Civil Law and Comparative AI Contract Governance

1. Introduction

AI contract governance refers to the legal principles, contractual rules, regulatory requirements, and dispute-resolution mechanisms governing the creation, negotiation, execution, performance, interpretation, and enforcement of contracts involving artificial intelligence (AI).

AI is transforming contractual relationships in several ways. Businesses use AI to draft commercial agreements, negotiate prices, evaluate counterparties, generate purchase orders, assess contractual risks, and monitor performance. Automated systems can also conclude transactions without direct human intervention, while generative AI tools may produce contractual language that contains errors, omissions, or unintended commitments.

These developments raise fundamental civil-law questions:

Can an agreement formed by an autonomous system constitute a legally binding contract?

Who is responsible when an AI agent makes an unauthorised commitment?

Can a party avoid a contract because an AI system generated a mistake?

How should courts interpret AI-generated contractual terms?

Who bears liability when an AI system fails to perform contractual obligations?

How should confidentiality, data protection, intellectual property, and cybersecurity be addressed in AI contracts?

Comparative AI contract governance examines how different jurisdictions answer these questions through contract law, agency principles, electronic-commerce legislation, consumer protection, data-protection rules, and civil liability.

An important distinction is that a contract involving AI is not necessarily the same as a contract concluded autonomously by AI. An AI software licence may be negotiated and signed by humans, while an autonomous trading agent may execute transactions automatically. The legal issues overlap, but the rules governing authority, consent, mistake, and attribution can differ significantly.

2. Meaning and scope of AI contract governance

AI contract governance has two principal dimensions.

A. Contracts for AI

These are agreements under which AI technology is developed, supplied, licensed, integrated, maintained, or used.

Examples include:

AI software licensing agreements.

Cloud computing and AI-as-a-service contracts.

AI model development and customisation agreements.

Data licensing and training-data agreements.

AI system maintenance and technical-support contracts.

Agreements for automated decision-making services.

AI procurement contracts between governments and private suppliers.

B. Contracts formed or performed by AI

These involve AI systems that perform contractual functions ordinarily undertaken by people.

Examples include:

Automated purchasing and supply-chain transactions.

Algorithmic trading and electronic market transactions.

AI-assisted price negotiation.

Automatically generated purchase orders.

Smart contracts that execute code when specified conditions occur.

AI agents authorised to book travel, purchase services, or negotiate commercial terms.

In these situations, courts must distinguish the technical operation of a system from the legal requirements of offer, acceptance, intention to create legal relations, consideration where required, capacity, authority, and certainty of terms.

3. Core principles governing AI contracts

3.1 Contract formation and consent

A contract ordinarily requires the elements recognised by the governing legal system, such as offer and acceptance, intention to create legal relations, and sufficiently certain terms. The precise requirements differ between common-law and civil-law jurisdictions.

AI involvement does not automatically invalidate an agreement. Where a business deliberately deploys an automated system to place orders or accept offers, the resulting transaction may be legally binding under applicable electronic-commerce and contract rules.

The principal issue is whether the system acted within the relevant authority and whether the parties objectively manifested legally sufficient agreement.

3.2 Attribution of AI-generated actions

AI systems generally do not possess independent legal personality merely because they can make decisions or generate communications.

Consequently, a contract formed through AI will usually be analysed by identifying the person or organisation that deployed, authorised, or operated the system, together with the applicable rules governing agency, attribution, and electronic transactions.

Relevant considerations include:

Who owned or controlled the system?

What authority was given to the system?

Were transaction limits imposed?

Could the counterparty reasonably rely on the system's apparent authority?

Did the operator fail to implement appropriate controls?

Did the governing legislation expressly address automated transactions?

The answers determine whether a transaction binds the principal, whether a party may seek avoidance or correction, and whether damages may be available.

3.3 Mistake and erroneous AI outputs

An AI system may generate an incorrect quotation, accept an unintended price, misread a contractual clause, or produce a misleading summary of an agreement.

The legal consequences depend on whether the error is attributable to a party, whether the other party knew or should have known of the mistake, whether the contract contains relevant risk-allocation provisions, and whether the governing law permits avoidance or rectification.

A party generally cannot assume that every AI-generated error automatically cancels an otherwise valid agreement.

3.4 Performance, breach, and liability

An AI system may fail to deliver a required result because of inaccurate data, software defects, model drift, service interruption, or inadequate integration.

Contractual liability will depend on the obligations actually undertaken. For example, a supplier may promise specified uptime, functionality, security, or accuracy, while expressly declining to guarantee particular business outcomes.

Courts may need to distinguish between:

A failure to perform an express contractual obligation.

A failure to meet an applicable statutory standard.

Negligent design, integration, or maintenance.

An unforeseeable external event.

A limitation or exclusion of liability that is valid under the governing law.

3.5 Good faith, fairness, and transparency

Many civil-law jurisdictions expressly recognise good faith in contractual dealings or performance. Common-law jurisdictions approach these issues through a mixture of implied terms, established doctrines, statutory controls, and context-specific duties.

AI does not eliminate these requirements. An organisation may face legal risk if it conceals a material system limitation, misrepresents an AI service's capabilities, or uses automated processes to frustrate contractual rights.

3.6 Evidence and interpretability

AI-related contract disputes may require evidence concerning prompts, system logs, model versions, transaction records, approval workflows, and generated communications.

The fact that an AI model is complex or proprietary does not automatically excuse a party from proving its case. Courts apply the relevant rules on disclosure, admissibility, confidentiality, and evidential burdens.

4. Comparative legal frameworks for AI contract governance

A. India

India's framework combines traditional contract law with legislation recognising electronic transactions.

Principal legal sources

Indian Contract Act, 1872: Governs offer, acceptance, consent, competence, agency, breach, and contractual remedies.

Information Technology Act, 2000: Section 10A provides that a contract is not unenforceable solely because proposals, acceptances, or revocations are expressed electronically. Section 11 addresses attribution of electronic records.

Consumer Protection Act, 2019: May apply to misleading representations, unfair trade practices, and qualifying consumer transactions involving AI.

Digital Personal Data Protection Act, 2023: Relevant to personal-data processing to the extent its provisions are in force and applicable to the transaction.

The central question is usually not whether an AI system is itself a legal person, but whether its acts can be attributed to a person or business and whether the resulting transaction satisfies the applicable legal requirements.

For example, if a company's AI procurement tool automatically accepts a supplier's offer, the court may examine the company's deployment of the tool, its authority limits, the electronic record, and the parties' conduct.

Main challenge: Indian law recognises electronic contracts, but questions involving autonomous AI negotiation, unpredictable outputs, and the allocation of responsibility between developers and users remain fact-sensitive.

B. European Union

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

The GDPR may affect AI systems processing personal data during contract formation or performance. Consumer-protection and unfair-commercial-practices rules can be relevant when AI systems make representations to consumers or personalise offers.

The EU AI Act imposes obligations on specified AI systems and actors, depending on the system's classification and the relevant provisions. It does not replace national contract law or automatically determine whether an AI-generated agreement is valid.

Main challenge: Businesses must reconcile contractual freedom with mandatory data-protection, consumer-protection, and applicable AI compliance obligations.

C. United States

The US model relies on state contract law, the Uniform Commercial Code where applicable, electronic-transactions legislation, federal statutes, and judicial interpretation.

The Uniform Electronic Transactions Act (UETA), as adopted by relevant states, and the federal Electronic Signatures in Global and National Commerce Act (E-SIGN Act) support the legal recognition of electronic records and signatures. UETA § 14 specifically addresses contracts formed by electronic agents.

An electronic agent's involvement does not by itself invalidate a contract. However, authority, mistake, assent, fraud, unconscionability, and other ordinary doctrines may still matter.

Main challenge: Rules differ between states and between categories of transactions. Businesses must therefore determine the governing law, the scope of electronic-agent provisions, and any applicable exceptions.

D. United Kingdom

The UK applies established contract doctrines alongside legislation governing electronic communications, consumer transactions, data protection, and unfair terms.

Contracts formed through automated systems are not generally invalid merely because no human personally performs every step of acceptance. The objective approach to agreement remains important: what did the parties communicate or objectively manifest through their conduct?

For consumer transactions, statutory controls may restrict unfair terms or misleading practices. For commercial transactions, express allocation of risk, authority limits, warranties, and liability provisions can be decisive.

Main challenge: The legal system must apply traditional rules to systems whose outputs may be probabilistic, difficult to predict, or generated without contemporaneous human review.

E. Singapore

Singapore is an important comparative jurisdiction because it has developed substantial case law concerning automated trading, electronic transactions, and contracts formed through technology.

Its Electronic Transactions Act supports electronic contracting, while common-law principles govern issues such as agreement, mistake, authority, and contractual interpretation.

Singapore's courts have considered disputes involving algorithms that execute cryptocurrency trades, providing a useful framework for understanding how ordinary contract law can apply when computer systems perform transactions automatically.

Main challenge: Courts must determine how conventional doctrines apply when neither party fully anticipates the precise output of the automated system.

F. Comparative summary

5. Case laws relevant to comparative AI contract governance

The following cases establish principles concerning electronic contracting, algorithmic execution, contractual assent, automated transactions, and representations made by digital systems. Most predate modern generative AI; they are relevant by analogy and should not be described as direct rulings on today's generative AI systems.

Case 1: Chwee Kin Keong v. Digilandmall.com Pte Ltd (2004)

Citation: [2004] SGHC 71; [2005] 1 SLR(R) 502 — High Court of Singapore.

Facts: An online retailer mistakenly listed printers at a dramatically reduced price. Buyers placed large orders after seeing the erroneous listing. The dispute concerned whether binding contracts had been formed and whether the buyers could rely on the pricing error.

Decision: The court examined the principles of offer, acceptance, unilateral mistake, and the parties' knowledge of the error. It concluded that the relevant contracts were affected by unilateral mistake and could not be enforced in the circumstances established.

Relevance to AI contract governance: An AI purchasing agent may exploit or respond to an erroneous price generated by another system. The legal issue is whether the transaction should be enforced where one party knew, or the circumstances indicated, that a serious mistake had occurred.

Civil-law significance: The case demonstrates that electronic ordering does not eliminate ordinary rules of mistake. Businesses deploying automated purchasing systems should use price thresholds, confirmation requirements, and error-detection controls.

Case 2: Quoine Pte Ltd v. B2C2 Ltd (2020)

Citation: [2020] SGCA(I) 02; [2020] 2 SLR 20 — Singapore Court of Appeal.

Facts: Cryptocurrency trades were executed through algorithmic trading software. Following a software-related event, transactions occurred at unusual exchange rates. The dispute concerned whether the trades could be reversed on the basis of unilateral mistake and what knowledge should be attributed to the parties operating the algorithms.

Decision: The Court of Appeal addressed the application of traditional contract doctrines to algorithmic trading. It upheld the result that the relevant trades could not be unwound on the pleaded basis of unilateral mistake in the circumstances of the case.

Relevance to AI contract governance: This is a particularly important authority for automated transactions. It shows how courts can analyse algorithmic execution without treating software as an independent legal person.

Civil-law significance: It highlights the importance of system design, attribution of knowledge, contractual rules governing execution, and risk allocation for unexpected algorithmic outcomes. The judgment does not establish a universal rule that all algorithmic transactions are enforceable regardless of error.

Case 3: Register.com, Inc. v. Verio, Inc. (2004)

Citation: 356 F.3d 393 — United States Court of Appeals for the Second Circuit.

Facts: Verio repeatedly accessed Register.com's domain-name database using automated software. Register.com sought to enforce restrictions governing use of its database. The dispute included whether Verio had sufficient notice of the conditions and whether its continued use amounted to acceptance.

Decision: The Second Circuit upheld relief against Verio, relying on the circumstances of its repeated access, its knowledge of the terms, and the legal significance of its conduct.

Relevance to AI contract governance: AI agents may repeatedly access online platforms, APIs, databases, and commercial services. Their operators must consider whether use of those systems is subject to contractual restrictions and whether notice and assent are sufficient.

Civil-law significance: The case illustrates the importance of clear terms, notice, repeated conduct, and evidence of knowledge. It is not a general ruling that every automated website visit constitutes acceptance of contractual terms.

Case 4: Specht v. Netscape Communications Corp. (2002)

Citation: 306 F.3d 17 — United States Court of Appeals for the Second Circuit.

Facts: Users downloaded software from a website. The software download page contained a link to licence terms, but users were not required to scroll to or otherwise encounter the terms before downloading.

Decision: The court held that the users had not manifested sufficient assent to the licence terms because the webpage did not provide adequate notice that the download would signify agreement.

Relevance to AI contract governance: AI systems frequently interact with websites, APIs, software interfaces, and digital marketplaces. A business cannot safely assume that an AI agent has accepted terms that were not adequately presented or incorporated.

Civil-law significance: The case supports careful design of digital contracting interfaces. Where contractual terms are important, the operator should ensure that the relevant user or authorised system encounters them and that acceptance is recorded appropriately.

Case 5: ProCD, Inc. v. Zeidenberg (1996)

Citation: 86 F.3d 1447 — United States Court of Appeals for the Seventh Circuit.

Facts: ProCD sold a database subject to licence terms supplied inside the software packaging. The purchaser used the software despite restrictions on commercial use and argued that the terms were not part of the original transaction.

Decision: The Seventh Circuit upheld the enforceability of the licence terms in the circumstances, recognising that contractual acceptance could occur through conduct after the purchaser received an opportunity to review the terms.

Relevance to AI contract governance: AI software, model APIs, and cloud platforms often impose terms governing permitted use, output ownership, data handling, and commercial deployment. The case is relevant to when terms supplied with digital products become part of the contract.

Civil-law significance: It illustrates that the timing and mechanism of acceptance matter. The result is not universal: enforceability depends on the governing law, notice, opportunity to reject, and any applicable consumer-protection rules.

Case 6: Air Canada v. Moffatt (2024)

Citation: 2024 BCCRT 149 — Civil Resolution Tribunal of British Columbia, Canada.

Facts: A passenger relied on information supplied by Air Canada's website chatbot concerning a bereavement-fare policy. The chatbot provided information inconsistent with the carrier's applicable policy. Air Canada argued that the passenger should follow the policy available elsewhere on its website.

Decision: The tribunal found Air Canada liable for negligent misrepresentation in the circumstances and rejected the attempt to treat the chatbot's information as legally irrelevant merely because it was generated by an automated system.

Relevance to AI contract governance: This is a prominent direct example involving a customer-facing chatbot. A business may face legal responsibility when its AI tool supplies misleading information in the course of customer interactions.

Civil-law significance: Businesses should test chatbots, ensure that material terms are accurate, provide escalation to human support, and preserve relevant conversation records. The decision is a tribunal ruling from British Columbia, not a universal rule that every inaccurate chatbot statement creates liability.

Case 7: Thornton v. Shoe Lane Parking Ltd (1971)

Citation: [1971] 2 QB 163 — Court of Appeal of England and Wales.

Facts: A motorist entered an automated car park and obtained a ticket from a machine. The operator relied on conditions seeking to limit liability, including for personal injury.

Decision: The Court of Appeal held that the contract was formed when the motorist put money into the machine. A particularly onerous exclusion clause could not be incorporated after the contract had already been concluded without adequate notice.

Relevance to AI contract governance: The case demonstrates that contracts can be formed through automated processes and that terms must be properly communicated before or at the legally relevant point of formation.

Civil-law significance: It supports the principle that automation does not remove the requirement to give appropriate notice of contractual terms. AI-driven checkouts and autonomous purchasing systems should be designed to present material conditions at the correct stage.

Case 8: Entores Ltd v. Miles Far East Corporation (1955)

Citation: [1955] 2 QB 327 — Court of Appeal of England and Wales.

Facts: The dispute concerned acceptance communicated through a telex system and the place at which the contract was formed.

Decision: The court treated telex as a form of instantaneous communication and examined where the acceptance was received. The judgment developed principles concerning the effectiveness of acceptance communicated through instantaneous methods.

Relevance to AI contract governance: AI agents communicate through APIs, messaging systems, automated emails, and other electronic channels. Contract formation may depend on when a communication is received, whether it is an effective acceptance, and which jurisdiction's law applies.

Civil-law significance: The case illustrates that electronic communications must be assessed under applicable rules of communication and acceptance. Modern legislation may alter or supplement the older common-law rules.

Case 9: Butler Machine Tool Co Ltd v. Ex-Cell-O Corporation (England) Ltd (1979)

Citation: [1979] 1 WLR 401 — Court of Appeal of England and Wales.

Facts: The parties exchanged commercial documents containing conflicting standard terms. The dispute concerned which terms governed the contract, including a price-escalation clause.

Decision: The Court of Appeal examined the exchange of offer and acceptance and concluded that the contract was formed on the buyer's terms in the circumstances.

Relevance to AI contract governance: Automated procurement systems may exchange quotations, purchase orders, and standard terms in milliseconds. If the systems use different templates, they may create a dispute about which version of the terms was accepted.

Civil-law significance: The case highlights the need for version control, clear precedence clauses, and automated detection of conflicting terms. It remains relevant to electronic procurement even though it did not concern AI.

6. Civil liability for AI-related contractual disputes

AI contract disputes can involve several parties, and the applicable legal duties may differ according to each party's role.

A. Liability of the AI deployer

A business deploying AI to negotiate, purchase, or communicate with customers may be responsible for transactions attributable to it under the governing law. Liability may arise where the business exceeds or misuses its authority controls, makes misleading representations, or fails to meet contractual obligations.

B. Liability of the AI developer or supplier

A developer or supplier may be liable where it breaches an express warranty, fails to meet agreed specifications, violates confidentiality obligations, or negligently performs services where a duty of care exists.

However, a developer is not automatically liable for every commercial loss caused by an AI system. The contract, applicable statutory rules, causation, and the parties' respective responsibilities must be examined.

C. Liability of an integrator or intermediary

An integrator may be responsible for improper configuration, insecure connections, inadequate testing, or failure to implement agreed controls. An intermediary may also have specific duties under applicable platform, consumer, or sectoral legislation.

D. Liability for unauthorised commitments

Where an AI agent enters into a transaction beyond its instructions, the legal consequences may depend on actual authority, apparent authority, attribution rules, ratification, reliance, and the conduct of the parties.

A company that sets a clear purchase limit but fails to enforce it technically may face a different analysis from a company whose AI system is manipulated by an external attacker.

E. Liability for inaccurate AI-generated information

If a chatbot or negotiation agent makes an inaccurate statement, possible claims may include breach of contract, misrepresentation, negligence, or a statutory consumer-protection claim. The relevant elements of each claim must be established.

The Air Canada v. Moffatt decision demonstrates why a business should not assume that an AI-generated statement is legally irrelevant simply because it came from a chatbot.

7. Essential clauses for AI contracts

A well-drafted AI agreement should allocate risks before a dispute occurs. The following clauses are particularly important.

These provisions should be tailored to the transaction. A generic statement that an AI system is supplied “as is” may not resolve every issue, particularly where mandatory consumer law, statutory duties, fraud, or other restrictions apply.

8. Recommended governance framework

A practical AI contract-governance programme can be divided into five stages.

Stage 1: Contract and risk assessment

Identify the parties, applicable law, transaction value, data involved, and consequences of an erroneous commitment.

Stage 2: Define authority

Set transaction limits, approved counterparties, prohibited actions, and human approval requirements.

Stage 3: Document the agreement

Specify the system's intended role, performance obligations, data rights, warranties, liability allocation, and governing law.

Stage 4: Monitor performance

Preserve transaction logs, detect unusual commitments, test for errors, and maintain a process to suspend or override the system.

Stage 5: Resolve disputes

Establish procedures for investigating errors, preserving evidence, correcting performance, allocating losses, and using appropriate dispute-resolution mechanisms.

9. Major legal challenges

Uncertainty about attribution: Traditional contract law was developed primarily around human or corporate conduct. Autonomous systems complicate the identification of whose actions and knowledge should count.

Unpredictable outputs: A system may produce a commitment that was not anticipated by its operator. Courts must determine whether existing doctrines adequately address the circumstances.

Unequal bargaining power: Businesses may use AI to personalise terms, assess customers, or negotiate at a scale that individual consumers cannot match. Consumer-protection and unfair-terms rules may be especially important.

Conflicting contractual documents: AI procurement agents may accept inconsistent terms or operate using outdated templates. This increases the importance of reliable version control.

Cybersecurity and manipulation: Prompt injection, compromised credentials, malicious data, and software vulnerabilities may cause unauthorised transactions. Contracts should address security responsibilities and incident management.

Cross-border enforcement: An AI system, its developer, the contracting parties, and the data infrastructure may be located in different countries. Choice-of-law, jurisdiction, and enforcement provisions can become decisive.

Limitations of disclaimers: A contract may allocate certain risks, but its exclusions and limitations remain subject to applicable mandatory law, statutory controls, and the precise wording of the agreement.

10. Conclusion

Comparative AI contract governance demonstrates that existing contract principles remain central to automated commercial activity. The Indian Contract Act, electronic-transactions legislation, EU digital and consumer rules, US electronic-agent provisions, UK common law, and Singapore's algorithmic-contract cases each provide elements of a framework for determining when AI-assisted transactions are binding and who bears responsibility for errors.

The cases of Chwee Kin Keong, Quoine, Register.com, Specht, ProCD, Air Canada v. Moffatt, Thornton, Entores, and Butler Machine Tool collectively illustrate the importance of assent, notice, mistake, attribution, representations, and conflicting contractual terms.

The most effective approach is not to assume that AI contracts are automatically valid or invalid. Instead, courts and businesses should assess the applicable law, the parties' conduct, the authority conferred on the system, the terms communicated, and the circumstances of the alleged error.

The central principle is that automation changes how contractual decisions are made, but it does not eliminate the need for legally sufficient agreement, fair dealing, accountability, and effective civil remedies.