Civil Law and Autonomous Negotiation Systems Liability in Europe

1. Introduction

Autonomous negotiation systems are AI or software systems capable of negotiating contractual terms with little or no real-time human intervention.

Examples include:

AI agents negotiating purchase prices;

procurement systems automatically negotiating with suppliers;

autonomous sales agents negotiating discounts;

algorithmic systems negotiating delivery dates;

AI systems negotiating financing terms;

machine-to-machine commercial contracting;

automated insurance or service negotiations;

autonomous purchasing systems;

AI agents negotiating licensing or technology agreements.

The central civil-law question is:

When an autonomous system makes, changes, accepts, rejects, or communicates contractual terms, who is legally bound and who bears responsibility if the system makes an error?

As of 2026, Europe does not have a fully developed, standalone doctrine called “autonomous negotiation system liability.” Existing rules must be combined from contract law, agency/representation, electronic contracting, product liability, tort/delict, consumer law, data protection, and increasingly AI regulation.

The new EU Product Liability Directive expressly treats software, including AI systems, as products for its purposes and can cover defects arising from software updates or upgrades under the manufacturer's control. (Eur-Lex)

2. Basic Legal Structure

An autonomous negotiation dispute normally contains three separate questions:

Question 1 — Was there a contract?

For example:

AI buyer offers €900,000 → AI seller accepts → system automatically generates purchase order.

The court must determine whether the electronic communications constituted legally effective offers and acceptance.

Question 2 — Who made the legal declaration?

Was it:

the company;

its employee;

its AI agent;

its software provider;

its system integrator?

The AI normally does not become a separate legal person merely because it negotiated autonomously.

Question 3 — Who bears the loss?

Possible responsible parties include:

principal/company;

software developer;

AI provider;

system integrator;

cloud provider;

employee;

cybersecurity provider;

other contracting party.

3. Autonomous Negotiation Does Not Automatically Mean Autonomous Legal Personality

A useful principle is:

The autonomy of the technology does not necessarily determine the legal identity of the contracting party.

If Company A deliberately deploys an AI procurement agent to negotiate contracts, the ordinary starting point is that the relevant legal consequences are attributed to Company A according to applicable contract and agency rules.

The difficult question arises when the system:

exceeds its instructions;

accepts an unintended price;

negotiates outside its authorised range;

misunderstands a counteroffer;

suffers a software error;

becomes compromised;

negotiates with another AI;

enters a contract that the company never intended to make.

4. Case Law

Case 1 — BGH, VIII ZR 79/04, 26 January 2005

German Federal Court of Justice

This is one of the most useful European cases for autonomous electronic contracting.

A computer system automatically transferred pricing information to an Internet sales database. Because of a technical error, a notebook that should have been listed at approximately €2,650 appeared at €245. An automatically generated communication followed the customer's order. (JurPC)

The German Federal Court of Justice considered the legal consequences of the erroneous electronic pricing process under the German rules on mistake.

Principle

A technical error in an automated electronic process does not necessarily disappear from contract law merely because a computer generated the relevant communication.

The legal analysis focuses on:

whether a legally relevant declaration occurred;

whether there was an error;

whether the error could legally be challenged;

what the applicable contract rules provide.

Application to autonomous negotiation

Imagine an AI procurement agent is instructed:

“Do not pay more than €100,000.”

Because of a software error, it negotiates and accepts:

€180,000.

The company cannot simply argue:

“The computer did it, so no legal consequences exist.”

Instead, the court would examine the applicable rules concerning:

authority;

mistake;

contractual formation;

reliance;

attribution;

damages.

Importance

This case is highly relevant to AI-generated offers, automated pricing and machine-generated contractual communications.

5. Case 2 — BGH, VIII ZR 289/09, 11 May 2011

German Federal Court of Justice

This case concerned contractual declarations made using another person's eBay account.

The BGH held that using another person's account to make contractual declarations can raise principles analogous to agency, apparent authority and tolerated authority. Mere failure to protect login credentials was not automatically enough to attribute the declaration to the account holder. (nu:legal Deutsches Recht)

Principle

Attribution of an electronic declaration requires more than simply asking:

“Which account was used?”

The court examines:

authority;

apparent authority;

conduct of the account holder;

circumstances surrounding the transaction.

Application to AI agents

Consider:

Company A → AI negotiation agent → supplier

If the AI exceeds its instructions, the court may need to determine:

Did Company A authorise the AI?

What powers were communicated?

Did Company A create an appearance of authority?

Did the counterparty reasonably rely on that authority?

Was the AI compromised?

Importance

The case provides a useful analogy for attribution of autonomous electronic contractual declarations.

6. Case 3 — El Majdoub v CarsOnTheWeb.Deutschland GmbH

CJEU, Case C-322/14

The CJEU considered the validity of contractual jurisdiction arrangements concluded electronically through a click-wrap mechanism.

The Court accepted that electronic contracting mechanisms can satisfy formal requirements where the relevant conditions are fulfilled and the terms can be stored/reproduced. (Infocuria)

Application to autonomous negotiation

An autonomous negotiation system may communicate:

“I accept the supplier's terms.”

The legal question is not merely whether a human physically clicked a button.

Instead, the relevant legal framework may ask:

Was electronic acceptance permitted?

Were the terms properly incorporated?

Was the system authorised?

Was there a durable record?

Were mandatory formalities satisfied?

Importance

This case supports the broader proposition that electronic means of contracting can produce legally significant contractual consequences.

It is therefore relevant to machine-to-machine negotiations.

7. Case 4 — Content Services Ltd v Bundesarbeitskammer

CJEU, Case C-49/11

The CJEU examined the information requirements applicable to distance contracts and whether information supplied through a website hyperlink satisfied the relevant consumer-law requirements.

The Court emphasised the importance of information being supplied in a form satisfying the applicable legal requirements, including durable-medium requirements under the then applicable framework. (curia)

Application

An autonomous AI negotiator might conclude a contract with a consumer.

The company cannot necessarily rely on:

“The AI displayed the terms somewhere on the platform.”

Consumer-law requirements concerning:

information;

transparency;

withdrawal rights;

contractual terms;

durable records

may continue to apply.

Importance

Autonomous negotiation does not remove mandatory consumer-protection requirements.

8. Case 5 — SCHUFA Holding, C-634/21

CJEU, 7 December 2023

This case concerned automated credit scoring and Article 22 GDPR.

The Court treated certain automated scoring activity as falling within the legal framework governing automated individual decision-making where the score was used by a third party in a way that effectively determined the outcome. (Infocuria)

Relevance to autonomous negotiation

Although SCHUFA concerned an individual rather than commercial contract negotiations, its reasoning is important for AI systems that independently determine:

whether to negotiate;

whether to accept a counteroffer;

whether to offer credit;

whether to terminate negotiations;

what price to propose.

Important qualification

GDPR Article 22 primarily protects natural persons. Therefore, it should not be incorrectly treated as a general corporate-contract liability rule.

Its value here is principally analogical, concerning automated decision-making and human involvement.

9. Case 6 — Dun & Bradstreet Austria, C-203/22

CJEU, 27 February 2025

The CJEU addressed automated credit assessment and the right to receive meaningful information about the logic involved in automated decision-making.

The Court stated that the information provided must allow the person concerned to understand and challenge the automated decision. (curia)

Application to autonomous negotiation

Suppose an AI commercial agent automatically rejects every supplier whose predicted price exceeds a particular threshold.

A dispute could concern:

how the threshold was calculated;

which data the system used;

whether the system operated correctly;

whether the decision was discriminatory;

whether the system followed its instructions.

Importance

The case demonstrates the increasing legal importance of:

algorithmic explainability + evidence + transparency.

Again, it is an analogical authority, not a case concerning an autonomous commercial negotiation agent.

10. Case 7 — Boston Scientific Medizintechnik

CJEU, Joined Cases C-503/13 and C-504/13

The CJEU examined defective-product liability where products belonging to the same production series presented an increased safety risk.

Application

An autonomous negotiation system could potentially be treated as a software product under the modern EU product-liability framework where the statutory conditions are satisfied.

For example:

AI negotiation software repeatedly produces materially erroneous contractual outputs because of a systematic software defect.

The new Product Liability Directive specifically recognises software and AI systems within its product concept. (Eur-Lex)

Importance

The case supports analysis of systemic software defects, although Boston Scientific itself was not an AI-negotiation case.

11. Case 8 — Sanofi Pasteur, C-621/15

CJEU

Sanofi Pasteur concerned product liability and causation, including difficult evidentiary questions where scientific certainty may be contested.

Application

Suppose an autonomous negotiation system:

contains a software defect;

generates an unintended contractual commitment;

causes the company to purchase goods at an excessive price;

the company subsequently suffers a loss.

The claimant must still establish:

defect → autonomous action → contractual consequence → economic loss.

A technical defect alone does not automatically establish the entire damages claim.

12. Contract Formation

Autonomous negotiations can generate several types of contractual communication:

Stage 1 — Initial proposal

“We offer 10,000 units at €20 each.”

Stage 2 — Counteroffer

“We can offer €18.50.”

Stage 3 — Automated response

“Accepted.”

Stage 4 — Automatic confirmation

“Contract concluded.”

A court must determine whether these communications constitute:

invitation to treat;

offer;

counteroffer;

acceptance;

preliminary agreement;

framework agreement.

The applicable national contract law remains crucial.

13. AI Exceeds Its Negotiation Mandate

This is probably the most important liability problem.

Suppose:

Human instruction:
Maximum price = €1 million.

AI action:
Agrees to €1.8 million.

Possible legal arguments include:

Company argument

“The AI exceeded its authority.”

Counterparty argument

“Your company deliberately deployed the AI and we reasonably relied on its apparent authority.”

The court may therefore have to distinguish:

internal limitation of authority

from

externally communicated authority.

14. Hidden Instructions

Consider:

Company tells AI: “Never disclose our reservation price.”

But the AI tells the other negotiator:

“Our maximum acceptable price is €5 million.”

This may generate claims involving:

breach of confidentiality;

breach of contract;

negligence;

trade-secret protection;

cybersecurity;

fiduciary obligations;

damages.

The question becomes whether the AI's disclosure can legally be attributed to the company and whether the company/provider failed to implement reasonable safeguards.

15. AI Hallucination During Negotiation

An autonomous system might falsely state:

“Our company has already approved this price.”

Or:

“The buyer has agreed to delivery within 24 hours.”

Or:

“Our company guarantees the product for ten years.”

Potential consequences include:

contractual dispute;

misrepresentation;

pre-contractual liability;

negligent misstatement;

breach of warranty;

reliance damages.

The legal consequences depend upon national law and the circumstances in which the statement was made.

16. Software Defect vs Negotiation Mistake

This distinction is essential.

Situation A — Business mistake

Company deliberately instructs AI to negotiate aggressively, but later dislikes the result.

This is primarily a contract/authority problem.

Situation B — Software defect

The AI was programmed to reject offers above €1 million but accepted €10 million because of a coding error.

This may additionally create:

software-provider liability;

product liability;

contractual indemnity;

professional negligence.

Situation C — Cyberattack

Attacker modifies the AI and causes it to negotiate disadvantageous contracts.

This creates a different causation and cybersecurity analysis.

17. Liability of the AI Developer

Potential allegations include:

negligent software design;

inadequate testing;

failure to implement safeguards;

inadequate documentation;

defective update;

failure to correct known errors;

inadequate cybersecurity;

foreseeable misuse not considered.

The modern Product Liability Directive is particularly significant because it treats software, including AI systems, as products for its purposes. It also addresses software updates/upgrades and certain defects arising after placing a product on the market where the software remains under the manufacturer's control. (Eur-Lex)

18. Liability of the AI User

The company deploying the AI can potentially be responsible for:

poor configuration;

excessive delegation;

failure to supervise;

failure to test;

ignoring known errors;

inadequate access controls;

failure to maintain logs;

failure to establish negotiation limits.

The user may therefore have responsibility even when the software itself is not defective.

19. Liability of the System Integrator

The integrator is particularly important where the system consists of:

AI model + CRM + procurement software + ERP + payment system + communication platform.

Each individual component may work correctly, but their interaction may produce an erroneous negotiation.

Example:

ERP shows €1 million authority → AI interprets figure as €10 million → negotiation agent concludes contract → payment system automatically processes transaction.

The integrator may face contractual or negligence claims depending upon the circumstances.

20. Joint Liability

Modern EU product-liability legislation expressly addresses situations in which multiple economic operators are liable for the same damage and provides for joint and several liability within the Directive's scope. (Eur-Lex)

This is particularly relevant to autonomous negotiation technology involving:

AI developer + software integrator + hardware manufacturer + cloud provider.

However, the precise availability of joint liability outside the Directive remains dependent on applicable national law.

21. Economic Loss

Autonomous negotiation disputes are unusual because the principal harm may be pure economic loss.

Examples:

buying goods above market value;

selling below authorised price;

agreeing excessive delivery charges;

accepting an unfavourable financing rate;

losing a valuable contract;

revealing a reservation price;

granting excessive discounts.

This is important because the EU product-liability regime and national contractual/tort rules do not necessarily treat all categories of pure economic loss identically.

Therefore:

A defective AI system does not automatically guarantee recovery of every economic consequence caused by its negotiation.

22. Damages

Potential damages may include:

Direct loss

Difference between:

contract price actually agreed

and

economically appropriate price, where legally recoverable and provable.

Consequential loss

Potential losses from:

resale;

missed production;

customer claims;

additional financing;

replacement suppliers.

Reliance expenditure

Costs incurred because the company relied on the AI-negotiated agreement.

Lost profits

Potentially recoverable under applicable law, but usually requiring strong proof of causation and foreseeability.

23. Mitigation of Loss

The injured company normally cannot simply allow losses to increase indefinitely.

Suppose an AI agrees to buy 100,000 units at an excessive price.

The company discovers the error immediately.

It may have to consider:

renegotiation;

cancellation;

resale;

substitute suppliers;

alternative buyers.

Failure to mitigate may reduce recoverable damages under applicable national law.

24. Autonomous Negotiation and Consumer Contracts

Consumer transactions create additional complications.

An AI sales agent cannot necessarily:

hide mandatory information;

eliminate statutory withdrawal rights;

use unfair terms;

create misleading representations;

bypass transparency obligations.

The CJEU's electronic-contract cases demonstrate that the digital method used to conclude the contract does not eliminate mandatory consumer-law requirements. Content Services and El Majdoub are useful authorities in this respect. (curia)

25. AI-to-AI Negotiation

A particularly futuristic scenario is:

Buyer AI ↔ Seller AI

Neither human directly participates.

Example:

Buyer AI: “€900 per unit.”

Seller AI: “€950.”

Buyer AI: “€925 accepted.”

Seller AI: “Confirmed.”

The legal system would still need to identify:

the principals;

authority of each system;

contractual terms;

applicable law;

jurisdiction;

whether acceptance occurred;

whether the systems acted within their mandates.

The fact that both parties used AI does not eliminate ordinary contract-law questions.

26. Autonomous Renegotiation

AI systems may not merely create contracts; they may automatically modify existing agreements.

Example:

“If raw-material prices rise by 10%, renegotiate the supply contract.”

The AI automatically changes:

price;

quantity;

delivery date;

payment terms.

Potential disputes include:

Was modification authorised?

Was the contractual modification mechanism satisfied?

Was consideration required under applicable national law?

Was the change within the AI's mandate?

Was the counterparty adequately notified?

Was the modification recorded?

27. Evidence and Audit Trails

Autonomous negotiation litigation will increasingly depend upon technical evidence.

Important evidence includes:

AI prompts;

system instructions;

model version;

negotiation logs;

API calls;

timestamps;

automated emails;

server logs;

identity authentication;

software updates;

configuration settings;

human approvals;

system alerts.

The absence of reliable logs can make causation and attribution substantially more difficult.

28. Black-Box Negotiation

A major problem occurs where a company cannot explain:

Why did the AI make that offer?

For example, an AI negotiator offers €7 million when its apparent instruction was to remain below €5 million.

Possible causes:

faulty training;

incorrect context;

prompt manipulation;

system integration;

software bug;

malicious input;

model drift.

Dun & Bradstreet demonstrates, in the GDPR context, the increasing legal significance of meaningful information about automated decision logic. (curia)

But this should not be overstated: Dun & Bradstreet does not establish a general civil-law right to demand the source code of every commercial AI negotiator.

29. Product Liability Under the Modern EU Framework

The Product Liability Directive (EU) 2024/2853 is particularly important for autonomous negotiation systems.

It expressly recognises:

software;

AI systems;

software updates;

upgrades;

AI-related defects.

It also treats an AI/software provider as a manufacturer for the Directive's purposes in relevant circumstances. (Eur-Lex)

This potentially creates an important route where the autonomous negotiation software itself is defective.

But an important limitation

The Directive is fundamentally a product-liability regime, not a universal rule making every bad commercial decision by AI compensable.

The claimant must still establish the statutory elements, including:

defect + legally relevant damage + causal link.

30. Six Core Legal Questions

A court considering autonomous negotiation liability could therefore ask:

1. Authority

Did the AI have authority to negotiate?

2. Attribution

Can its communication legally be attributed to the company?

3. Contract formation

Did offer and acceptance create a contract?

4. Defect

Was the AI/software defective?

5. Causation

Did the defect cause the contractual/economic loss?

6. Damages

What loss is legally recoverable?

31. Hypothetical Example

Company A gives its autonomous purchasing AI this instruction:

“Purchase 10,000 components, maximum €100 each.”

The AI negotiates with Company B.

Due to a software error, it agrees:

€160 per component.

Company A discovers the error after the contract is concluded.

Potential legal analysis

Contract formation:
Was the AI's acceptance legally effective?

Authority:
Was the AI authorised to conclude contracts?

Internal limitation:
Was the €100 ceiling communicated externally?

Mistake:
Does national contract law permit avoidance or another remedy?

Software defect:
Was the AI incorrectly designed?

Developer liability:
Did the software provider breach its contractual obligations?

Damages:
What loss resulted?

Mitigation:
Could Company A cancel, renegotiate or obtain substitutes?

32. Case-Law Classification

The first four are particularly useful for the contract-formation and attribution dimension. The later cases are primarily useful for AI, automated decision and product-liability analogies.

33. Important Distinction: Direct vs Analogical Authorities

There is currently no established CJEU case saying that an autonomous AI negotiation agent is itself legally liable for a contract it negotiated.

Therefore, it would be incorrect to present SCHUFA or Dun & Bradstreet as direct autonomous-negotiation cases.

A legally careful analysis instead combines:

Electronic contracting cases

Agency/attribution principles

Automated-decision cases

Software/product-liability rules

National contract and tort/delict law.

34. Overall Legal Framework

The liability structure can be represented as:

Autonomous AI negotiator

↓

Electronic communication

↓

Authority / attribution

↓

Offer / counteroffer / acceptance

↓

Contract

↓

AI error / software defect / unauthorised action

↓

Causation

↓

Economic or other legally recognised damage

↓

Contract / product liability / tort-damages remedy

35. Conclusion

Autonomous negotiation systems are likely to produce a new category of civil disputes, but European law currently addresses them through existing legal doctrines rather than a single autonomous-negotiation liability regime.

The most important principles are:

AI autonomy does not automatically create separate legal personality.

Electronic communications can constitute legally relevant contractual declarations.

Authority and attribution are central where an AI exceeds its instructions.

A technical error does not automatically erase contractual consequences.

Software and AI systems are increasingly covered by European product-liability legislation.

A software defect and a mere commercially bad negotiation must be distinguished.

Causation and proof of economic loss remain essential.

Consumer-protection rules continue to apply to automated contracting.

AI developers, users and integrators may have different responsibilities.

The same dispute may involve contract, tort/delict, product liability and regulatory rules simultaneously.

The particularly important authorities for examination are BGH VIII ZR 79/04, BGH VIII ZR 289/09, El Majdoub (C-322/14), Content Services (C-49/11), SCHUFA (C-634/21), Dun & Bradstreet (C-203/22), Boston Scientific (C-503/13 and C-504/13), and Sanofi Pasteur (C-621/15). (JurPC)

Exam / Revision Keywords

Autonomous negotiation — AI contracting — electronic agent — machine-to-machine contract — contractual attribution — agency — apparent authority — offer — acceptance — counteroffer — automated contracting — software defect — AI liability — product liability — Product Liability Directive — contractual mistake — algorithmic error — AI hallucination — cybersecurity — system integrator — developer liability — causation — economic loss — lost profits — mitigation — consumer protection — electronic evidence — audit logs — black-box AI — AI-to-AI contracting — autonomous renegotiation — contractual authority — digital contracting.