1. Introduction

Artificial intelligence is changing the way contracts are drafted, negotiated, executed, interpreted, monitored, and enforced. Businesses increasingly use AI systems to generate contractual clauses, review commercial agreements, assess counterparty risk, automate procurement, monitor performance, and execute transactions through smart contracts.

AI contract governance is the legal and institutional framework governing the use of AI throughout the contractual lifecycle. It addresses whether an AI-generated agreement is legally binding, who has authority to enter into a contract, how errors and misleading outputs are treated, who bears responsibility for automated decisions, and what remedies are available when an AI-supported contractual process causes loss.

The central civil-law question is:

When an AI system contributes to the formation, interpretation, performance, or breach of a contract, which party should bear the resulting legal responsibility?

Potentially responsible parties include:

Contracting parties: for their own promises, representations, performance, and breaches.

AI developers and providers: for contractual failures, misleading assurances, or negligence where legally established.

Businesses deploying AI: for inadequate supervision, unauthorised automated commitments, or failures in their own contractual processes.

Intermediaries and agents: for conduct within their actual or apparent authority.

Smart-contract developers and platform operators: depending on their contractual obligations, control, representations, and applicable law.

The use of AI does not automatically invalidate a contract. Nor does automation automatically excuse a party from performance. The legal outcome depends on the governing jurisdiction, the parties' intentions, the authority of the person or system acting for them, and the circumstances of the dispute.

2. Core principles of AI contract governance

2.1 Contract formation and consent

Traditional contract law generally examines offer, acceptance, consideration where required, intention to create legal relations, capacity, and compliance with any applicable formalities.

An AI system may generate or communicate an offer, recommend acceptance, or transmit a purchase order. A key issue is whether the system was authorised to act on behalf of the business and whether its conduct objectively manifested contractual assent.

A company should not assume that every AI-generated statement creates a binding contract. Equally, it should not assume that an automated acceptance is legally ineffective merely because no person manually clicked an approval button.

2.2 Authority and attribution

AI systems are not ordinarily treated as independent legal persons merely because they can make decisions or conduct transactions autonomously. Contractual rights and duties usually attach to the natural or legal persons operating through the system.

Important questions include:

Who deployed and configured the AI?

What authority was delegated to it?

Were transaction limits or approval requirements established?

Did the counterparty reasonably understand that the system could commit the business?

Was the transaction within the system's actual or apparent authority?

A business that allows an AI procurement system to place orders may be bound by transactions made within its authorised scope, depending on the applicable agency and contract rules.

2.3 AI-generated errors and misrepresentation

AI systems may produce inaccurate descriptions, incorrect prices, fabricated contractual terms, or misleading summaries of obligations.

A legal dispute may involve mistake, misrepresentation, fraud, negligent advice, breach of warranty, or failure to disclose material information. These are distinct doctrines, and their elements vary by jurisdiction.

An AI-generated error does not automatically entitle a party to cancel an agreement. The legal consequences depend on the nature of the error, its materiality, the parties' knowledge and reliance, and the remedies permitted by law.

2.4 Performance and breach

AI may monitor delivery schedules, calculate payments, assess service-level compliance, or trigger contractual remedies. Errors in data or system configuration may lead to wrongful rejection of performance, incorrect invoicing, or premature termination.

Contract governance should establish who verifies automated decisions, how disputes are escalated, and whether human review is required before imposing significant consequences.

2.5 Smart contracts

A smart contract is code that automatically executes specified actions when programmed conditions are satisfied. It may implement some or all of a legally enforceable agreement, but code and legal contract are not necessarily identical.

A smart contract can raise questions about programming errors, ambiguous terms, unauthorised transactions, remedies for unintended execution, and the relationship between code and written contractual terms. Automated execution does not itself establish that a transaction is valid or immune from judicial review.

3. Comparative AI contract governance models

A. United States: electronic-agent and commercial-contract model

The United States generally relies on established contract doctrine supplemented by electronic-transactions legislation and state-specific commercial law.

The Uniform Electronic Transactions Act (UETA), adopted in many states, recognises electronic records and signatures and contains provisions addressing contracts formed through electronic agents. The federal Electronic Signatures in Global and National Commerce Act (E-SIGN Act) also prevents a contract or signature from being denied legal effect solely because it is electronic, subject to statutory conditions and exceptions.

The Uniform Commercial Code (UCC), as enacted in the relevant state, may govern sales of goods and related commercial transactions. Its rules on offer, acceptance, mistake, warranties, and remedies may apply when AI systems negotiate or process orders.

Governance approach: Establish the authority of automated systems, define transaction limits, preserve electronic records, and allocate responsibility for errors through enforceable agreements.

B. United Kingdom: objective-assent and common-law model

English law generally assesses contract formation by examining the parties' outward words and conduct in context. A contract need not be invalid merely because a machine or automated system facilitated its formation.

Traditional principles of offer, acceptance, mistake, agency, interpretation, and remedies can be applied to AI-mediated transactions. The UK Jurisdiction Taskforce's legal statement on cryptoassets and smart contracts also concluded that existing English law can, in principle, accommodate smart legal contracts, subject to the facts and applicable legal requirements.

Governance approach: Make the intended legal agreement clear, establish the relationship between natural-language terms and code, and specify how errors, termination, and dispute resolution will be handled.

C. European Union: consumer protection and technology regulation

EU contract governance combines national contract law with harmonised rules concerning consumer protection, digital services, data protection, and AI regulation.

Where an AI system is subject to the EU AI Act, the applicable obligations depend on its intended purpose and regulatory classification. AI used in contract administration is not automatically high-risk in every case. Consumer transactions may also be subject to rules concerning unfair terms, misleading commercial practices, distance contracts, and digital content.

Governance approach: Ensure transparency, lawful processing of personal data, fair consumer treatment, appropriate human review where legally required, and effective remedies for defective performance.

D. India: general contract law and electronic-contract recognition

India's principal legal foundation includes the Indian Contract Act, 1872, and the Information Technology Act, 2000.

Section 10A of the Information Technology Act provides that a contract is not to be deemed unenforceable solely because electronic means were used in its formation. The Contract Act continues to govern essential questions such as consent, competence, lawful consideration and object, misrepresentation, mistake, and breach.

The main challenge is not simply whether electronic contracting is valid. It is how existing legal principles should be applied where an AI system independently selects terms, exceeds transaction limits, makes inaccurate representations, or executes a transaction without contemporaneous human review.

Governance approach: Link each AI system to an identifiable business or principal, document the authority delegated to it, and maintain reliable evidence of offers, acceptances, instructions, and performance.

E. Singapore: commercial certainty and automated transactions

Singapore combines established contract law with electronic-transactions legislation and a developed commercial dispute-resolution framework. Its courts have considered algorithmic transactions and the consequences of computer-generated mistakes.

Governance approach: Promote commercial certainty while preserving established remedies for mistake, fraud, and other legally recognised grounds for relief. Technical automation should not be confused with a complete answer to the question of legal intention.

4. Case laws: judicial principles applicable to AI contracts

The following decisions are relevant to AI contract governance because they address automated transactions, electronic contracting, objective assent, mistake, and contractual responsibility. They are not all direct AI cases. In particular, algorithmic-trading cases concern automated systems but should not be described as rulings on generative AI.

Case 1: Thornton v. Shoe Lane Parking Ltd. [1971] 2 QB 163

Jurisdiction: England and Wales.

Facts: A motorist entered an automatic car park where the ticket machine issued a ticket after payment. The dispute concerned contractual formation and the incorporation of an exclusion clause.

Decision and principle: The Court of Appeal treated the transaction as one formed through an automated process. The terms had to be assessed in the context of when and how the contract was made, and unusually onerous terms required adequate notice.

Relevance to AI: A business can form a contract through an automated system. Terms that significantly restrict liability should be presented with appropriate prominence and timing, rather than being hidden in a later screen or obscure set of conditions.

Case 2: Butler Machine Tool Co Ltd v. Ex-Cell-O Corp (England) Ltd. [1979] 1 WLR 401

Jurisdiction: England and Wales.

Facts: The parties exchanged commercial documents containing conflicting standard terms, including different approaches to price variation. The issue was which terms governed the resulting agreement.

Decision and principle: The Court of Appeal examined the exchange of offer, counter-offer, and acceptance documents to determine the terms of the contract. The judgments illustrate the importance of analysing the parties' communications as a whole.

Relevance to AI: AI negotiation systems may exchange competing standard terms at high speed. Organisations should preserve the full negotiation record and establish which version of the terms was ultimately accepted.

Case 3: ProCD, Inc. v. Zeidenberg, 86 F.3d 1447 (7th Cir. 1996)

Jurisdiction: United States Court of Appeals, Seventh Circuit.

Facts: A purchaser acquired software subject to licence terms presented with the product and later used it in a way the seller argued violated those terms.

Decision and principle: The court upheld the enforceability of the licence arrangement in the circumstances, recognising that contractual terms could be accepted through conduct after an opportunity to review them.

Relevance to AI: Terms governing AI software, model access, data use, and usage restrictions may be enforceable when properly presented and accepted. Enforceability is fact-specific and may be affected by applicable consumer-protection law and the adequacy of notice.

Case 4: Specht v. Netscape Communications Corp., 306 F.3d 17 (2d Cir. 2002)

Jurisdiction: United States Court of Appeals, Second Circuit.

Facts: Users downloaded software from a webpage, but the arbitration terms were located below the download button and were not adequately brought to users' attention.

Decision and principle: The court found that the users had not assented to the arbitration terms because the interface did not provide adequate notice that downloading the software would constitute acceptance of those terms.

Relevance to AI: An AI platform cannot safely assume that users accepted terms merely because they interacted with a chatbot, uploaded a document, or used an automated service. The interface should clearly communicate material contractual terms and the action that signifies acceptance.

Case 5: Meyer v. Uber Technologies, Inc., 868 F.3d 66 (2d Cir. 2017)

Jurisdiction: United States Court of Appeals, Second Circuit.

Facts: Users challenged whether they had agreed to Uber's terms, including an arbitration provision, during the app-registration process.

Decision and principle: The court held that the interface gave reasonably conspicuous notice of the terms and that the users' registration conduct manifested assent in the circumstances.

Relevance to AI: AI-enabled platforms should design interfaces so that material terms are accessible and the user's action clearly signifies agreement. This case complements Specht: digital contracts may be enforceable, but the adequacy of notice and manifestation of assent matters.

Case 6: Chwee Kin Keong v. Digilandmall.com Pte Ltd. [2005] 1 SLR(R) 502

Jurisdiction: Singapore.

Facts: An online seller listed printers at a price that was substantially lower than the apparent intended price. Buyers placed orders, and the seller argued that the pricing error prevented enforcement.

Decision and principle: The Singapore High Court considered the law of unilateral mistake in the context of an online transaction. It addressed the significance of whether a buyer knew, or ought to have known, that the displayed price resulted from a mistake.

Relevance to AI: An AI procurement agent may accept an erroneous quote or generate an anomalous price. The legal outcome may depend on the governing mistake doctrine, the parties' knowledge, and the transaction context. Automated acceptance does not invariably eliminate the possibility of relief for mistake.

Case 7: Quoine Pte Ltd v. B2C2 Ltd. [2020] SGCA(I) 02

Jurisdiction: Singapore International Commercial Court of Appeal.

Facts: Cryptocurrency trades were executed by algorithmic trading systems during unusual market conditions. One party sought to reverse transactions, raising questions about mistake and the role of automated systems.

Decision and principle: The Court of Appeal considered how established contract doctrines, including unilateral mistake, apply to algorithmic trading and the knowledge attributable to parties operating automated systems. It did not establish that all algorithmically executed transactions are automatically reversible or invalid.

Relevance to AI: This is particularly important for autonomous contracting. Courts may need to determine whose knowledge is legally relevant, what the system was designed to do, and whether the applicable requirements for mistake are satisfied. Algorithmic operation alone does not settle those questions.

Case 8: Register.com, Inc. v. Verio, Inc., 356 F.3d 393 (2d Cir. 2004)

Jurisdiction: United States Court of Appeals, Second Circuit.

Facts: Verio repeatedly queried Register.com's database and obtained information subject to stated terms. The dispute involved the enforceability of those terms in the context of repeated automated access.

Decision and principle: The court upheld enforcement of the terms in the circumstances, taking account of the notice provided and the user's continued conduct.

Relevance to AI: Automated agents that repeatedly access APIs, databases, or commercial platforms may be subject to terms governing that access. Businesses should not assume that automation removes contractual obligations or that repeated access is always permitted.

Case 9: Trimex International FZE Ltd. Dubai v. Vedanta Aluminium Ltd., (2010) 3 SCC 1

Jurisdiction: Supreme Court of India.

Facts: The parties negotiated a commercial transaction through email communications. A dispute arose over whether the communications had resulted in a concluded contract despite the absence of a subsequently signed formal document.

Decision and principle: The Supreme Court recognised that a binding commercial agreement may arise from electronic communications where the correspondence establishes agreement on the essential terms and the parties intended to be bound. The absence of a later formal contract is not invariably decisive.

Relevance to AI: AI-assisted negotiations can produce enforceable agreements through electronic exchanges. Businesses should establish when negotiations become binding, which terms are essential, and whether final human approval is required before an AI system communicates acceptance.

Case 10: Bhagwandas Goverdhandas Kedia v. Girdharilal Parshottamdas & Co., AIR 1966 SC 543

Jurisdiction: Supreme Court of India.

Facts: The dispute involved a contract negotiated through telephone communications and the place where the contract was formed.

Decision and principle: The Supreme Court examined when and where acceptance communicated through instantaneous communication becomes effective, distinguishing such communications from acceptance by post.

Relevance to AI: Automated negotiations may involve real-time exchanges between parties located in different jurisdictions. The principles governing communication of acceptance can help determine formation and jurisdiction, although the application to particular AI systems depends on the facts and applicable law.

Overall significance of the case law

Taken together, these decisions support five central propositions:

Automated systems can facilitate legally binding contracts.

The parties must still satisfy the applicable requirements for formation and consent.

Material terms need adequate notice and acceptance.

Mistake, misleading conduct, and lack of authority may affect enforceability or remedies.

A party cannot necessarily escape contractual responsibility merely because a computer system performed the relevant action.

None of these authorities establishes a general rule that an AI system is itself a legal person with independent contractual capacity.

5. Civil liability arising from AI contract governance

A. Liability of AI developers and vendors

An AI provider may face a claim where it breaches a contractual promise, supplies a defective system, makes actionable misrepresentations, or negligently performs an agreed service. Liability depends on the contract, the governing law, the applicable product-liability regime, and proof of causation.

B. Liability of the business deploying AI

A business may be responsible for transactions made by an AI system within its authorised scope. It may also face claims arising from inadequate controls, misleading automated communications, or failure to supervise contractual processes.

For example, if an authorised purchasing system accepts a supplier's offer within an established spending limit, the business may be bound even though no employee personally approved that individual order. If the system exceeds its authority, the outcome may depend on the scope of actual authority, representations to the counterparty, and other applicable agency principles.

C. Liability for unauthorised or erroneous transactions

Where an AI system accepts an offer by mistake, several questions arise:

Was the error obvious to the counterparty?

Did the counterparty know of the mistake?

Did the system have authority to accept the offer?

Did the business establish safeguards appropriate to the transaction?

Does the contract provide an effective correction or cancellation procedure?

The answers may determine whether the contract is enforceable, whether it can be rescinded, or whether damages are available.

D. Consumer protection

AI-powered consumer services may generate misleading product descriptions, omit material terms, or recommend unsuitable contractual options. Consumer-protection legislation may restrict unfair terms and misleading practices, even where a user technically clicked an acceptance button.

E. Smart-contract errors

Where a smart contract executes a payment or transfers a digital asset in accordance with code, the parties may still dispute whether the transaction was authorised or whether the underlying agreement was valid. Technical irreversibility does not necessarily extinguish legal remedies, although recovering assets or reversing a transaction may be difficult in practice.

6. Recommended AI contract governance framework

A sound governance model should address the full contractual lifecycle.

1. Define authority

Specify the transactions the AI may negotiate, approve, or execute; set monetary limits; and identify actions requiring human approval.

2. Establish contract-formation rules

State when an offer becomes binding, what constitutes acceptance, which terms prevail, and whether an agreement requires a signature or additional approval.

3. Preserve an audit trail

Retain prompts and relevant inputs, generated terms, version histories, approval records, timestamps, transaction logs, and communications, subject to applicable privacy and retention laws.

4. Control high-risk actions

Require additional verification for large payments, unusual pricing, termination, waiver of legal rights, changes to liability clauses, and irreversible transactions.

5. Allocate contractual risks

Define warranties, liability limits, indemnities, security obligations, correction procedures, and responsibility for software updates. Ensure that clauses comply with mandatory law.

6. Provide effective dispute resolution

Specify governing law, jurisdiction or arbitration, procedures for disputed automated transactions, and the handling of evidence held by AI vendors or third-party platforms.

7. Comparative evaluation

The table describes general tendencies, not uniform rules across every state or transaction. The applicable law depends on the particular jurisdiction, contract type, and circumstances.

8. Major legal challenges

1. Uncertain attribution: It can be difficult to determine whether a particular output is attributable to the user, business, developer, or intermediary.

2. Lack of transparency: Parties may not know how an AI system arrived at a contractual recommendation or why it accepted particular terms.

3. Conflicting versions: Automated systems may produce or accept different versions of an agreement, creating uncertainty over which terms govern.

4. Cross-border jurisdiction: AI systems and counterparties may operate in different countries, making governing law and enforcement especially important.

5. Irreversible execution: Automated payments or blockchain transactions may be technically difficult to reverse even when legal grounds for recovery exist.

6. Unequal bargaining power: Consumers and small businesses may be unable to negotiate the terms of AI-enabled platforms, raising questions about fairness and transparency.

9. Conclusion

Comparative AI contract governance demonstrates that established contract law remains central to automated commercial relationships. The key questions are not simply whether AI was used, but whether a valid agreement was formed, whether the system had authority to act, whether the terms were properly communicated, and whether a recognised ground for relief exists.

The decisions in Thornton, Butler Machine Tool, ProCD, Specht, Meyer, Chwee Kin Keong, Quoine, Register.com, Trimex International, and Bhagwandas Goverdhandas Kedia offer important principles for analysing automated and electronic contracting. They should be treated as authorities by analogy where they did not directly concern modern AI.

A robust governance framework should combine clear delegation of authority, transparent contractual terms, human review for high-impact transactions, reliable records, proportionate technical safeguards, fair risk allocation, and effective remedies. The objective is to enable the efficiency of AI-mediated commerce without undermining consent, accountability, and contractual certainty.

This explanation is intended for academic and general legal research. The legal effect of an AI-generated or AI-executed contract depends on the governing law and the facts of the particular transaction.