Civil Law and AI Content Moderation Liability in Europe

1. Introduction

AI content-moderation liability concerns civil and related legal responsibility arising when an AI system:

• removes lawful content;
• fails to remove unlawful content;
• incorrectly classifies speech as hate speech, defamation, terrorism or illegal material;
• leaves unlawful material online;
• demonetises or restricts a user's account;
• automatically suspends an account;
• amplifies harmful content;
• makes discriminatory moderation decisions;
• fails to detect illegal content;
• over-removes lawful speech;
• causes reputational, financial or other legally recognised harm.

Modern platforms increasingly use machine-learning classifiers, large language models, image-recognition systems, automated risk scoring and hybrid human-AI moderation.

European liability is consequently divided among several legal fields:

1. Digital Services Act (DSA);
2. national civil/tort/delict law;
3. defamation and personality rights;
4. privacy and data protection;
5. intellectual-property law;
6. consumer law;
7. contractual law;
8. Article 8 and Article 10 ECHR;
9. intermediary-liability principles.

A crucial point is that Europe does not have a single general civil-liability code for AI moderation. EU legislation establishes important harmonised obligations, while national law continues to determine many questions of damages, causation and civil liability.

2. Meaning of AI Content Moderation

Content moderation means identifying, reviewing, restricting, ranking or removing digital content.

AI moderation can operate at several stages:

Stage 1 — Detection

AI identifies potentially unlawful or harmful content.

Stage 2 — Classification

The system classifies it as:

• hate speech;
• terrorism;
• pornography;
• defamation;
• misinformation;
• copyright infringement;
• harassment;
• illegal goods.

Stage 3 — Decision

The system automatically decides:

• remove;
• block;
• restrict;
• demonetise;
• reduce visibility;
• suspend account;
• refer to human moderator.

Stage 4 — Appeal

The user challenges the decision.

Stage 5 — Human review

A moderator may confirm or reverse the AI decision.

The legal problem is therefore not simply:

"Was the AI correct?"

It is:

Did the platform comply with its applicable legal duties, and did the AI-assisted moderation process unlawfully cause legally recoverable harm?

3. Digital Services Act

The most important current EU framework is the Digital Services Act, Regulation (EU) 2022/2065.

The DSA applies to intermediary services and establishes rules concerning:

• illegal-content notices;
• hosting liability;
• statements of reasons;
• internal complaint mechanisms;
• out-of-court dispute settlement;
• systemic-risk management;
• transparency;
• recommender systems;
• very large online platforms and search engines.

The DSA's hosting safe-harbour provision generally protects a provider where it lacks actual knowledge of illegal content and, after obtaining such knowledge, acts expeditiously to remove or disable access.

Importantly, Article 54 DSA expressly provides a right to seek compensation under Union and national law for damage or loss caused by an intermediary provider's infringement of its DSA obligations.

Thus, DSA compliance can become relevant to a private civil claim.

4. AI Moderation and Civil Liability

A useful liability structure is:

AI moderation decision

↓

Platform's legal/contractual duty

↓

Wrongful removal OR failure to remove

↓

Damage

↓

Causation

↓

Civil remedy

Examples:

Wrongful removal

A journalist's legitimate article is incorrectly classified as hate speech.

Potential consequences:

• loss of audience;
• contractual loss;
• reputational harm;
• loss of advertising revenue.

Failure to remove

A platform leaves clearly unlawful defamatory content online despite receiving an adequate notice.

Potential consequences:

• continuing reputational injury;
• financial loss;
• privacy harm.

The applicable civil remedy depends heavily on national law.

5. The Importance of the DSA Safe Harbour

The DSA does not establish a rule that platforms are automatically liable for everything users publish.

Article 6 provides a conditional hosting liability exemption.

A hosting provider can generally rely upon the exemption if:

1. it lacks actual knowledge of illegal activity/content; or
2. for damages, it lacks awareness of facts or circumstances from which the illegality is apparent; and
3. once it obtains the required knowledge, it acts expeditiously. 

Therefore:

AI moderation itself does not automatically destroy safe-harbour protection.

The important question is how the provider's system and conduct interact with its statutory obligations.

6. No General Monitoring Obligation

European intermediary law has traditionally resisted imposing a general obligation on providers to monitor all information.

The CJEU's Glawischnig-Piesczek judgment expressly addressed Article 15 of the former E-Commerce Directive and the absence of a general monitoring obligation.

The DSA maintains this fundamental principle.

This creates an important distinction:

General monitoring

"Check everything users post."

Generally prohibited.

Specific moderation duty

"Remove this particular unlawful content after valid notice."

Potentially required.

AI therefore can be used for moderation, but European law does not simply transform the platform into an unlimited general surveillance system.

7. Case Law

Case 1 — Delfi AS v Estonia

Court: European Court of Human Rights, Grand Chamber
Application No.: 64569/09
Judgment: 16 June 2015

This is one of Europe's foundational intermediary-liability cases.

Facts

Delfi operated a major online news portal.

Readers posted offensive comments concerning an individual.

The Estonian courts imposed liability on Delfi for the third-party comments.

Delfi argued that this violated its freedom of expression under Article 10 ECHR.

The Grand Chamber examined whether imposing liability was compatible with Article 10.

Principle

The Court considered several factors, including:

• context of the comments;
• nature of the comments;
• measures taken by the platform;
• identity and position of the persons responsible;
• consequences for the victim;
• effectiveness of the domestic proceedings.

The case concerned particularly serious comments, including hate speech and incitement to violence.

AI relevance

An AI moderation system could be used to:

• detect hate speech;
• prioritise dangerous comments;
• identify threats;
• trigger human review.

Delfi demonstrates that platform liability must be considered in the context of the competing rights of:

victim ↔ platform ↔ speaker.

Important limitation

Delfi was decided before the DSA and did not concern AI moderation.

Classification: Analogical but foundational intermediary-liability authority.

8. Case 2 — Magyar Tartalomszolgáltatók Egyesülete and Index.hu Zrt v Hungary

Court: ECtHR
Application No.: 22947/13
Judgment: 2 February 2016

Facts

Internet users posted vulgar and offensive comments on Hungarian online portals.

The national courts imposed liability on the operators.

The ECtHR examined whether this interfered disproportionately with Article 10 freedom of expression.

Decision

The Court found a violation of Article 10.

The Court distinguished the case from Delfi, particularly because the comments were not the same type of clearly unlawful hate speech or incitement to violence involved in Delfi.

AI relevance

This case is highly useful for AI moderation because automated systems frequently face the difficult boundary between:

• offensive speech;
• insulting speech;
• controversial opinions;
• unlawful hate speech.

An AI classifier may classify all three as "harmful."

But legally they are not necessarily equivalent.

Key lesson

AI moderation should not treat every offensive statement as automatically unlawful content.

Classification: Direct online-comment liability authority; analogical for AI.

9. Case 3 — Eva Glawischnig-Piesczek v Facebook Ireland Ltd, C-18/18

Court: CJEU
Date: 3 October 2019

Facts

The dispute concerned allegedly defamatory content posted on Facebook.

An Austrian court required Facebook to remove unlawful material.

The CJEU considered:

• hosting-provider liability;
• injunctions;
• removal of identical content;
• equivalent content;
• territorial scope;
• the prohibition on general monitoring.

The CJEU confirmed that an injunction can, in appropriate circumstances, require removal or blocking of identical or equivalent unlawful information while remaining subject to the limits of EU law.

AI significance

This is one of the most important cases for automated moderation.

An AI system may be used to identify:

"essentially identical" or legally equivalent unlawful material.

However, the distinction between:

• identical content;
• equivalent unlawful content;
• merely similar content

is extremely important.

Example

If a court declares:

"Statement X is defamatory."

An automated system may potentially identify reposts of the same statement.

But an AI system should not automatically conclude:

"Every criticism of the claimant is defamatory."

That could raise over-removal and freedom-of-expression problems.

Classification: Direct platform-content-moderation authority.

10. Case 4 — L'Oréal SA and Others v eBay International AG, C-324/09

Court: CJEU Grand Chamber
Date: 12 July 2011

Facts

L'Oréal brought proceedings concerning allegedly infringing goods offered by users through eBay.

The CJEU considered intermediary liability under the former E-Commerce Directive and intellectual-property enforcement.

The Court examined whether an online marketplace operator played a sufficiently active role to affect its position under the intermediary-liability framework.

Principle

A provider engaged in merely technical, automatic and passive processing can benefit from the hosting exemption where the relevant conditions are satisfied.

However, an active role giving knowledge of or control over the information can affect the availability of the exemption.

AI relevance

Modern AI moderation systems can involve:

• ranking;
• classification;
• recommendation;
• prioritisation;
• automated intervention.

This creates an important legal question:

Does extensive technological involvement make the provider an "active" participant in the underlying content?

The answer cannot be derived simply from the fact that AI is being used. The legal analysis concerns the nature of the provider's role.

Classification: Foundational intermediary-liability authority.

11. Case 5 — Google France SARL and Google Inc. v Louis Vuitton, Joined Cases C-236/08 to C-238/08

Court: CJEU Grand Chamber
Date: 23 March 2010

Facts

Google's AdWords system displayed advertisements triggered by keywords corresponding to trademarks.

The dispute concerned the liability of the search-engine operator.

The CJEU considered the conditions under which an information-society service provider could benefit from the intermediary liability exemption.

Principle

The CJEU distinguished the provider's technological role from circumstances in which it acquires knowledge or control relevant to the stored information.

AI relevance

The case is useful by analogy for AI systems that:

• classify;
• rank;
• recommend;
• select;
• display;
• target content.

The important legal question is not simply:

"Does an algorithm make the decision?"

It is:

What role does the platform play, and what knowledge or control does it have concerning the allegedly unlawful material?

Classification: Analogical AI-platform liability authority.

12. Case 6 — YouTube and Cyando, Joined Cases C-682/18 and C-683/18

Court: CJEU
Date: 22 June 2021

Facts

The cases concerned copyright-infringing content uploaded by users to online platforms.

The CJEU considered when platform operators can benefit from the E-Commerce Directive intermediary-liability framework and the circumstances in which their conduct goes beyond merely providing a platform.

Principle

The CJEU examined whether the platform operator's conduct was sufficiently connected with the communication of protected works to the public to affect liability.

It reiterated the importance of whether the provider plays a role going beyond merely providing the platform.

AI relevance

This is relevant where AI is heavily integrated into:

• upload filtering;
• content ranking;
• recommendation;
• copyright detection;
• automated blocking.

An AI moderation platform may need to distinguish between:

technical hosting

and

active participation in the content process.

Classification

Direct platform-liability authority; indirect AI-moderation authority.

13. Case 7 — Sanchez v France

Court: ECtHR Grand Chamber
Application No.: 45581/15
Judgment: 15 May 2023

Facts

The case concerned comments posted by third parties on a public Facebook wall.

French courts held the applicant criminally responsible in relation to the failure to remove certain comments.

The Grand Chamber considered the compatibility of that responsibility with Article 10 ECHR.

Significance

The case demonstrates that liability for third-party online comments can involve an individual who controls a social-media page, not only a large technology platform.

The ECtHR's approach considered:

• the context;
• the nature of the comments;
• the applicant's role;
• the accessibility of the page;
• the measures that could reasonably have been taken;
• the applicable sanctions.

The case has subsequently been discussed extensively in relation to the development of European online-intermediary liability.

AI relevance

It highlights the importance of determining who actually has control over the relevant online space.

For AI systems, this translates into questions such as:

• Who configured the moderation rules?
• Who controls the model?
• Who receives moderation notices?
• Who can override the system?
• Who is legally responsible for the account or platform?

Classification: Analogical social-media liability authority.

14. Case 8 — Tamiz v United Kingdom

Court: ECtHR
Application No.: 3877/14

Facts

The case concerned allegedly defamatory comments posted on a Google-hosted blog.

The applicant complained about the availability of the comments and the intermediary's response.

Significance

The case addressed the question of when intermediary responsibility may arise in relation to third-party comments and the practical consequences of delay in removal.

AI relevance

AI moderation frequently involves notice-response timing.

For example:

User reports defamatory content → AI identifies it as potentially unlawful → platform delays human review for several days.

The legal relevance of the delay will depend upon the applicable legal regime and facts.

Classification: Analogical intermediary-liability authority.

15. What These Cases Show Collectively

The European case law demonstrates several recurring principles.

Principle 1 — Platform status matters

A provider performing neutral hosting functions is treated differently from a provider that actively participates in unlawful content.

Principle 2 — Knowledge matters

Notice, awareness and obviousness of illegality can affect intermediary liability.

Principle 3 — Content type matters

Hate speech and incitement to violence can be treated differently from vulgar or offensive opinions.

Principle 4 — Context matters

The same words may have different legal significance depending on:

• context;
• target;
• audience;
• public interest;
• purpose.

Principle 5 — Freedom of expression matters

Content moderation can affect Article 10 ECHR rights.

Principle 6 — Victims' rights also matter

Article 8 ECHR can protect:

• reputation;
• private life;
• personal identity;
• dignity.

Thus, courts frequently have to balance Article 8 and Article 10.

16. AI Moderation Creates New Civil-Liability Problems

Traditional human moderation generally involves:

Human → Content → Decision.

AI moderation may involve:

Data → Model → Classification → Automated decision → User consequence.

This creates additional liability questions.

17. Algorithmic False Positives

A false positive occurs when lawful content is classified as unlawful.

Example:

"The government should ban this religion."

The AI incorrectly classifies the statement as terrorist content and permanently suspends the account.

Potential consequences:

• lost income;
• lost business;
• reputational harm;
• contractual losses;
• interference with expression.

The claimant may need to identify a legal cause of action under:

• DSA;
• contract;
• national civil law;
• data protection law;
• other applicable legislation.

18. False Negatives

A false negative occurs when unlawful content is incorrectly classified as lawful.

Example:

AI fails to identify a credible threat against an individual.

The platform leaves the content online.

Potential claims may concern:

• privacy;
• personality rights;
• reputation;
• personal safety;
• negligence under national law.

Again, liability is not automatic.

The claimant would need to establish the relevant duty and causation.

19. Defamation

AI moderation can create two separate defamation problems.

A. Failure to remove defamatory content

The platform leaves the content online.

B. AI-generated moderation label

The platform labels a user:

"fraudster"

or

"extremist"

or

"dangerous individual."

If that label is communicated to others and causes legally recognised harm, national defamation/personality-rights law may become relevant.

The platform's own AI-generated classification can therefore potentially become the subject of a civil claim.

20. Privacy and Personal Data

AI moderation systems process large amounts of data.

Potential data include:

• usernames;
• IP addresses;
• biometric information;
• facial images;
• voice;
• location;
• political opinions;
• religious opinions;
• sexual-orientation information;
• inferred characteristics.

GDPR may therefore apply.

The platform must consider:

• lawful basis;
• purpose limitation;
• data minimisation;
• accuracy;
• retention;
• transparency;
• security;
• automated decision-making rules.

21. Automated Decision-Making

Where moderation produces a decision with legal or similarly significant effects concerning an individual, Article 22 GDPR may become relevant depending on the circumstances.

Examples:

• permanent account suspension;
• termination of a seller account;
• exclusion from a platform;
• employment-related account decisions;
• removal of monetisation.

The CJEU's recent automated-decision cases, including SCHUFA (C-634/21) and CK v Dun & Bradstreet Austria (C-203/22), are important for understanding automated scoring, meaningful information and challengeability.

These cases are not specifically content-moderation cases, but their principles can be relevant where moderation is based on automated individual decisions.

22. DSA Notice-and-Action Mechanism

Article 16 DSA establishes mechanisms through which individuals and entities can notify platforms of allegedly illegal content.

The platform must process qualifying notices under the statutory framework.

This changes the practical liability question.

Previously:

"Did the platform know?"

Under the modern framework:

"When did it receive legally sufficient notice, and what did it do afterwards?"

This makes evidence concerning notice and response extremely important.

23. Statement of Reasons

The DSA imposes transparency obligations concerning restrictions on content.

When a platform restricts content, it generally has to provide the affected recipient with a statement of reasons under Article 17, subject to the Regulation's exceptions.

This creates an important evidence trail.

For AI moderation:

AI says "hate speech."

The platform should be able to explain the relevant basis for the restriction sufficiently to satisfy the applicable legal obligation.

24. Internal Complaint Mechanism

The DSA requires relevant platforms to provide an internal complaint-handling mechanism.

A user may challenge:

• content removal;
• account suspension;
• visibility restrictions;
• monetisation restrictions.

This is particularly important for AI decisions because automated systems can make large numbers of mistakes.

25. Out-of-Court Dispute Settlement

The DSA also provides mechanisms for users to challenge moderation decisions through certified out-of-court dispute-settlement bodies.

Therefore, a user need not necessarily begin with a traditional civil lawsuit.

The legal structure increasingly becomes:

AI decision → internal complaint → out-of-court mechanism → regulator/court → civil damages where available.

26. Very Large Online Platforms

Very Large Online Platforms and Very Large Online Search Engines have additional DSA obligations.

These include systemic-risk assessment and mitigation.

Relevant risks can include:

• dissemination of illegal content;
• fundamental-rights risks;
• effects on civic discourse;
• discrimination;
• manipulation;
• systemic effects arising from platform design.

AI moderation systems can therefore be examined not merely as individual tools but as part of a platform's broader risk-management architecture.

27. AI Bias and Discrimination

AI moderation can disproportionately flag certain:

• languages;
• dialects;
• ethnic expressions;
• religious expressions;
• political discussions;
• minority communities.

For example:

A moderation model trained mainly on English-language content may incorrectly classify an expression in another European language as hate speech.

Potential legal consequences may involve:

• DSA;
• GDPR;
• equality/non-discrimination law;
• contractual law;
• national civil law;
• fundamental rights.

The legal assessment depends on the specific facts and applicable legislation.

28. Lack of Explainability

One of the most important problems is:

Why did the AI remove this content?

A platform may know:

"Probability of hate speech = 97%."

But that does not necessarily explain:

• which words triggered the result;
• which contextual factors were considered;
• whether irony was detected;
• whether quotation was detected;
• whether the system understood the language;
• whether human review occurred.

Explainability can therefore become relevant to:

• DSA statements of reasons;
• GDPR transparency;
• procedural fairness;
• civil evidence.

29. AI Hallucinations

Generative-AI moderation systems can create another problem.

An AI may invent:

"The user posted a threat."

when the user never posted one.

If the platform acts on the fabricated classification and suspends the account, potential issues include:

• negligent system design;
• inadequate verification;
• contractual breach;
• DSA procedural obligations;
• damages.

A key legal question would be whether the platform had reasonable safeguards against such errors.

30. Human Oversight

A platform may reduce legal risk by maintaining meaningful human review.

For example:

Low-risk content

AI can automatically remove obvious spam.

Medium-risk content

AI flags content for review.

High-risk content

Human moderator makes final decision.

Extremely sensitive content

Specialist human review + appeal.

This does not create automatic legal immunity, but it can be relevant to assessing whether the platform complied with applicable duties.

31. Evidence in AI Content-Moderation Litigation

Important evidence can include:

Technical evidence

• model version;
• training data;
• classification threshold;
• confidence score;
• moderation logs;
• model outputs;
• system prompts;
• safety filters.

Platform evidence

• community standards;
• moderation policies;
• internal guidelines;
• notice records;
• human-review records;
• appeal decisions.

Legal evidence

• DSA notices;
• statement of reasons;
• regulatory decisions;
• court orders.

User evidence

• original content;
• screenshots;
• account history;
• financial loss;
• correspondence.

32. Causation

Causation can be difficult.

Suppose:

AI incorrectly removes a journalist's post.

The journalist claims:

"I lost €50,000 because of the removal."

The court may have to examine:

1. Was the removal unlawful?
2. Was the platform legally responsible?
3. Would the post actually have generated €50,000?
4. Did other factors contribute?
5. Was the loss foreseeable?
6. Is the loss recoverable under applicable law?

Therefore:

wrongful moderation ≠ automatically €50,000 damages.

33. Damages

Potential damages vary considerably between legal systems.

Possible categories include:

Economic loss

• lost advertising;
• lost sales;
• lost subscriptions;
• lost contracts;
• lost platform income.

Reputational loss

• damage to professional reputation;
• loss of commercial goodwill.

Privacy/personality harm

National law may permit compensation for certain non-material harms.

Contractual loss

A platform user may have a contractual claim where platform obligations have been breached.

Data-protection damages

GDPR Article 82 may provide a separate route where the statutory conditions are met.

34. Defences

Platforms may rely on several arguments.

1. Safe harbour

The platform may argue that the DSA conditions for exemption from liability are satisfied.

2. No actual knowledge

The platform did not have the legally relevant knowledge.

3. Prompt removal

After receiving sufficient notice, it acted expeditiously.

4. Lawful moderation

The content violated valid platform rules.

5. No causation

The claimed loss resulted from another cause.

6. No recoverable damage

The claimant cannot establish legally compensable loss.

7. User contractual obligations

The user agreed to applicable platform rules.

These defences depend upon the applicable law and facts.

35. Platform Terms and Conditions

A user's contract with a platform can be significant.

Terms may regulate:

• prohibited content;
• suspension;
• termination;
• appeals;
• moderation;
• monetisation;
• account restrictions.

However, contractual terms do not necessarily allow a platform to disregard mandatory EU or national law.

A contractual clause cannot simply eliminate statutory rights that cannot legally be waived.

36. AI Moderation and Fundamental Rights

Two ECHR rights are particularly important.

Article 8

Protection of:

• private life;
• reputation;
• personal identity.

Article 10

Freedom of:

• expression;
• information.

Content moderation can therefore create competing interests:

Victim's reputation/privacy

versus

speaker's freedom of expression

versus

platform's freedom to operate its service.

Cases such as Delfi and MTE/Index.hu demonstrate how the ECtHR approaches this balancing exercise.

37. Important Difference Between AI Moderation and AI Generation

These should not be confused.

AI moderation

AI decides whether someone else's content should remain online.

Generative AI

AI itself creates the content.

AI-assisted moderation + generative AI

AI may:

1. generate content;
2. classify it;
3. determine whether it violates rules;
4. remove or modify it.

Each stage can create different liability questions.

38. The "Active Role" Problem

European intermediary jurisprudence repeatedly asks whether the platform has remained a neutral intermediary or has taken an active role.

This concept originated under the E-Commerce Directive and remains important background to modern DSA interpretation.

Google France, L'Oréal/eBay, and later platform cases demonstrate this development.

For AI systems, the question becomes particularly difficult because algorithms inherently:

• rank;
• filter;
• classify;
• personalise.

But algorithmic functionality alone should not be equated automatically with legal responsibility for every piece of user content.

39. AI Content Moderation and DSA Compensation

One of the most important provisions for civil-law analysis is:

Article 54 DSA

A recipient of an intermediary service has a right to seek compensation, under Union and national law, for damage or loss suffered because of an intermediary provider's infringement of its DSA obligations.

This potentially provides an important bridge between:

DSA regulatory obligations

and

private civil claims.

However, Article 54 does not create a simple automatic damages formula. The claimant still has to satisfy the applicable requirements concerning:

• infringement;
• damage;
• causation;
• recoverability;
• applicable national law.

40. Case-Law Revision Table

41. Key Legal Principles

Principle 1

AI moderation does not automatically make a platform liable for user-generated content.

Principle 2

The DSA establishes important conditional intermediary-liability protections.

Principle 3

Actual knowledge and legally sufficient notice can be highly important.

Principle 4

After obtaining the relevant knowledge, the provider may have to act expeditiously.

Principle 5

EU law does not impose a general obligation to monitor all user content.

Principle 6

A platform can nevertheless be required to remove specific unlawful material.

Principle 7

AI classification does not eliminate the need to distinguish unlawful content from merely offensive or controversial expression.

Principle 8

Freedom of expression under Article 10 ECHR remains relevant to moderation disputes.

Principle 9

Protection of reputation and private life under Article 8 ECHR can justify restrictions on harmful content.

Principle 10

Article 54 DSA provides an express route to seek compensation for damage caused by a provider's infringement of DSA obligations, subject to Union and national law.

42. Exam-Style Conclusion

AI content-moderation liability in Europe is governed by a developing combination of EU intermediary law, the Digital Services Act, data-protection law, intellectual-property law, national civil liability and fundamental-rights jurisprudence.

The historical European case law provides the foundation. Delfi AS v Estonia demonstrates circumstances in which liability for third-party comments can be compatible with Article 10, while Magyar Tartalomszolgáltatók Egyesülete and Index.hu v Hungary illustrates the importance of distinguishing clearly unlawful material from merely offensive comments. Glawischnig-Piesczek establishes important principles concerning court-ordered removal and equivalent unlawful content. L'Oréal v eBay, Google France, and YouTube/Cyando develop the distinction between neutral intermediary activity and a more active role. Sanchez v France adds an important perspective concerning responsibility for third-party comments on controlled social-media spaces.

The DSA materially changes the modern framework by introducing detailed notice-and-action, statement-of-reasons, complaint-handling, transparency and systemic-risk obligations while retaining conditional hosting liability protection. Most importantly for civil litigation, Article 54 expressly recognises compensation claims for damage or loss resulting from an intermediary provider's infringement of its DSA obligations.

The central legal sequence can therefore be remembered as:

AI detection → AI classification → moderation decision → notice/knowledge → platform response → legal duty → unlawful conduct → causation → damage → civil remedy.

For examination purposes, the most important authorities are:

Delfi → MTE/Index.hu → Glawischnig-Piesczek → L'Oréal/eBay → Google France → YouTube/Cyando → Sanchez → Tamiz.

The principal modern question is no longer simply "Is the platform liable for user content?" It is more precisely:

What legal role did the platform and its AI moderation system play, what statutory or civil duty applied, what did the provider know or reasonably have to address, and can the claimant establish legally recoverable damage caused by the resulting moderation or non-moderation?