Law Of Agency And AI Agent: Can An Autonomous AI Agent Bind Its Human Principal?
“Agentic AI” is rapidly transforming the relationship between AI and the legal and commercial world. An AI agent can decide on a series of actions, engage with external systems, make choices, and execute actions with limited human involvement, unlike traditional generative AI, which typically reacts to a user's instruction. Agentic AI emerged as a key governance concern at the India AI Impact Summit 2026, focusing on accountability, security, and human oversight. This raises an intriguing issue under Indian contract law that can an autonomous AI agent create a binding contract on behalf of a person? It involves whether an autonomous system is capable of carrying out a legal function that traditionally has to be carried out by an agent and, if the autonomous system cannot, whether the human behind the autonomous system is capable of being bound by its actions.
Traditional Concept of Agency under Indian Law
The principle of agency is primarily based on legal maxim 'Qui facit per alium facit per se' meaning acts performed by an authorised agent are treated as acts of principal. According to Section 182 of the Indian Contract Act, 1872, an “agent” is a person employed to do any act for another or to represent another in dealings with third persons. The person for whom the act is done or represented is the “principal”. Further, Sections 183, 184, 185, 186, 187 & 188 cover appointment of agents and extent of authority granted to them. Section 186 acknowledges express and implied authority. The real problem lies under Section 184. It provides that, as between principal and third persons, any person may become an agent, but a person who is not of the age of majority and of sound mind cannot become an agent to be responsible to the principal. The statutory language therefore, seems to refer to a natural person and not to software. An AI agent does not have to be a legal entity to act in a technological capacity to enable the principal to communicate with third parties. The law has already validated the ability of electronic systems to enable the creation of contracts. Section 10A of the Information Technology Act, 2000 acknowledges that a contract cannot be deemed unenforceable simply because a proposal, acceptance or other communication was made electronically. So, the better question is that do AI agents have capacity to contract? Can the behaviour of the AI be legally traced back to the person who authorised its use? The answer is no, an AI agent can't independently enter into a contract as it lacks legal personhood and contractual capacity under Indian Contract Act, 1872. As a result, it can neither sue nor be sued. However, automated contracts have a strong legal backbone under Uniform Electronic Transactions Act (UETA) Section 14 and Electronic Signatures in Global and National Commerce Act (ESIGN) Section 101, makes it clear that the use of an electronic agent to form a contract is not a basis for denying it legal effect or enforceability. In the Moffatt v. Air Canada[1] case, airline's customer-service chatbot told a passenger that he could apply for a bereavement fare after the trip, which was not the airline's policy, and British Columbia's Civil Resolution Tribunal ruled that the airline was responsible for the words of its chatbot, rejecting the idea that the bot was a 'separate legal entity'.
How and when does an AI Agent bind itself to its Principal?
For instance, the user instructs an AI agent to buy office supplies when they are under ₹50,000. The agent observes several websites, finds a product, contacts the seller and accepts an offer of ₹48,000. The human has not actually made a purchase, but rather intentionally delegated purchasing authority to the AI agent. The difficulty arises when the AI goes beyond the instructions given to it. If AI decides it would be better to spend ₹2 lakh on more expensive item because of the added value it offers in the long run, whereas principal approves spending of not more than ₹50,000. Would the principal be held liable? Sections 227 and 228 provide minor clarity on this point. If the agent acts beyond their powers, the outcome of such an incident depends on whether the authorised and unauthorised aspects of the transaction can be distinguished. If the excess cannot be separated, the principal will not be liable for the transaction. But with the advent of an autonomous AI agent, this classification becomes more challenging. Human agents typically interpret and make decisions based on human logic. An AI system can misinterpret a general goal. In the landmark case of Pannalal Jankidas v. Mohanlal[2], the Supreme Court held that an agent must obey to lawful instructions of principle and is liable for compensation and consequences of its negligence. Similarly, in Keighley, Maxted & Co. v. Durant[3], the House of Lords held that if agent acts beyond authority and doesn't disclosed to principal, the principal can't sue third party. But again, the issue still persists as these cases undermine the liability of human agency, not an AI agent.
The issue of ratification and AI autonomy
The doctrine of ratification provides an alternative remedy. The statutory conditions under Sections 196 to 199, allows a principal to adopt an act if it is performed without authorisation. The ratification may be either express or implied. This may be relevant if an AI agent makes an unauthorised transaction, but the principal then agrees to its advantages. For example, if an AI agent buys shares without explicit permission, and the principal is aware of the transaction and benefits from it, the law could review the situation to determine if it constitutes ratification or not. But depending too much on ratification would lead to uncertainty. It would leave companies in doubt if an AI-generated transaction is binding or not until the principal accepts or rejects it. If the AI performs actions that are within a specific scope of authority and the person authorising its use, the actions of AI are to be considered as those of the principal. There should be a separation between the autonomy of execution and legal authority based on the principal's authorised objective and commercially sensible limits, instead of on whether or not each AI action was personally authorised. In a prominent precedent, Amazon.com Services, LLC v. Perplexity AI, Inc.[4], the Ninth Circuit gave a crucial principle that “Al is a tool, not a person”. The primary issue here was that the agentic Al is acting on behalf of whom? The company that created it (Perplexity) or the individual users who are directing it? The Ninth Circuit further ruled that it is the user, not Perplexity, who accesses Amazon's computers, with the help of the AI agent. Therefore, Amazon fails to prove the CFAA violation's threshold requirement i.e. intentional access to the computer. Since the Al agent is not a person, it should be regarded just a tool in the user's hands. The case is still ongoing. However, each step introduces a new agentic Al jurisprudence.
The need for a more precise framework of Indian Law
The present position is largely a consequence of applying traditional legal concepts to a technology that did not exist at the time of enactment of the Contract Act. In the Indian legal context, there is a growing debate revolving around concerning AI-driven transactions and agency. If India gives a legal personality to AI, this may only make things worse. The legal landscape has already raised questions about AI agents, contract formation, authority and liability. It is important to move beyond the question of can AI be an agent. Instead, consider how principles of agency should be applied when agents are acting autonomously with the aid of AI. If AI agent goes beyond those bounds, courts would consider factors including principal's directions, technical competence of the AI agent, foreseeability of the action, protections put in place, and knowledge of third party of the agent's authority. An AI agent may have autonomy in executing tasks that have been delegated to it, but legal liability for the conduct of such a task should be traceable to the person or organisation that placed the AI agent in the legal and commercial environment. The Indian Contract Act may thus be able to deal with the first generation of autonomous AI transactions without the need to recognize AI as a legal entity. However, when agents start to negotiate and buy, contract and communicate with other autonomous systems, there could be issues beyond what judicial interpretation can achieve.