Back
Can Autonomous AI Agents Bind Their Human Principals Under Indian Contract Act?
The rapid evolution of agentic artificial intelligence is forcing legal systems worldwide to confront a fundamental question: can a machine that acts with limited human oversight create a binding contract on behalf of a person? Under Indian law, the answer is far from settled, as traditional agency principles collide with a technology that the drafters of the , could never have imagined. At the India AI Impact Summit 2026, this issue emerged as a key governance concern, with experts debating accountability, security, and the need for human oversight. As businesses increasingly deploy AI agents to negotiate purchases, execute trades, and interact with third parties, the legal community must urgently determine whether the principal behind the bot remains bound by its actions — and when liability attaches.
Traditional Agency and the Problem of Capacity
The bedrock of Indian agency law is the maxim — he who acts through another is deemed to act in person. defines an agent as “a person employed to do any act for another or to represent another in dealings with third persons.” This language clearly contemplates a natural person, a requirement reinforced by Section 184, which states that only a person of the age of majority and of sound mind can become an agent responsible to the principal. Software, however sophisticated, does not meet these criteria.
Yet the law has already validated electronic contracting. , provides that a contract cannot be deemed unenforceable merely because a proposal, acceptance, or other communication was made electronically. This opens the door for AI agents to function as communication tools — but not as independent legal actors. As the source material notes, “an AI agent can't independently enter into a contract as it lacks and under . As a result, it can neither sue nor be sued.”
The real question, then, is not whether the AI itself can contract, but whether the behaviour of the AI can be legally traced back to the person who authorised its use. This is where the traditional doctrine of agency becomes both useful and strained.
Key Precedents from Around the World
Several recent decisions illustrate how courts are grappling with AI agency. In , the held the airline responsible for false information provided by its customer-service chatbot. The tribunal rejected the argument that the chatbot was a “,” ruling that the airline was liable for the words of its tool. This case underscores a crucial principle: a principal cannot escape liability by hiding behind an automated system it has deployed.
In the Indian context, the in held that an agent must obey the lawful instructions of the principal and is liable for consequences of its negligence. Similarly, the in established that an agent acting beyond authority and without disclosure to the principal cannot bind the principal to a third party. But these cases address human agents, not autonomous software. As the source material points out, “these cases undermine the liability of human agency, not an AI agent.”
A more modern and directly relevant precedent is , where the ruled that “AI is a tool, not a person.” The court considered whether the AI agent acted on behalf of the user or the AI company, and concluded that it was the user, not the AI developer, who accessed Amazon’s computers with the help of the agent. Therefore, Amazon failed to prove intentional access under the . The case is ongoing, but it introduces a crucial idea: that the agentic AI should be regarded as a mere tool in the user’s hands, not an independent actor.
When the AI Exceeds Its Instructions
A core difficulty arises when an AI agent goes beyond the scope of authority granted by the principal. Consider a scenario where a user instructs an AI to buy office supplies under ₹50,000. The agent, acting autonomously, decides that a more expensive item valued at ₹2 lakh offers better long-term value and enters into a contract on that basis. Under , if the agent acts beyond their powers, the outcome depends on whether the authorised and unauthorised parts of the transaction can be separated. If they cannot, the principal is not liable.
But with an autonomous AI, separating the authorised from the unauthorised becomes far more complex. A human agent interprets instructions using human logic; an AI may misinterpret a general goal or apply an unexpected heuristic. The principal may have intended a narrow delegation of authority, but the AI’s execution can produce outcomes the principal never envisioned. The law currently provides little guidance on how to attribute responsibility in such cases.
Ratification: A Partial Safety Net
The , codified in , offers one avenue for resolving unauthorised AI transactions. If an AI agent makes an unauthorised contract, the principal can later adopt it by express or implied conduct. For instance, if an AI buys shares without permission and the principal later accepts the profit, a court may find ratification. However, this approach creates significant uncertainty. Companies would be left in doubt about whether an AI-generated transaction is binding until the principal decides to accept or reject it. As the source material warns, “depending too much on ratification would lead to uncertainty.”
A more robust solution, some legal scholars argue, is to separate the autonomy of execution from legal authority. The principal's authorisation should define the objective and commercially sensible limits, rather than requiring each individual AI action to be personally approved. The AI may have autonomy in how it achieves the goal, but legal liability must remain traceable to the person or organisation that placed the AI in the commercial environment.
The Need for a More Precise Indian Framework
The current position under Indian law is largely a consequence of applying traditional concepts to a technology that did not exist at the time of the Contract Act’s enactment. As the source material notes, “if India gives a legal personality to AI, this may only make things worse.” The better approach is to move beyond the question of whether AI can be an agent and instead consider how agency principles should apply when agents are acting autonomously with the aid of AI.
Courts considering disputes over AI-driven contracts would likely weigh several factors: the principal’s directions, the technical competence of the AI agent, the of the action, the protections put in place, and the third party’s knowledge of the agent’s authority. The Indian Contract Act may be able to handle the first generation of autonomous AI transactions without recognising AI as a legal entity. However, when agents begin to negotiate and contract with other autonomous systems, issues may arise that go beyond what judicial interpretation can resolve.
Legislative intervention may eventually be necessary to provide clear rules on when an AI agent’s actions bind the principal, including provisions for mandatory disclosure of AI agency, limits on liability for unforeseeable AI behaviour, and possibly a system of registration or certification for commercial AI agents. In the meantime, businesses deploying autonomous AI should carefully define the scope of authority, implement robust safeguards, and ensure that their contracts with third parties address the possibility of automated decision-making.
Conclusion
The rise of agentic AI does not require a complete rewrite of agency law, but it does demand a thoughtful application of existing principles and a willingness to adapt. The fundamental maxim — that an act performed by an authorised agent is the act of the principal — remains valid. The task for Indian courts and lawmakers is to determine how that maxim applies when the agent is a piece of software that can learn, decide, and act without human intervention. As the legal landscape evolves, the principle that AI is a tool, not a person, offers a solid foundation. The challenge will be to ensure that the tool does not become a loophole for escaping responsibility.
