AI Hallucination Damages Claims in Europe

1. Introduction

AI hallucination damages claims concern legal claims arising when an artificial-intelligence system produces false, fabricated, materially misleading, or unsupported information and a person or organization suffers legally recognizable harm as a consequence.

Examples include an AI system:

• inventing a court case;
• fabricating a quotation;
• falsely accusing someone of criminal conduct;
• creating a nonexistent medical diagnosis;
• giving incorrect financial information;
• inventing contractual terms;
• generating false information about a company;
• producing a fabricated professional report;
• providing false information to a public authority;
• generating defamatory statements;
• supplying inaccurate information that causes economic loss.

There is no single autonomous European cause of action called “AI hallucination damages.” Liability is generally constructed from existing doctrines such as:

• negligence/delict;
• defamation;
• product liability;
• professional negligence;
• contractual liability;
• consumer protection;
• data protection;
• privacy and personality rights;
• misrepresentation;
• economic-loss doctrines;
• medical liability;
• financial-services regulation; and
• fundamental-rights law.

The central question is:

Who was responsible for the AI output, what duty existed to prevent or correct the hallucination, could the error reasonably have been foreseen, and did reliance on the hallucinated output cause legally recoverable damage?

2. What Is an AI Hallucination?

An AI hallucination occurs when an AI system generates information that appears plausible but is:

• factually incorrect;
• fabricated;
• unsupported by its underlying data;
• presented with unjustified confidence; or
• inconsistent with the available evidence.

Examples include:

Fabricated legal authority

AI invents:

“The Supreme Court held in X v Y that…”

when no such case exists.

Fabricated person

AI states that a person committed misconduct when there is no evidence.

Fabricated financial information

AI claims:

“Company X reported a €500 million profit”

when the company never made such a statement.

Medical hallucination

AI states:

“Your test results prove that you have disease X”

when the underlying medical information does not support the conclusion.

Contractual hallucination

AI invents a contractual clause and presents it as part of an existing agreement.

3. Hallucination Alone Does Not Automatically Create Liability

This is a crucial principle.

An incorrect AI output does not automatically mean that the provider or user is liable for damages.

A claimant generally needs to establish some combination of:

1. a legal duty;
2. breach;
3. causation;
4. legally recognizable damage;
5. foreseeability;
6. and, depending on the cause of action, reliance or other required elements.

Thus:

False output + no reliance + no legally recognized damage ≠ automatic damages.

Conversely:

False output + foreseeable reliance + breach of duty + provable damage = potentially strong claim.

4. The Basic Liability Chain

AI hallucination claims can be analyzed through this sequence:

AI system

↓

Hallucinated information

↓

Publication / communication

↓

Reliance or dissemination

↓

Harm

↓

Legal duty / breach

↓

Causation

↓

Damage

↓

Remedy

The most difficult elements are usually:

• identifying the legally responsible actor;
• establishing the applicable duty;
• proving that the information was false;
• proving reasonable reliance;
• proving causation;
• quantifying the resulting loss.

5. Who Can Be Liable?

Potential defendants include:

AI developer

The developer created the model.

AI provider

The provider operates the service and makes it available to users.

Platform operator

The platform incorporates AI into its services.

Professional user

A lawyer, doctor, accountant, financial adviser or consultant relies on AI and communicates its output to a client.

Employer

An employee uses an AI system within employment.

Publisher

A business publishes AI-generated content.

Public authority

An authority uses AI-generated information in administrative decision-making.

Corporate directors

Potential liability may arise where directors personally breach their own duties.

6. The Importance of Intended Use

Liability is heavily influenced by how the AI system was intended to be used.

An AI system marketed as:

“a creative writing assistant”

presents a different legal situation from one marketed as:

“a reliable legal research assistant.”

Similarly:

“AI brainstorming tool”

is different from:

“AI medical diagnostic system.”

The more an AI provider represents that the system is suitable for a professional or high-risk purpose, the stronger the argument may become that reasonable safeguards were required.

7. European Legal Framework

Relevant legal areas include:

Contract

Where AI services are supplied under a contract.

Tort/delict

Where careless AI-generated information causes damage.

Defamation

Where false statements damage reputation.

Consumer protection

Where consumers receive misleading AI-generated information.

Product liability

Where defective AI software or an AI-enabled product causes legally recognized damage.

GDPR

Where hallucinated information concerns an identifiable individual and is processed or disseminated as personal data.

Professional negligence

Where a professional relies on or communicates AI-generated information without appropriate verification.

Fundamental rights

Particularly:

• privacy;
• reputation;
• fair trial;
• property;
• physical integrity.

8. Case Law

European courts have not yet developed a large body of judgments specifically titled “AI hallucination damages.”

Accordingly, the following cases are important because they establish legal principles that can apply directly or by analogy.

Case 1 — Nowak v Data Protection Commissioner

CJEU, Case C-434/16, 20 December 2017

Principle

The CJEU interpreted the concept of personal data broadly.

Information relating to an identifiable individual can fall within data-protection law.

Relevance to AI hallucinations

Suppose an AI system generates:

“John Smith was convicted of fraud.”

when John Smith was never convicted.

If the information identifies John Smith, the AI-generated information may create data-protection implications depending upon the circumstances in which it is processed and disseminated.

Potential issues include:

• accuracy;
• rectification;
• unlawful processing;
• disclosure;
• profiling.

Damages relevance

The case is important because it demonstrates that incorrect information concerning an identifiable individual is not merely an abstract “AI error.”

It can become a legally significant data-processing problem.

Authority type: Directly relevant data-protection analogy.

Case 2 — Google Spain SL v AEPD and Mario Costeja González

CJEU, Case C-131/12, 13 May 2014

Principle

The CJEU recognized significant obligations concerning personal information processed through technological information systems.

AI hallucination relevance

Imagine that an AI system generates a false statement about a person's:

• criminal history;
• professional misconduct;
• financial condition;
• medical condition.

If that information is incorporated into an automated search, profile or decision-making system, the legal consequences can extend beyond the original hallucination.

Governance lesson

Organizations operating information-processing technologies must consider:

• accuracy;
• correction;
• deletion;
• dissemination;
• individual rights.

Authority type: Technology/data authority highly relevant by analogy.

Case 3 — GC and Others v Commission nationale de l'informatique et des libertés

CJEU, Joined Cases C-136/17 and C-137/17, 24 September 2019

The CJEU addressed sensitive categories of personal data and search-engine processing.

Relevance

AI hallucinations may concern extremely sensitive information, including:

• health;
• criminal allegations;
• political information;
• sexuality;
• religion.

False AI-generated information in these categories can cause particularly serious harm.

Governance implication

The more sensitive the subject matter, the greater the importance of:

• accuracy;
• safeguards;
• lawful processing;
• proportionality;
• correction mechanisms.

Authority type: Data-protection analogy.

Case 4 — Delfi AS v Estonia

ECtHR Grand Chamber, 16 June 2015

Importance

The case concerned liability of an internet portal for unlawful user-generated comments.

Although it did not involve AI, it is highly relevant to questions concerning online intermediaries and harmful automated content.

Principle

The ECtHR examined the balance between:

• freedom of expression;
• reputation;
• protection of others;
• intermediary responsibility.

AI relevance

Consider an AI platform that generates a false defamatory statement about an identifiable person.

The legal question may include:

• Who published the statement?
• Who controlled the platform?
• Was the harm foreseeable?
• Was there an effective correction mechanism?
• What responsibility did the platform assume?

Delfi demonstrates that intermediary status does not necessarily resolve liability automatically.

Authority type: Strong analogical authority for AI-generated harmful content.

Case 5 — MTE and Index.hu Zrt v Hungary

ECtHR, 2 February 2016

Principle

The ECtHR further examined intermediary liability and freedom of expression.

The Court emphasized the need to balance:

• protection of reputation;
• freedom of expression;
• intermediary responsibility;
• context and proportionality.

AI hallucination relevance

An AI-generated false statement can spread rapidly.

The legal analysis may need to distinguish between:

• a private AI interaction;
• publication to the public;
• automated indexing;
• republication;
• commercial dissemination.

The greater the dissemination, the greater the potential reputational harm.

Authority type: Analogical platform/content-liability authority.

Case 6 — Axel Springer AG v Germany

ECtHR Grand Chamber, 7 February 2012

Principle

The ECtHR considered the balance between freedom of expression and protection of reputation/private life.

Relevant considerations include:

• contribution to a matter of public interest;
• status of the person concerned;
• method of obtaining information;
• accuracy;
• consequences of publication.

AI hallucination relevance

These factors can become particularly important when AI generates false information concerning a public figure.

For example:

AI falsely states that a politician was involved in corruption.

The legal analysis cannot simply stop at:

“The statement was generated by AI.”

It must examine:

• whether the statement was factual;
• whether it was false;
• how it was disseminated;
• the public interest;
• reputational consequences.

Authority type: Defamation/free-expression analogy.

Case 7 — Von Hannover v Germany

ECtHR Grand Chamber, 7 February 2012

Principle

The ECtHR addressed protection of private life and reputation against publication.

AI relevance

AI systems may generate information involving private individuals.

False AI outputs concerning:

• relationships;
• family circumstances;
• medical information;
• private conduct;

can interfere with Article 8 rights.

Liability significance

The case reinforces that technological publication can engage privacy and personality rights.

Authority type: Privacy/reputation analogy.

Case 8 — Mosley v United Kingdom

ECtHR, 10 May 2011

The case concerned privacy and publication of personal information.

AI relevance

AI systems can automatically create or distribute highly personal content.

The governance problem becomes especially serious when an AI system:

• generates intimate false information;
• publishes it automatically;
• reproduces it repeatedly;
• fails to remove it after notification.

Principle

European law recognizes significant protection for private life against unjustified publication.

Authority type: Privacy authority applied by analogy.

Case 9 — Hedley Byrne & Co Ltd v Heller & Partners Ltd

House of Lords, 1964

Importance

This is a foundational common-law authority concerning negligent misstatements.

Principle

A person who provides information in circumstances involving an appropriate assumption of responsibility may potentially be liable where the information is negligently supplied and relied upon.

AI relevance

Suppose a bank provides an AI financial assistant and tells customers:

“You can rely upon our AI for personalized investment recommendations.”

The AI generates materially false information.

The legal question could include:

• Was responsibility assumed?
• Was reliance foreseeable?
• Was the system reasonably supervised?
• Was the information negligently generated or communicated?
• Did the investor suffer financial loss?

Authority type: Strong comparative authority for AI-generated misinformation.

Case 10 — Caparo Industries plc v Dickman

House of Lords, 1990

Principle

The case developed the framework for determining whether a duty of care exists.

Important factors include:

• foreseeability;
• proximity;
• fairness, justice and reasonableness.

AI relevance

The case is useful when determining whether an AI developer owes a duty directly to an end user.

For example:

AI developer → corporate customer → consumer

The developer may argue that it had no sufficiently proximate relationship with the ultimate consumer.

The claimant may argue that the developer:

• specifically designed the system for that use;
• knew how it would be deployed;
• controlled safety mechanisms;
• assumed responsibility through representations.

The precise outcome will depend on national law.

Authority type: Comparative negligence authority.

9. Defamation and AI Hallucinations

Defamation may be one of the most obvious areas of AI hallucination liability.

Suppose an AI system falsely states:

“A is a convicted fraudster.”

Potential legal elements may include:

1. statement concerning the claimant;
2. publication to a third party;
3. defamatory meaning;
4. falsity or other jurisdiction-specific requirements;
5. legally recognized harm.

National European defamation law varies significantly.

Therefore, the precise cause of action depends on the jurisdiction.

10. AI Hallucination and Reputation

Reputational damage may include:

• loss of employment;
• loss of customers;
• professional embarrassment;
• business losses;
• social exclusion;
• psychological distress;
• loss of professional opportunities.

A false AI-generated allegation may be particularly harmful because AI outputs can be:

• copied;
• indexed;
• shared;
• incorporated into reports;
• reproduced by other AI systems.

This creates a propagation problem.

11. AI Hallucination and Data Accuracy

GDPR Article 5 includes the principle of accuracy.

An organization processing personal data must take appropriate steps concerning inaccurate information.

This becomes important where AI generates or stores false personal information.

Example

An AI recruitment system creates the profile:

“Candidate has previously been dismissed for misconduct.”

The information is false.

If the employer uses the AI-generated profile in hiring, possible legal issues include:

• inaccurate personal data;
• unlawful profiling;
• discriminatory decision-making;
• employment-law violations;
• damages.

12. AI Hallucination and Automated Decision-Making

The SCHUFA line of case law is particularly important here.

An AI hallucination may not merely produce false information.

It may trigger an automated decision.

For example:

AI hallucination

↓

“Applicant has serious debt problems.”

↓

Credit score falls.

↓

Loan refused.

↓

Applicant loses opportunity to purchase property.

This is much more legally significant than an isolated inaccurate chatbot response.

The system has transformed false information into a consequential decision.

13. AI Hallucination in Healthcare

Healthcare creates especially serious liability risks.

Examples:

• fabricated diagnosis;
• incorrect medication information;
• false interpretation of imaging;
• invented medical research;
• fabricated clinical guidelines.

Potential claims may involve:

• medical negligence;
• product liability;
• professional negligence;
• informed consent;
• contract;
• fundamental rights.

Important principle

A doctor who relies on AI remains responsible for exercising appropriate professional judgment.

Likewise, a medical AI provider may face responsibility depending upon:

• product design;
• intended use;
• warnings;
• validation;
• safety controls.

14. AI Hallucination in Financial Services

Examples include:

• false stock prices;
• fabricated earnings;
• nonexistent investment products;
• false statements about regulatory approval;
• incorrect risk assessments.

Potential claims include:

• negligent misstatement;
• breach of advisory contract;
• financial-services misconduct;
• consumer protection;
• professional negligence.

The Hedley Byrne principle concerning reliance on professional information is particularly useful as a comparative authority.

15. AI Hallucination in Legal Services

Legal AI may fabricate:

• cases;
• statutes;
• quotations;
• legal propositions;
• procedural rules.

This can cause:

• missed limitation periods;
• incorrect pleadings;
• wasted litigation costs;
• adverse judgments;
• professional disciplinary consequences.

A lawyer cannot necessarily avoid responsibility by saying:

“The AI gave me the wrong case.”

The professional's own duties concerning verification and competence remain relevant.

16. AI Hallucination and Professional Negligence

The professional-negligence question can be expressed as:

Would a reasonably competent professional using AI have independently verified the information before relying upon or communicating it?

The answer will depend upon:

• the profession;
• risk level;
• complexity;
• available verification;
• representations made by the AI provider;
• urgency;
• professional standards.

The greater the potential harm, the stronger the argument for independent verification.

17. AI Hallucination and Consumer Protection

Consumer law becomes important where an AI product is marketed using statements such as:

• “always accurate”;
• “expert-level advice”;
• “error-free”;
• “professionally verified”;
• “safe to rely upon.”

If such claims are materially misleading, consumer-protection law may become relevant.

A disclaimer saying:

“AI can make mistakes”

does not necessarily cure every misleading representation made elsewhere in the service.

18. AI Hallucination and Product Liability

The emerging EU product-liability framework is important for AI-enabled products.

However, a crucial distinction must be maintained:

Not every incorrect AI response automatically constitutes a defective product causing recoverable damage under product-liability law.

The legal classification of the software and the type of damage matter.

The claim becomes stronger where AI is incorporated into a product whose defective operation causes:

• personal injury;
• property damage;
• other legally recognized harm.

19. Economic Loss

Pure financial loss is particularly complicated.

Suppose:

AI gives false stock advice → investor loses €50,000.

The investor cannot simply assume that every European legal system provides tort damages.

The claimant may need to rely upon:

• contract;
• financial-services regulation;
• negligent misstatement;
• professional negligence;
• consumer law;
• specific statutory compensation mechanisms.

The availability of damages depends heavily on national law.

20. Causation

Causation is often the hardest part.

Suppose:

AI says Buy X

↓

Investor buys X

↓

Market falls 30%

Does the AI provider automatically owe 30% of the loss?

No.

The claimant may need to demonstrate:

• reasonable reliance;
• that the information was materially wrong;
• that the error caused the transaction;
• that the loss resulted from the error rather than ordinary market conditions.

21. Counterfactual Causation

Courts may effectively ask:

What would have happened if the AI had provided accurate information?

Example:

Actual scenario

AI says:

“Company X has secured regulatory approval.”

Investor purchases €100,000 of shares.

The statement was false.

Shares collapse when the truth emerges.

Counterfactual

Had accurate information been provided, the investor would not have purchased the shares.

The claimant's causal case is considerably stronger.

22. Reliance

Reliance is particularly important in information-based claims.

Factors may include:

• Was the AI presented as authoritative?
• Did the provider encourage reliance?
• Did the customer know it could hallucinate?
• Were warnings prominent?
• Was human verification recommended?
• Was the information independently verified?
• Was the user sophisticated?

A sophisticated professional may be expected to verify high-risk information more carefully than an ordinary consumer.

23. Foreseeability

A key question is:

Was the possibility of hallucination reasonably foreseeable?

Modern generative AI systems are known to produce inaccurate information.

Therefore, an organization deploying such a system in a high-risk environment may have difficulty arguing that hallucination was entirely unforeseeable.

The more serious the use case, the stronger the expectation of safeguards.

24. Human Oversight

Human oversight can substantially affect liability.

Weak governance

AI generates legal advice.

Employee automatically forwards it to client.

No verification.

Stronger governance

AI generates draft.

Lawyer verifies:

• cases;
• statutes;
• quotations;
• factual claims.

Only verified information is communicated.

If an error nevertheless occurs despite reasonable safeguards, the defendant may have a stronger defence.

25. AI Hallucination and Warnings

Warnings can be relevant but are not automatically decisive.

A warning saying:

“This AI may occasionally produce inaccurate information”

may be helpful.

But if the provider simultaneously markets the system as:

“professional-grade and reliable for legal decisions,”

the overall presentation may still create liability issues.

The court may examine the substance and context of the service, not merely the existence of a disclaimer.

26. AI Hallucination and Corporate Governance

Boards should establish controls concerning:

• approved AI use cases;
• prohibited high-risk uses;
• verification requirements;
• incident reporting;
• employee training;
• model testing;
• vendor management;
• customer warnings;
• record retention.

If management receives repeated reports that:

“The AI is inventing facts”

but does nothing, the resulting governance failure may become significant evidence in subsequent litigation.

27. AI Hallucination and Directors

Directors are not automatically personally liable for every AI failure.

Personal liability may become relevant where directors themselves:

• authorize unlawful conduct;
• knowingly ignore serious risks;
• fail to perform their own duties;
• misrepresent AI performance;
• conceal material incidents.

Traditional authorities concerning directors' duties remain relevant by analogy.

The key principle is:

Corporate AI liability and individual director liability are separate questions.

28. Evidence in Hallucination Claims

Evidence may include:

AI output

The exact generated response.

Prompt

What the user asked the AI.

System instructions

How the model was configured.

Model version

Which model generated the answer.

Logs

When and how the output was generated.

Training/retrieval information

Where applicable.

Verification records

Whether a human reviewed the information.

Warnings

What the user was told about reliability.

Communications

Emails, messages and reports showing reliance.

Damage evidence

Financial records, medical records, employment consequences or reputational evidence.

29. Importance of Preserving the Original AI Output

AI systems can change rapidly.

A later version may produce a different response.

Therefore, litigation may require preservation of:

• original prompt;
• original output;
• model version;
• date/time;
• retrieval context;
• source material;
• system instructions.

Without this evidence, it may become difficult to establish exactly what the AI said.

30. Defences

A defendant may argue:

1. No duty

The service was not intended to provide authoritative information.

2. No reasonable reliance

The user was expressly warned to verify the output.

3. Independent decision

The claimant independently acted.

4. No causation

The loss resulted from another event.

5. Contributory negligence

The claimant ignored obvious warnings.

6. Adequate safeguards

The defendant had reasonable testing and monitoring.

7. Third-party fault

Another person modified or misused the AI output.

8. Freedom of expression

Relevant particularly to defamation claims.

9. Lack of legally recoverable damage

An incorrect output may not have caused compensable harm.

31. Remedies

Depending upon the jurisdiction and legal basis, remedies may include:

Damages

For proven economic or non-economic loss.

Correction

Particularly important for inaccurate personal information.

Deletion

Where legally available.

Injunction

Preventing further publication.

Retraction

Correcting defamatory information.

Contractual damages

For breach of service obligations.

Regulatory penalties

Against providers or deployers.

Reassessment

Where an AI-generated decision affected an individual.

Restoration

Where AI-generated misinformation caused an identifiable loss capable of being reversed.

32. Special Problem: AI-to-AI Propagation

A distinctive AI problem is recursive propagation.

Example:

AI A fabricates information.

↓

Website publishes it.

↓

AI B scrapes website.

↓

AI B treats it as factual.

↓

AI C trains or retrieves from AI B's output.

↓

False information becomes widespread.

This creates difficult questions concerning:

• original responsibility;
• intermediary responsibility;
• reasonable verification;
• causation;
• contributory fault.

The legal system may need to distinguish between:

creating false information

and

republishing false information after reasonable notice.

33. AI Hallucination and Defamation: A Typical Claim

Example

An AI chatbot states:

“Person A committed financial fraud.”

Person A has never committed fraud.

The statement is published publicly.

Person A loses clients.

A possible claim might involve:

False AI output

→

publication

→

reputational harm

→

economic loss

→

defamation/privacy claim

The claimant may seek:

• removal;
• correction;
• injunction;
• damages.

The precise requirements depend upon national defamation law.

34. AI Hallucination and Financial Loss: A Typical Claim

Example

AI financial assistant says:

“Investment X is government guaranteed.”

The statement is false.

Customer invests €100,000.

Investment collapses.

The customer may potentially argue:

1. the provider assumed responsibility;
2. the statement was false;
3. reliance was foreseeable;
4. safeguards were inadequate;
5. the investment was made because of the statement;
6. financial loss resulted.

The defendant may argue:

• customer received risk warnings;
• the AI was general information only;
• the customer independently invested;
• market losses were unforeseeable;
• no legally recognized duty existed.

35. AI Hallucination and Legal Services: A Typical Claim

Suppose an AI legal assistant invents a limitation period.

A lawyer relies upon the output.

The limitation period expires.

The client loses the claim.

Potential liability may involve:

• lawyer's professional negligence;
• breach of retainer;
• failure to verify;
• causation;
• loss of litigation opportunity.

The lawyer may not be able simply to transfer responsibility to the AI provider.

36. AI Hallucination and Public Authorities

A government agency might use AI to summarize:

• evidence;
• applications;
• criminal records;
• welfare information.

If the AI invents a material fact and the authority relies upon it, potential issues include:

• administrative unlawfulness;
• procedural fairness;
• failure to give adequate reasons;
• discrimination;
• data-protection violations;
• judicial review;
• damages where available.

The M.M. line of CJEU authority is useful in demonstrating the importance of effective procedural safeguards in administrative decision-making.

37. Important Distinction: Error vs Negligence

This distinction is essential.

AI error

The system produces an incorrect statement.

AI negligence

The responsible actor failed to take precautions that a reasonable actor should have taken.

For example:

A research AI unexpectedly generates one incorrect citation despite reasonable controls.

This is different from:

A company knows the AI routinely fabricates citations but markets it as a professional legal research tool without verification controls.

The second situation presents a much stronger negligence argument.

38. Important Distinction: Provider vs User

A user who knowingly misuses AI may bear significant responsibility.

For example:

Company uses a general-purpose chatbot to make medical decisions despite explicit warnings against doing so.

The provider's liability may be weaker.

Conversely:

Provider specifically markets the system as an autonomous medical diagnostic system.

The provider may face considerably greater scrutiny.

Intended use and representations therefore matter greatly.

39. Comparative Case Table

40. The Strongest Liability Scenarios

AI hallucination liability becomes particularly strong where several factors coexist:

1. High-risk use

Medical, legal, financial or governmental use.

2. Express representation of reliability

The provider markets the system as authoritative.

3. Foreseeable reliance

The provider knows users will rely upon the output.

4. Known hallucination risk

The provider knows the model produces false information.

5. Inadequate safeguards

No verification, monitoring or human review.

6. Material falsehood

The error significantly affects the claimant.

7. Actual reliance

The claimant acted because of the information.

8. Provable damage

Financial, reputational, physical, professional or other legally recognized harm.

41. A Ten-Part Legal Test

When evaluating an AI hallucination damages claim, ask:

1. What exactly did the AI say?
2. Was the statement objectively false?
3. Who supplied or operated the AI?
4. What purpose was the AI intended for?
5. What legal duty applied?
6. Was hallucination reasonably foreseeable?
7. Were adequate safeguards implemented?
8. Did the claimant reasonably rely upon the output?
9. Did the output cause legally recognizable damage?
10. What remedy is available under the applicable national law?

42. Conclusion

AI hallucination damages claims are an emerging European liability field rather than a fully autonomous cause of action. Existing legal doctrines provide the principal foundation.

The most important areas are:

• negligence and negligent misstatement;
• defamation;
• privacy and personality rights;
• GDPR accuracy and automated decision-making;
• consumer protection;
• professional negligence;
• contractual liability;
• product liability;
• administrative law.

The most relevant authorities include Nowak, Google Spain, GC and Others, Delfi, MTE and Index.hu, Axel Springer, Von Hannover, Mosley, Hedley Byrne, and Caparo.

The central legal distinction is:

An AI hallucination is an error; liability arises only when the error falls within a legally recognized duty and produces legally attributable harm.

The strongest claim generally follows this structure:

AI hallucination → foreseeable risk → inadequate governance/verification → publication or reliance → causation → economic, physical, reputational or privacy harm → legally recognized damage → remedy.

For European litigation, the most important practical evidence will usually be the original prompt and output, model/version information, system configuration, warnings, audit logs, verification records, communications showing reliance, and evidence of the resulting damage.