and jurisprudence, where the difficulty is not of stretching the doctrine, but of a statute that does not in true terms contemplate the role of an AI agent at all. Section 182 of the Indian Contract Act, 1872, defines an Agent, as a person who is employed to do any act for another, which on plain reading excludes an AI agent entirely, since an AI agent is quintessentially not a “person” within the meaning contemplated by the section. In the context of India, one more problem persists, that is, whether Section 188 of the Companies Act, 2013, is equipped to detect a transaction that was initiated by an AI system rather than a director.
Section 188 requires a company to obtain a board approval and in certain cases a shareholder approval before entering into specified transaction with a related party, a term defined under Section 2(76) of the Companies Act, 2013, to include directors, key managerial personnel and entities in which such other person holds a substantial interest. However, this presupposes that a related party transaction is one that a human being aware of the relationship in question has knowingly entered into on behalf of the company. Where a procurement autonomously selects a counter party, and negotiates terms and finalises a deal thereof, without any human reviewing the deal, the question that will stand in a court of will be, whether that counterparty happens to be a “related party” is never, in true terms, put to a human mind capable of triggering the disclosure obligation in the first place.
An AI Agent that is trained to select vendors on the basis of price, delivery time and other relevant data, could, select a counterparty that would qualify as a related party under Section 2(76), without the AI system possessing any means of recognising this fact as legally significant, since the AI system has no legal obligation to check for such a relationship at all.
It can be clearly seen that Section 188 imposes a disclosure obligation upon the company acting through its directors and not upon any transaction as such is unremarkable when every transaction is negotiated by a human being, becomes a genuine loophole once the negotiating party is a system that owes no fiduciary duty and bears no statutory obligation to flag the relationship at all.Related party ( Section 2(76), Companies Act, 2013),Directors, KMP & their relatives Individuals directly connected to the company Firms & companies with a stake
Partnership firms; private/public cos where director or relative holds interest Entities acting on advice Corporate group,Holding, subsidiary, associate & fellow subsidiary companies.
Body corporate or person whose board acts on the director’s or manager’s adviceBut, it may be argued that the board’s ultimate approval requirement under Section 188 already accounts for this difficulty, since no transaction, however initiated, escapes the requirement of board or shareholder ratification.Board approval, in ordinary corporate practice, is sought on the basis of what is disclosed to the board by management, and the disclosure itself depends on someone within the company having identified the counterparty as a related party in the first place. Where an AI system, rather than a human negotiator, is the one selecting and finalising the counterparty, the very information that would ordinarily trigger the disclosure to the board may never be generated at all, since the AI system has no obligation, statutory or otherwise, to flag the relationship for board consideration. The gap, therefore, does not lie in the board’s ultimate authority to approve or reject a transaction, but in the absence of any mechanism compelling the relevant information to reach the board in the first place once a human negotiator is removed from the transaction.
Illustration: Company X deploys an AI procurement agent to source packaging material, configured to select vendors on the basis of price, delivery time, and past order history alone. The agent autonomously selects Vendor Y, whose bid is marginally cheaper than its competitors. Neither the agent nor any human reviewing the resulting purchase order is aware that Vendor Y is, in fact, 28% owned by the brother-in-law of one of Company X’s directors, a relationship that would squarely bring Vendor Y within the definition of “related party” under Section 2(76) of the Companies Act, 2013. No board approval is sought, since no one within the company ever identified the relationship as one requiring disclosure in the first place, and the transaction proceeds entirely outside the framework Section 188 was designed to govern.
Therefore, an amendment to the section is much needed, it should be on these lines,
Model Section
Provided also that where a company employs an automated or artificial intelligence system (AI agent) to initiate, negotiate, or conclude any contract (as defined in the Indian Contract Act, 1972) or arrangement referred to in sub-section (1), such company shall ensure that the system is configured to verify, prior to the conclusion of such contract or arrangement, whether the counterparty falls within the definition of ‘related party’ under clause (76) of section 2 of this act; and any contract or arrangement concluded without such verification shall be voidable at the option of the company, without prejudice to the liability of the directors under sub-section (5)
Counter Argument
It may be argued that this entire essay proceeds on a false premise. Based on this view, an AI Agent is not an agent in any legal sense at all, but is merely a sophisticated instrument through which a principal continues to act and therefore, it is no different in kind from a calculator or a vending machine perhaps, and therefore, agency law when strictly construed has no role to play here whatsoever. Vending machines and AI Systems and agents have since long been treated by the law of contracts as merely the channels of an offer, wherein, the human principal’s assent is given in advance, at the point of programming rather that at the end point, that is, the transaction.
If this view is correct, the company itself remains a party to the contract and Section 188 of the companies Act, 2013, is triggered only in the same manner as it would be for any ordinary contract and since it is the company and not the AI agent that is contracting all the time.This argument, however, in the opinion of the author, proves too much. A vending machine’s conduct is entirely predetermined by its programming and it therefore does not have any discretion. An AI agent by contrast selects a counterparty an negotiates a deal, which it thereafter finalises and all this is done without any human intervention and therefore, a degree of discretion that the instrumentality analogy simply cannot account for. It is clearly seen that treating such conduct as indistinguishable from a fixed tool that is mechanical, does not, in true terms resolve the disclosure gap identified hereinabove, it merely relocates the same problem back onto a human who, by hypothesis, was never consulted on the contract at all.This goes on to show that the instrumentality argument, whatever force it may carry in the narrow context of contractual attribution, offers no answer whatsoever to the disclosure problem that lies at the centre of this essay, since a human principal cannot be said to have identified a related party that it never turned its mind to in the first place.
Conclusion
This essay was set out to examine whether an AI agent, acting without a direct human intervention at the point of the contract can be said to bind a company in the same manner that a human agent binds a principal under apparent authority. It can be clearly seen that the doctrine survives comfortably, since it turns both outward manifestations rather than the agent’s own capacity for intention, and both English and American law have found room to accommodate this without any reform in the statute.The real difficulty lies in total silence of India’s related party disclosure jurisprudence on contracts negotiated by an AI agent, and this is a gap that the instrumentality counterargument does not answer.Therefore, this shows that, what is needed is not an overhaul of the agency law at all but, an amendment to Sections 188 that requires companies that deploy AI Agents to have a related party check at the very beginning of the transaction rather than relying on humans to notice the relationship after the contract. Quintessentially, making this as an oversight waiting to be corrected.