Civil Law and AI Agent Contract Formation Validity Disputes in Europe

1. Introduction

AI-agent contract formation refers to situations where an AI system, autonomous software agent, or AI-enabled purchasing system performs actions that may create, modify, accept, reject, or terminate a contract on behalf of a person or company.

Examples include:

an AI procurement agent ordering agricultural commodities;

an AI purchasing agent accepting a supplier's price;

an autonomous logistics system contracting for transport;

an AI sales agent negotiating prices;

an AI system renewing a software or service contract;

an AI agent accepting standard terms;

an AI agent making a purchase based on predefined parameters;

two automated systems negotiating with each other and concluding a transaction.

The central civil-law question is:

When an AI agent generates or communicates an offer or acceptance, when can that conduct legally bind the human or company behind the system?

European law does not generally treat an AI agent as a separate legal person merely because it operates autonomously. The more conventional approach is to attribute the agent's electronic acts to the person or enterprise that deployed, authorised, configured, or controlled it, subject to the applicable national contract law.

Importantly, there is presently no general CJEU judgment establishing that an AI agent itself possesses independent contractual legal personality. The existing European case law instead provides rules concerning electronic contracting, electronic signatures, consumer consent, automated ordering and attribution. Those authorities are therefore used by analogy when analysing AI-agent contracts.

2. Basic Legal Model

An AI-agent transaction can be represented as:

Principal

↓ instructions / authority

AI Agent

↓ electronic communication

Counterparty's system

↓ offer/acceptance

Contract

The dispute may arise at any of these stages.

Example

Company A tells its AI procurement agent:

“Purchase up to 1,000 tonnes of wheat if the price is below €240 per tonne.”

The AI agent encounters a supplier offering wheat at €235.

It automatically accepts.

The company subsequently says:

“We did not intend to buy the wheat.”

The legal dispute becomes:

Was the AI's acceptance legally attributable to Company A?

3. European Legal Framework

Several areas of European law are relevant.

A. National contract law

The fundamental requirements generally concern:

offer;

acceptance;

intention to contract;

authority;

capacity;

consent;

certainty;

lawful object;

consideration where relevant under the applicable system;

formal requirements where applicable.

These requirements differ among European legal systems.

B. E-Commerce Directive

Directive 2000/31/EC provides rules concerning electronic commerce and electronic contracting.

It facilitates electronic transactions but does not create a general rule that an AI system has independent legal personality.

C. eIDAS

Regulation (EU) No 910/2014, as amended by the European digital identity framework, is important for electronic signatures and trust services.

Article 25 establishes that an electronic signature cannot be denied legal effect or admissibility as evidence solely because it is electronic.

A qualified electronic signature has the legal effect equivalent to a handwritten signature under Article 25(2).

But eIDAS itself expressly leaves much of the law concerning contract conclusion and validity to EU or national law. (Infocuria)

D. Consumer Rights Directive

Directive 2011/83/EU becomes especially important where an AI agent purchases goods or services for a consumer.

The ordering process must provide sufficient information and, where payment is involved, the consumer must clearly acknowledge the payment obligation.

E. Unfair Contract Terms Directive

Directive 93/13/EEC may apply where AI systems accept standard terms on behalf of consumers.

An AI's acceptance does not remove the requirement that contractual terms comply with mandatory consumer protection rules.

4. The First Fundamental Question: Who Is the Contracting Party?

Suppose an AI agent sends:

“We accept your offer to purchase 500 tonnes of wheat.”

The recipient may ask:

“Who actually accepted?”

Possible answers include:

the company;

an employee acting through the AI;

the AI acting as an authorised electronic instrument;

nobody, if the system acted outside its authority and national law does not attribute the act to the principal.

The most important legal question is therefore attribution.

5. AI Agent Does Not Necessarily Become a Separate Legal Person

A software system generally does not become a legal person simply because it:

negotiates;

communicates;

makes decisions;

learns;

acts autonomously.

This is different from saying that an AI-generated transaction has no legal effect.

A company can deliberately use software as its contracting instrument.

The practical legal model is therefore closer to:

AI agent = sophisticated electronic instrument used by a contracting party

rather than:

AI agent = independent legal person

This distinction is particularly important for damages and enforcement.

6. Case 1 — El Majdoub v CarsOnTheWeb

C-322/14, CJEU, 21 May 2015

This is one of the most important European authorities for AI-agent contract formation by analogy.

Facts

A car was purchased online.

The seller's general terms contained a jurisdiction clause.

The purchaser accepted the terms using a click-wrap process.

Issue

Could electronic acceptance through click-wrapping create the necessary legal record for the jurisdiction agreement?

Decision

The CJEU accepted that click-wrapping can satisfy the relevant formal requirements where the terms can be accessed, saved and printed before the contract is concluded. (Infocuria)

AI-agent significance

Suppose an AI procurement agent clicks:

“I agree to the terms.”

If the system is authorised to conclude contracts, the electronic mechanism itself is not necessarily invalid merely because the act is performed electronically.

The difficult issue becomes:

Was the AI authorised to make that acceptance?

Principle

Electronic acceptance can satisfy contractual formal requirements; the AI dispute shifts toward authority, attribution and evidence.

7. Case 2 — Fuhrmann-2 v B.

C-249/21, CJEU, 7 April 2022

Facts

A consumer booked hotel rooms through an online platform.

The final button said:

“complete booking”

rather than clearly stating that clicking the button created an obligation to pay.

Decision

The CJEU held that the ordering button must clearly indicate that placing the order creates an obligation to pay. If the required formulation is not satisfied, the consumer is not bound by the contract/order under the relevant consumer-law rule. (Infocuria)

AI-agent significance

Imagine a consumer tells an AI:

“Find me a hotel.”

The AI subsequently books the room.

The system's technical ability to click the booking button does not eliminate mandatory consumer-protection requirements.

Principle

AI automation cannot circumvent mandatory rules concerning informed consumer consent.

This is especially important for:

AI shopping assistants;

travel agents;

subscription agents;

automated purchasing tools.

8. Case 3 — Content Services v Bundesarbeitskammer

C-49/11, CJEU, 5 July 2012

Facts

Consumers entered into online contracts.

The trader provided important contractual information through a hyperlink rather than delivering it in the legally required durable form.

Decision

The CJEU distinguished merely making information available on a website from providing it on a legally sufficient durable medium. (Infocuria)

AI-agent significance

An AI agent might receive contractual information from:

a website;

API;

dynamic terms;

a chatbot;

a URL;

a changing database.

A company cannot necessarily argue:

“The AI could technically access the terms.”

The legally relevant question may be whether the required information was actually supplied in the manner required by applicable law.

Principle

AI accessibility is not automatically equivalent to legally sufficient communication.

9. Case 4 — V.B. Trade

C-466/22, CJEU, 29 February 2024

Issue

The case concerned electronic signatures and the concept of a qualified electronic signature under eIDAS.

Decision

The Court examined the conditions required for a signature to qualify as a qualified electronic signature and confirmed that national courts may examine whether the statutory requirements are actually satisfied. (Infocuria)

AI-agent significance

Consider an AI system that automatically applies a digital signature to a contract.

Questions arise:

Who is the signatory?

Was the signature device controlled by the company?

Was the AI authorised to sign?

Was a qualified signature actually used?

Does national law require personal signature?

Was the relevant credential securely controlled?

An AI-generated digital mark does not automatically become a qualified electronic signature.

Principle

Electronic form alone does not establish qualified-signature status; the statutory requirements and attribution must be established.

10. Case 5 — Ekofrukt

C-362/21, CJEU, 20 October 2022

Facts

The dispute concerned the legal status of an electronic signature under eIDAS.

Decision

The CJEU held that an electronic signature cannot be declared ineffective merely because it is electronic, but national courts may examine whether it actually satisfies the requirements for a qualified electronic signature. (Infocuria)

AI-agent significance

This is important when an AI agent signs or authenticates a contract.

The argument:

“The signature was generated electronically, therefore it is invalid”

is insufficient.

But the opposite argument:

“The AI generated a digital signature, therefore it is equivalent to a handwritten signature”

is also too broad.

The relevant legal classification must be established.

Principle

Electronic signature and legally equivalent qualified signature are not necessarily the same thing.

11. Case 6 — Jarocki

C-302/23, CJEU, 17 October 2024

Facts

The case concerned electronic signatures and the lodging of electronic documents before courts.

Decision

The CJEU reiterated that electronic signatures cannot be denied legal effect merely because they are electronic, while national law retains an important role in determining formal requirements and legal effect. (Infocuria)

AI-agent significance

This reinforces an important principle for AI contracting:

EU electronic-signature law does not completely replace national contract-formation law.

Therefore, an AI-generated transaction must still be analysed under the applicable national law concerning:

authority;

consent;

form;

capacity;

contract formation.

Principle

eIDAS facilitates electronic transactions but does not create a complete autonomous AI-contract law.

12. Case 7 — Verein für Konsumenteninformation v Amazon EU

C-191/15, CJEU, 28 July 2016

Facts

Amazon used standard contractual terms in cross-border online consumer transactions.

An Austrian consumer organisation challenged the terms.

Decision

The CJEU considered the interaction between:

online contracts;

choice-of-law clauses;

unfair contractual terms;

mandatory consumer protection.

The Court emphasised that a contractual choice of law cannot simply deprive consumers of mandatory protections that would otherwise apply. (curia)

AI-agent significance

Suppose a consumer AI automatically accepts:

“All disputes shall be governed exclusively by the law of Country X.”

That does not automatically make the clause enforceable.

Principle

An AI's acceptance does not eliminate mandatory consumer protection or unfair-term controls.

13. Case 8 — Möbel Kraft

C-529/19, CJEU, 21 October 2020

Issue

The case concerned distance contracts and the information that must be provided before consumers become bound.

The CJEU emphasised the importance of giving consumers the information necessary to understand the contractual terms and consequences before the contract is concluded. (Infocuria)

AI-agent significance

An AI assistant may make purchasing extremely fast.

But speed does not eliminate:

pre-contractual information;

withdrawal rights;

price disclosure;

mandatory consumer warnings.

Principle

Automation cannot replace legally required pre-contractual information.

14. What These Cases Tell Us About AI Agents

The cases do not establish a single rule such as:

“AI contracts are valid.”

Instead, they produce a layered framework.

Layer 1 — Electronic communication

Is the electronic method legally recognised?

El Majdoub

Layer 2 — Information

Was the required information supplied?

Content Services / Möbel Kraft

Layer 3 — Consent

Did the contracting party clearly agree?

Fuhrmann-2

Layer 4 — Signature

Does the electronic signature satisfy applicable requirements?

Ekofrukt / V.B. Trade / Jarocki

Layer 5 — Contractual terms

Are the terms enforceable?

VKI v Amazon

Layer 6 — Attribution

Was the AI acting for the person or company?

This remains heavily dependent on national contract and agency law.

15. AI Agent as an Electronic Agent

A useful conceptual model is:

Principal → AI Agent → Counterparty

The AI is functioning as an electronic agent or automated contracting tool.

The critical questions are:

Who deployed the AI?

Who configured it?

What authority was given?

What limits were imposed?

Did the agent exceed those limits?

Could the counterparty reasonably rely upon the agent?

Did the principal know that the agent was operating autonomously?

Did the principal allow the system to communicate acceptance?

Was the AI's action technically foreseeable?

16. Actual Authority

Actual authority may exist where a company expressly authorises the AI to contract.

Example:

“The procurement AI may purchase up to €100,000 of agricultural products.”

The AI buys €80,000 of wheat.

There is a strong factual basis for attributing the transaction to the company, subject to applicable national law.

17. Apparent Authority

A more difficult case occurs when the AI exceeds its internal instructions.

Example:

Internal instruction:

Maximum purchase: €100,000.

AI purchases:

€500,000.

The supplier does not know the internal limit.

The supplier sees:

“Official Procurement Agent — Company X.”

The dispute may concern apparent authority, reliance and allocation of risk under the applicable national law.

This is likely to become one of the most important AI-contract disputes.

18. AI Hallucination and Contract Formation

AI systems can produce fabricated:

prices;

product descriptions;

supplier identities;

contractual terms;

delivery dates;

warranties.

Suppose an AI tells a buyer:

“The seller offers 1,000 tonnes at €200.”

But the actual seller offered:

€260.

The AI accepts €200.

Questions include:

Was €200 an offer?

Did the AI have authority?

Was the communication objectively attributable to the principal?

Was the counterparty aware of the error?

Can the contract be avoided for mistake?

Was the error caused by the AI provider?

19. AI Agent Mistake

Traditional civil-law systems recognise various doctrines concerning:

mistake;

error;

misrepresentation;

fraud;

lack of authority;

mistake in transmission.

AI introduces a new factual mechanism for producing the error.

But the underlying legal doctrine does not necessarily disappear.

The court may ask:

Who bears the legal risk of an automated system's mistake?

Possible answers depend on:

national law;

contractual allocation;

foreseeability;

system configuration;

negligence;

notice to the counterparty.

20. AI-to-AI Contract Formation

A particularly difficult scenario is:

AI Agent A

↓

negotiates

↓

AI Agent B

Both systems independently generate offers and acceptances.

Example:

AI A:

“Accept €210.”

AI B:

“Accepted.”

Neither human has reviewed the transaction.

Is there a contract?

Potentially yes, if the relevant national law recognises:

the electronic communications as legally attributable to the parties;

sufficient offer and acceptance;

authority of the systems;

sufficiently certain terms.

The absence of real-time human intervention does not necessarily prevent contract formation.

But there is no general EU rule yet stating that every AI-to-AI exchange automatically creates a contract.

21. Smart Contracts and AI Agents

AI agents can also interact with blockchain smart contracts.

Example:

AI agent:

detects wheat price;

↓

automatically sends transaction;

↓

smart contract:

transfers payment;

↓

digital asset:

transferred automatically.

A legal dispute may arise because:

code executed successfully

but

the underlying legal contract may be disputed.

This produces two separate questions:

Technical question

Did the code execute?

Legal question

Was there a valid legal obligation requiring that execution?

These are not necessarily identical.

22. Contract Formation vs Contract Performance

This distinction is important.

Formation dispute

“Did a contract ever come into existence?”

Performance dispute

“A contract existed, but the AI performed incorrectly.”

Example:

AI purchases:

100 tonnes of Grade A wheat.

The contract is valid.

The AI later orders:

Grade B wheat.

That may be a performance/authority issue rather than a formation issue.

23. AI Agent Contracting in B2B Transactions

B2B transactions are particularly important because many consumer-specific protections do not apply.

Example:

German food company

uses

French AI procurement platform

to buy

Spanish agricultural products

The AI:

negotiates;

accepts price;

agrees delivery;

accepts arbitration clause.

The court may need to determine:

applicable law;

authority;

incorporation of terms;

arbitration agreement;

electronic evidence;

validity of AI-generated acceptance.

24. AI Agent Contracting in Consumer Transactions

Consumer transactions require greater caution.

An AI shopping agent may:

select a product;

compare prices;

place the order;

agree to terms;

subscribe to services.

But the consumer-protection framework may require:

clear price information;

clear contractual terms;

confirmation;

payment acknowledgement;

withdrawal information.

Fuhrmann-2 is particularly important because the CJEU held that failure to meet the required ordering-button standard can mean that the consumer is not bound by the contract/order. (Infocuria)

25. AI Agent and Dynamic Pricing

Suppose an AI agent is authorised to:

“Buy when price falls below €200.”

The market price changes every second.

The AI accepts at €199.

The seller argues:

“That price was only displayed for a fraction of a second.”

The dispute may concern:

when the offer was made;

when acceptance became effective;

whether the platform's system created the offer;

whether an automated quote was binding;

timing of electronic communications.

Digital timestamps and server logs become critical.

26. Evidence in AI Contract Disputes

Courts may need to examine:

AI records

prompts;

instructions;

system messages;

agent memory;

tool calls;

decision logs;

model version.

Transaction records

timestamps;

APIs;

emails;

order confirmations;

authentication records.

Corporate records

authority matrix;

procurement policy;

AI governance policy;

employee instructions;

system configuration.

Technical records

audit logs;

database records;

blockchain entries;

API logs.

27. Burden of Proof

A central dispute may be:

“The AI made the contract.”

The other party responds:

“Our company never authorised this.”

Evidence may include:

access credentials;

system architecture;

authorised-user records;

AI deployment documentation;

transaction history;

company policies.

A company that gives an AI system unrestricted contractual authority may face a more difficult factual argument than a company that can demonstrate carefully defined transaction limits.

28. AI Agent and Electronic Signature

Not every AI-generated contract requires a qualified electronic signature.

This depends on:

applicable national law;

type of transaction;

statutory form;

contractual requirements;

eIDAS requirements.

However, where a qualified electronic signature is required, merely saying:

“The AI digitally signed it”

does not establish that the signature satisfies the qualified-signature requirements.

Ekofrukt and V.B. Trade demonstrate this distinction. (Infocuria)

29. AI Agent and Mandatory Formalities

Certain transactions can require special form.

Examples may include transactions involving:

immovable property;

registered rights;

guarantees;

consumer credit;

securities;

corporate acts.

An AI cannot eliminate a statutory formality merely because the parties used an automated platform.

eIDAS expressly preserves national and EU rules concerning contractual validity and form. (Infocuria)

30. Unauthorised AI Contract

Consider:

Company A authorises its AI to purchase up to €50,000.

AI purchases:

€2 million.

Possible legal outcomes include:

Possibility 1

Company is bound because the counterparty reasonably relied on the AI's apparent authority.

Possibility 2

Company is not bound beyond its actual authority.

Possibility 3

Company must compensate the counterparty for reliance losses.

Possibility 4

Company ratifies the transaction afterwards.

The answer is primarily a matter of the applicable national law of agency, contract and representation.

31. AI Provider Liability

A separate dispute may exist between:

Company → AI provider

Suppose the AI provider promised:

“The system will never make purchases above your authorised limit.”

The system purchases €10 million.

The company may consider:

breach of contract;

negligence;

indemnification;

service-level obligations;

product/service defect;

limitation clauses.

This is separate from whether the third-party contract was valid.

32. Three-Layer Litigation Model

AI contract litigation can therefore involve three separate relationships:

Layer 1 — Principal vs Counterparty

Was the AI-generated contract valid?

Layer 2 — Principal vs AI Provider

Did the AI provider breach its obligations?

Layer 3 — Principal vs AI Agent/User

Was the system used contrary to internal authority?

A court should not automatically merge these three questions.

33. AI Contract Validity Matrix

34. Important Case-Law Revision Table

35. Key Legal Principles

Principle 1

AI itself does not automatically become a legal person capable of contracting in its own name.

Principle 2

An AI can nevertheless function as an electronic contracting instrument of a human or company.

Principle 3

Electronic contract formation is not inherently invalid.

Principle 4

The crucial issue is attribution and authority.

Principle 5

Electronic signatures and AI-generated communications are different legal questions.

Principle 6

A qualified electronic signature has special legal status under eIDAS, but an ordinary electronic signature cannot be dismissed merely because it is electronic. (Infocuria)

Principle 7

Consumer protection rules cannot be bypassed by automation.

Principle 8

An AI agent cannot eliminate mandatory statutory formalities.

Principle 9

AI hallucination does not automatically determine whether a contract exists; attribution, authority, mistake and national contract law must be analysed.

Principle 10

Code execution and legal contract formation are separate questions.

36. Practical Example

Facts

A French agricultural company deploys an AI procurement agent.

Its instruction is:

“Purchase wheat below €250/tonne, maximum 500 tonnes.”

The AI negotiates with a German supplier.

It accidentally sends:

“We accept 5,000 tonnes at €280/tonne.”

The German supplier accepts.

Legal questions

Question 1: Was the AI authorised?

Only up to 500 tonnes and €250.

Question 2: Did the supplier know of the limitation?

If not, apparent-authority principles may become relevant.

Question 3: Was the AI's message an acceptance?

Depends on applicable contract law and the preceding communications.

Question 4: Can the company rely on AI error?

Potential mistake/authority doctrines may be relevant.

Question 5: Who bears the loss?

Potentially:

company;

supplier;

AI provider;

depending on the legal relationships and applicable law.

Question 6: What evidence matters?

AI instruction;

system logs;

supplier communications;

timestamps;

authentication;

transaction history.

37. Strongest Litigation Questions

When an AI-agent contract is disputed, courts are likely to need answers to these factual questions:

Who owned the AI system?

Who deployed it?

Who authorised it?

What authority was granted?

What limits were programmed?

Could the counterparty see those limits?

What exactly did the AI communicate?

Was there a valid offer?

Was there a valid acceptance?

Was the transaction subject to special formalities?

Were standard terms incorporated?

Was the AI-generated signature legally sufficient?

Was the transaction consumer or B2B?

Which national law applies?

What loss resulted from the AI's action?

38. Future Direction of European AI Contract Law

The legal development is likely to focus less on the question:

“Can AI sign a contract?”

and more on:

“Who should bear the legal risk of autonomous machine decisions?”

Three possible approaches are particularly important:

Approach A — Principal-risk model

The company using the AI bears the risk of its authorised automated actions.

Approach B — Fault model

Liability depends on:

poor configuration;

inadequate safeguards;

negligence;

failure to supervise.

Approach C — Reliance model

Where an innocent counterparty reasonably relies on an apparently authorised AI agent, the principal may bear the consequences.

The actual legal answer will depend heavily on national contract and agency law.

39. Exam-Style Conclusion

AI-agent contract formation disputes in Europe are currently governed principally through existing contract, agency, electronic-commerce, consumer-protection and electronic-signature rules rather than through a comprehensive European law granting AI systems independent contractual personality.

The CJEU's decisions in El Majdoub, Fuhrmann-2, Content Services, Ekofrukt, V.B. Trade, Jarocki, VKI v Amazon and Möbel Kraft demonstrate that European law already recognises extensive forms of electronic contracting while maintaining requirements concerning consent, information, durable records, electronic signatures, consumer protection and national contractual formalities. (Infocuria)

For AI agents, the central legal formula is:

AI Action + Authority/Attribution + Offer + Acceptance + Required Form + Genuine Consent + Applicable Mandatory Rules = Contract Formation Analysis

And where the AI exceeds its authority:

Actual Authority + Apparent Authority + Counterparty Reliance + AI Error + Applicable National Law = Liability/Validity Dispute

The most important unresolved area is therefore attribution: determining when an autonomous AI decision should legally be treated as the decision of the human or company that deployed it. Current EU electronic-contract case law provides the foundations for this analysis, but does not yet establish a comprehensive autonomous-AI contract-formation regime.