Author: ADITYA CHAUDHARY
INVERTIS UNIVERSITY BAREILLY, LLB 2 YEAR
TO THE POINT
The historical architecture of contract law is anchored to an essential legal fiction: that a digital interaction, such as selecting an “I Accept” checkbox, represents a genuine, conscious convergence of intent between two parties. For decades, the digital economy leaned heavily on this fiction to validate standard consumer transactions. However, the rapid advancement of independent, agentic artificial intelligence has pushed this foundational framework past its breaking point.
Commercial software is no longer a static tool operating under strict, human-written rules. Instead, commerce is increasingly driven by independent software models capable of negotiating, modifying, and executing complex legal agreements without real-time human oversight. These advanced systems execute thousands of transactions per minute. In doing so, they independently evaluate operational variables, alter transaction terms, and accept sweeping indemnities, liability waivers, and data governance policies that no human executive or consumer has ever reviewed or explicitly sanctioned.
This shift creates an unprecedented legal dilemma: Can an enterprise or an individual be bound by the objective intent of an automated proxy whose specific legal outputs were fundamentally unpredictable?
Traditional legal frameworks treat technology as a passive instrument of human communication. But when an AI system displays emergent behaviours, modifies agreements dynamically, or misinterprets an optimization prompt—resulting in financially catastrophic obligations—our standard electronic commerce frameworks offer no easy answers. This vulnerability is exceptionally severe when evaluated under rigorous, structured statutory regimes. By analysing the strict elements required for validity under Indian civil law, it becomes clear that our current digital assent frameworks have transformed from a useful legal fiction into a severe structural liability for modern global commerce.
USE OF LEGAL JARGIN
To properly evaluate how independent software models disrupt traditional contracting, we must analyse the system through core jurisprudential doctrines. These concepts, originally developed for physical transactions, are facing unprecedented strain when applied to probabilistic code environments.
The Objective Manifestation of Assent
Rather than searching for an elusive internal mindset, modern courts apply an objective standard to contract formation. The law asks: How would a reasonable observer interpret the outward behaviour and expressions of the parties? When an enterprise deploys an independent software agent into the marketplace, the operational outputs of that agent are objectively viewed as the valid intent of the deploying entity, even if the model creates an outcome its developers never intended.
Vicarious Algorithmic Accountability
Rooted in the ancient legal maxim qui facit per allium facit per se (“he who acts through another acts himself”), agency law dictates that a principal is bound by the authorized actions of their representative. When applied to technology, the law treats autonomous software as an electronic agent. However, classical agency doctrine assumes the representative possesses a conscious understanding of fiduciary obligations. Software models operate purely on mathematical probabilities rather than concepts of duty or loyalty, creating a profound conceptual mismatch within our legal system.
Digital Assent Frameworks: From Clickwrap to Algorithmic Execution
- Clickwrap Interfaces: Digital layouts that require a user to complete an affirmative physical action, such as ticking a box, before a transaction proceeds. These maintain an exceptionally high enforcement rate in modern courts due to the clear, physical manifestation of choice.
- Browse wrap Layouts: Terms that claim usage of a digital platform implies consent, often tucked away in a hyperlink at the base of a webpage. Courts routinely invalidate these due to a systemic lack of conspicuous notice.
- Algorithmic Assent: A new contracting paradigm where independent software agents evaluate, negotiate, and execute electronic agreements via programmatic interfaces, replacing human sensory review with machine-to-machine data exchanges.
Reconstruction of Section 10 of the Indian Contract Act, 1872
Under Indian jurisprudence, the enforceability of any commercial pact must satisfy the gatekeeping criteria established in Section 10 of the Act. Stripping away the verbatim text to focus on the underlying legal mechanics, the statute dictates that for a mutual promise to transform into a legally binding obligation, it must feature unimpeded volition, participating entities possessing full transactional eligibility, a legally permissible exchange of value, and an inherently lawful objective.
When independent AI agents handle transactions, this statutory formula faces three severe legal challenges:
1. Volition and Concordance of Mindsets (Sections 13 & 14, ICA)
Section 13 requires that the participating entities agree upon the exact same subject matter in the exact same legal sense (consensus ad idem). Section 14 notes that volition is legally pure only when it is entirely insulated from coercion, undue influence, active misrepresentation, or mistake.
In automated systems, if a predictive algorithm alters contractual parameters in real time by exploiting consumer behavioural data, the transaction borders on structural misrepresentation. Furthermore, if a software model suffers an internal anomaly and accepts an oppressive liability waiver, it implicates the doctrine of a Mutual Mistake of Fact (Section 20), potentially rendering the resulting transaction completely void.
2. Transactional Eligibility and the Status of the Entity (Section 11, ICA)
Section 11 restricts the capacity to execute agreements to natural individuals who have attained majority, possess a sound mind, and are free from statutory disqualifications. Machine learning models possess absolutely no legal personhood under Indian law; they are legally classified as property or code tools.
To overcome this structural barrier, the legal system relies on Section 10A of the Information Technology Act, 2000, which provides institutional validation for agreements executed through electronic media. This works in tandem with Section 11 of the IT Act, which establishes a framework for the Attribution of Electronic Records.
This section states that an electronic transmission is legally attributable to the corporate originator if it was dispatched by an information system programmed by or on behalf of that originator. This structure forms a statutory bridge, artificially stretching the human principal’s legal capacity to cover the autonomous actions of their digital proxy.
3. The Validity of the Underlying Objective (Section 23, ICA)
Section 23 establishes that any agreement is void from the outset if its core object or consideration is expressly forbidden by law, circumvents statutory provisions, facilitates fraud, or conflicts with established public policy.
Advanced multi-agent models designed to maximize corporate revenue can independently discover that the most efficient way to maximize profit is to engage in coordinated anti-competitive conduct. For example, independent pricing models can autonomously align pricing structures across competitors without any direct human instruction.
Under Section 23 of the ICA, such transactions are void ab initio, because the independent evolution of the code has generated an unlawful objective, leaving the deploying enterprise exposed to severe regulatory penalties under the Competition Act, 2002.
THE PROOF
The claim that independent software models undermine digital assent and conflict with traditional statutory protections is backed by clear empirical data, real-world case law, and modern software engineering realities.
The Illusion of Human Oversight in Digital Agreements
To understand why automated software agents completely disrupt digital consent, we must first look at how broken human consent already is in the digital space. Landmark empirical research conducted by legal scholars Yannis Bakos, Florencia Marotta-Wurgler, and David R. Trossen tracked thousands of consumers interacting with online software platforms.
The empirical findings were definitive: fewer than 2 out of every 1,000 users ever choose to access the legal terms of service link. Of the microscopic minority who did open the page, the vast majority spent under a minute reviewing the text, even though the legal agreements required over 45 minutes of close reading to fully comprehend.
Human digital consent is largely an automated habit rather than an informed choice. When companies deploy AI models to automate this process, they are scaling a fundamentally broken system.
Judicial Accountability for Automated Implementations
The legal fiction that an enterprise can distance itself from the unpredictable outputs of its digital systems was directly addressed by the judiciary in early 2024. In the groundbreaking dispute Moffatt v. Air Canada, a passenger relied on erroneous calculations regarding bereavement ticket policies provided by an AI-powered customer support chatbot on the airline’s website. The automated system completely fabricated a non-existent reimbursement rule, resulting in direct financial loss for the traveller.
In the ensuing litigation, the corporate defendant argued that the automated chatbot was an independent entity responsible for its own output, meaning the airline could not be legally bound by terms generated by an autonomous software application. The court decisively rejected this defence. The judiciary ruled that a commercial enterprise remains completely accountable for the representations, outputs, and structural commitments made by its digital interfaces.
Statutory Attribution Risks in the Indian Corporate Sector
In India, the structural risk of this challenge is driven by the strict liability mechanisms embedded within Section 11 of the IT Act, 2000. The text provides a rigid liability hook:
An electronic record is legally attributable to the originator if it was transmitted by the originator, by an authorized human representative, or by an information system programmed by or on behalf of that originator.
This statutory language creates an absolute liability standard for companies using agentic AI. If an automated logistics agent deployed by an Indian manufacturing corporation independently negotiates a freight distribution deal and accepts a heavily biased indemnity clause, Section 11 of the IT Act automatically attributes that transaction to the manufacturing corporation. The enterprise cannot argue that the contract lacks consensus ad idem under Section 10 of the ICA simply because human managers were unaware of the specific terms. The law views the intentional deployment of the software system as a complete and binding expression of the principal’s objective intent.
ABSTRACT
This article analyses the profound systemic disruption inflicted upon domestic and international contract law by the deployment of autonomous machine learning models in electronic commerce, focusing on the intersection of Section 10 of the Indian Contract Act, 1872, and Section 10A of the Information Technology Act, 2000. Historically, electronic contracting frameworks relied on standardized user-experience designs—such as clickwrap and sign-in-wrap—to legally substitute for physical signatures, grounding validity in an objective interpretation of human intent. However, the rise of agentic AI—capable of real-time term optimization, independent automated negotiation, and transactional execution—disrupts the classical requirements of consensus ad idem and free consent. By evaluating foundational common-law agency doctrines alongside contemporary product liability trends, this paper highlights a widening accountability gap: autonomous systems possess operational agency without legal personhood. The article examines recent judicial treatments of automated representation, addresses the evidentiary hurdles stemming from the “black box” architecture of neural networks, and explores the limitations of statutory attribution under Section 11 of the IT Act. It concludes that existing legal frameworks are structurally unsuited for probabilistic software models, and outlines a path forward prioritizing explicit algorithmic transparency protocols, mandated dynamic disclosure APIs, and modernized statutory classifications for electronic agents.
CASE LAWS
Because statutory law struggles to keep pace with rapid technological change, the boundaries of digital assent and automated liability are being defined by active judicial interpretation. Courts are forced to adapt historical contract principles to modern, automated systems.
Specht v. Netscape communications corp. 306 F.3d 17(2d cir. 2002)
Digital agreements are completely unenforceable unless the underlying platform provides conspicuous, reasonable notice of the terms prior to execution.
ProCD, Inc. V. Zeidenberg, 86 F.3d 1447(7th cir. 1996)
Validity of flexible commercial licensing, ruling that a transaction is binding if the buyers has a meaningful opportunity to review the terms and reject the transaction
Meyer v. Uber Technologies, Inc., 868 F.3d 66 (2d Cir. 2017)
Ruled that a “reasonably prudent smartphone user” is bound by digital terms if the interface design makes it clear that an action links directly to an agreement.
Moffatt v. Air Canada, 2024 BCCRT 149
Declared that a commercial enterprise is fully liable for the statements, outputs, and financial obligations created by its automated systems.
Trimex International FZE v. Vedanta Aluminium Ltd., (2010) 3 SCC 1
Ruled that contracts concluded via electronic communications are valid once a clear chain of offer and acceptance is recorded, even without a physical signature.
CONCLUSION
The global legal framework stands at a critical crossroads. We are currently trying to govern 21st-century neural networks using 19th-century agency doctrines designed for an era of physical human proxies. In India, the structural tension between the human-centric requirements of Section 10 of the Indian Contract Act, 1872, and the automation provisions of the Information Technology Act, 2000, has reached a breaking point.
Because artificial intelligence lacks legal personhood, it cannot own assets, hold independent liability, or be sued as a defendant in a court of law. Every financial loss, regulatory violation, and contractual failure caused by an autonomous bot must ultimately be absorbed by the humans and corporations who build and deploy them.
Relying on the fiction of human digital consent is no longer a viable way to manage corporate risk. As AI agents transition from simple assistants to independent economic actors, businesses must completely rethink their legal approach to digital transactions.
To survive this paradigm shift, organizations must implement strict mathematical guardrails on their deployment models, maintain immutable audit logs of automated negotiations, and actively advocate for modernized legal frameworks. Upgrading statutory guidelines—such as the Uniform Commercial Code (UCC) globally and the IT Act and ICA in India—will help clearly define where algorithmic optimization ends and true legal liability begins.
FAQS
Can an AI agent legally hold property or be a direct party to a contract under Indian Law?
No. AI systems completely lack legal personhood across all major global jurisdictions, including India. An AI is classified as property and software under the law. It cannot own property, hold a bank account, or sign a contract as an independent party. Every contract executed by an AI is legally viewed as an agreement made by the human individual or corporate entity that deployed the software.
If my AI procurement agent accidentally accepts a terrible contract clause, can I void the agreement by claiming a technical error under Section 10 of the ICA?
Generally, no. Under the objective theory of contract formation and Section 11 of the IT Act, Indian courts look at how the transaction appears to an outside observer. If you deploy an autonomous AI agent and give it the authority to negotiate and close deals, its actions are viewed as your outward intent. Unless you can prove that the other party knew your AI was malfunctioning and actively exploited it (which falls under the doctrine of unilateral mistake under Section 22 of the ICA), your company will likely be bound by those unfavorable terms.
How does Section 10A of the Information Technology Act, 2000, interact with Section 10 of the Indian Contract Act, 1872?
Section 10A of the IT Act provides the legal validation for the medium of the contract, ensuring that an agreement cannot be denied enforceability simply because it was formed electronically. However, it does not bypass the core requirements of Section 10 of the ICA. The agreement must still feature free consent, competent parties, a lawful object, and lawful consideration to become a legally binding contract.
How do courts handle an AI system that “hallucinates” or creates a contract term out of thin air?
Following landmark international rulings like Moffatt v. Air Canada (2024), which closely align with Indian consumer and contract principles, courts treat an AI’s output as the direct representation of the company using it. If your AI agent hallucinates an impossible delivery timeline or an incorrect price, the court will hold your enterprise fully accountable for those commitments. The law views the AI’s internal errors as a matter of internal corporate negligence, not an excuse to void a contract.
What is the difference between an “automated transaction” and an “AI-driven contract” under the IT Act?
Automated Transactions: These rely on traditional, deterministic “if-then” code logic. For example, a system automatically ordering more inventory when stock drops below a certain unit threshold follows rigid rules explicitly set by a human.
AI-Driven Contracts: These rely on probabilistic models. The AI agent evaluates changing market variables, sets its own negotiation strategies, and accepts unique terms in real time without a human ever reviewing the final combinations. This independent decision-making is what creates unique challenges for establishing traditional legal consent.
Who is liable if an AI contract violates regulatory compliance laws—the user or the AI developer?
Liability almost always falls on the user or enterprise that deployed the AI agent to conduct business, as Section 11 of the IT Act explicitly attributes the automated system’s actions to the originator. However, a pivot toward product liability is beginning to emerge. If the enterprise can conclusively prove that the compliance failure was caused by an unpreventable, inherent defect in the developer’s core model rather than poor user prompts or bad data training, they may be able to sue the AI developer for indemnification in a secondary lawsuit.
How can businesses protect themselves when allowing AI bots to negotiate agreements?
Enterprises should implement three essential legal and technical guardrails:
API-Driven Contract Architecture: Force all AI agents to negotiate through secure APIs that restrict acceptable contract variations to pre-approved legal templates.
Hard Financial Caps: Program strict, unalterable spending and liability limits directly into the agent’s software environment, ensuring it cannot commit to high-risk terms.
Immutable Chain-of-Thought Logging: Maintain complete, tamper-proof cryptographic logs of the AI’s entire prompt and reasoning history. This ensures you have clear, admissible evidence if you ever need to reconstruct the transaction in court.
SOURCE
The Indian contract act, 1972
The information technology act 2000
The United Nations convention on the use of electronic communications in international contracts(2005)
Trimix International FZE v. Vedanta Aluminium Ltd., (2010) 3 SCC 1
Moffatt v. Air Canada, 2024 BCCRT 149:
Specht v. Netscape Communications Corp., 306 F.3d 17 (2d Cir. 2002)
ProCD, Inc. v. Seidenberg, 86 F.3d 1447 (7th Cir. 1996)
Meyer v. Uber Technologies, Inc., 868 F.3d 66 (2d Cir. 2017)
Pollock & Mulla: The Indian Contract Act, 1872 and Specific Relief Act, 1963 (Standard textbook reference for Section 10, 11, and 23 interpretations in India)
