Algorithmic Propaganda Liability in Europe

1. Meaning and Scope

Algorithmic propaganda liability concerns legal responsibility arising when algorithms, artificial intelligence, recommender systems, automated advertising tools, bots, ranking systems, or profiling technologies are used to create, target, amplify, suppress, personalise, or distribute propaganda in a manner that causes legally cognisable harm.

The concept is particularly relevant to:

political micro-targeting;

automated amplification of political content;

recommender-system manipulation;

AI-generated propaganda;

coordinated bot networks;

disinformation campaigns;

manipulation of electoral information;

discriminatory political targeting;

suppression or down-ranking of opposing viewpoints;

foreign-influence operations;

personalised political advertising;

automated manipulation of vulnerable groups;

deepfake political content.

There is no single European cause of action called “algorithmic propaganda liability.” Liability is instead constructed from several bodies of law, including:

freedom of expression and information;

privacy and data protection;

electoral law;

political advertising regulation;

consumer and unfair-commercial-practice law;

platform/intermediary regulation;

competition law;

discrimination law;

tort/delict and civil liability;

administrative/public law;

national criminal law in serious cases;

European human-rights law.

A central legal question is:

Who is legally responsible when an algorithm does not merely host propaganda, but materially amplifies, targets, recommends, or optimises its dissemination?

2. Principal Forms of Algorithmic Propaganda

A. Algorithmic amplification

A platform's recommender system identifies highly engaging political material and repeatedly recommends it to users.

Potential issues include:

deliberate amplification of inflammatory material;

excessive optimisation for engagement;

amplification of demonstrably false information;

failure to manage systemic risks;

manipulation of elections.

B. Political micro-targeting

Algorithms analyse personal information and classify individuals into political audiences.

For example:

Users predicted to be economically anxious receive one political message, while users predicted to be socially conservative receive another.

The principal legal questions involve:

lawful basis for processing;

profiling;

transparency;

special-category data;

political advertising rules;

discrimination;

meaningful consent;

manipulation.

C. Automated propaganda generation

Generative AI may produce:

fake political speeches;

synthetic images;

deepfake videos;

fabricated news articles;

automated social-media posts;

fake comments;

large-scale bot content.

Liability may arise where the content is defamatory, deceptive, unlawful, discriminatory, fraudulent, or otherwise prohibited.

D. Algorithmic suppression

Algorithms may deliberately or improperly:

down-rank political opponents;

suppress particular viewpoints;

demonetise political speakers;

suspend accounts;

restrict distribution.

This creates a difficult balance between:

freedom of expression + platform autonomy + protection against unlawful content.

E. Automated political advertising

Advertising algorithms can determine:

who receives an advertisement;

when it appears;

how frequently it appears;

which message is displayed;

which users are excluded.

This creates particular concerns where political advertising is opaque or excessively personalised.

3. European Legal Framework

A. Article 10 ECHR

Article 10 protects:

freedom of expression;

freedom to receive information;

freedom to impart information.

It is fundamental to propaganda disputes because political expression receives especially strong protection.

However, Article 10 does not protect every form of manipulation or unlawful conduct. Restrictions can be justified where prescribed by law and necessary in a democratic society.

B. Article 11 EU Charter

Article 11 of the Charter protects freedom of expression and information.

It is particularly relevant where EU legislation regulates:

online platforms;

political advertising;

content moderation;

digital services;

data processing.

C. GDPR

Algorithmic propaganda frequently depends upon profiling.

Relevant provisions include:

Articles 5 and 6 — data-processing principles and lawful bases;

Articles 9 and 10 — sensitive/special-category and criminal data;

Articles 12–15 — transparency and information;

Article 21 — objection;

Article 22 — automated individual decision-making;

Article 25 — data protection by design and default;

Article 35 — data protection impact assessments;

Article 82 — compensation.

Political opinions constitute special-category personal data under Article 9 GDPR.

Consequently, political profiling can create particularly serious legal issues.

4. Digital Services Regulation

The EU's Digital Services Act framework is highly significant for algorithmic propaganda.

Large online platforms and search engines have enhanced responsibilities concerning systemic risks, including risks associated with:

manipulation;

electoral processes;

fundamental rights;

dissemination of illegal content;

recommender systems;

advertising transparency.

The important legal development is that responsibility can extend beyond merely asking:

“Did the platform publish the propaganda?”

toward:

“Did the platform's system design, recommender architecture, targeting system or risk-management practices materially contribute to systemic dissemination or manipulation?”

This does not, however, mean that platforms become automatically liable for every unlawful statement posted by users.

5. Political Advertising and Algorithmic Targeting

Political advertising creates a special category of risk because algorithmic systems can exploit highly detailed information about voters.

A problematic system might combine:

browsing behaviour;

location;

political preferences;

social connections;

purchasing behaviour;

psychological characteristics;

inferred political opinions.

The legal question becomes whether the use of such information constitutes lawful political communication or impermissible manipulation/profiling.

6. Main Elements of an Algorithmic Propaganda Liability Claim

A claimant normally needs to establish some combination of:

1. Algorithmic activity

There must be an identifiable system:

recommender;

advertising algorithm;

profiling system;

bot system;

generative AI;

ranking system.

2. Unlawful or defective conduct

Examples:

unlawful profiling;

discriminatory targeting;

unlawful processing;

deceptive advertising;

failure to remove illegal content;

unlawful suppression;

inadequate safeguards;

negligent system design.

3. Causation

The claimant must establish a connection between:

algorithmic activity → propaganda dissemination/manipulation → legally relevant harm.

This is often the most difficult element.

4. Recognised harm

Potential harm includes:

financial loss;

reputational injury;

privacy infringement;

discrimination;

psychological injury;

electoral-rights interference;

political participation impairment;

infringement of freedom of expression;

unlawful data processing.

7. Important European Case Law

Because algorithmic propaganda is a relatively new legal category, European courts have not yet produced six major judgments specifically titled “algorithmic propaganda liability.” The following cases therefore combine direct digital-platform/data cases with highly relevant freedom-of-expression, surveillance and intermediary-liability authorities.

Case 1 — Delfi AS v Estonia

ECtHR Grand Chamber, Application No. 64569/09, 16 June 2015

Facts

Delfi operated a major online news portal. Users posted offensive comments beneath an article, including serious threats and hateful statements.

The Estonian courts imposed liability on the portal.

Principle

The European Court of Human Rights accepted that an internet intermediary could, in particular circumstances, bear responsibility for unlawful user-generated content.

The Court examined:

nature of the comments;

platform's role;

measures available to the platform;

identity of the perpetrators;

consequences for the victims.

Relevance to algorithmic propaganda

This is important for platforms whose technology goes beyond passive hosting.

If a platform:

recommends;

ranks;

amplifies;

targets;

monetises

content, the question of its responsibility may become more significant.

However, Delfi should not be interpreted as establishing automatic liability for platforms whenever unlawful material appears on their services.

Case 2 — MTE and Index.hu Zrt v Hungary

ECtHR, Applications Nos. 22947/13 and 23607/13, 2 February 2016

Principle

The Court distinguished the circumstances from those in Delfi and emphasised the importance of:

context;

nature of the comments;

seriousness of the harm;

platform's role;

proportionality of liability.

Importance

This case prevents an excessively broad approach to intermediary liability.

For algorithmic propaganda, it demonstrates that liability requires a context-sensitive proportionality analysis.

A platform should not automatically be treated as the author of every user statement merely because its algorithm distributes content.

Case 3 — Glawischnig-Piesczek v Facebook Ireland

CJEU, Case C-18/18, 3 October 2019

Facts

A politician sought removal of defamatory material published on Facebook.

The CJEU considered whether an injunction could require removal not only of the specific unlawful statement but also identical or, in appropriate circumstances, equivalent material.

Principle

The Court accepted that national courts can, subject to the relevant legal conditions, impose orders requiring removal of identical or equivalent unlawful content.

Relevance

This is highly relevant to automated content moderation.

An AI system can potentially be used to:

detect duplicates;

identify equivalent unlawful content;

automate removal;

prevent repeated dissemination.

Liability significance

The case supports the proposition that technology can be part of the legal mechanism for controlling unlawful online content.

But it does not establish that every recommender algorithm must eliminate all political misinformation.

Case 4 — Google Spain SL, Google Inc. v AEPD and Mario Costeja González

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

Principle

The CJEU recognised significant responsibilities of search engines in relation to personal-data processing and the presentation of search results.

The case established the famous European “right to be forgotten”/de-referencing framework in appropriate circumstances.

Relevance

Algorithmic propaganda may involve the repeated algorithmic presentation of information concerning individuals.

For example, an automated political campaign might repeatedly cause searches for an individual to surface:

false accusations;

obsolete information;

personal information;

misleading political claims.

Google Spain demonstrates that an algorithmic intermediary can have legally significant responsibilities arising from the effects of its processing and presentation system.

Case 5 — Google LLC v CNIL

CJEU, Case C-507/17, 24 September 2019

Principle

The CJEU addressed the territorial scope of de-referencing obligations.

It held, broadly, that EU law did not require a search engine to carry out worldwide de-referencing in every case, while EU-wide de-referencing could be required under the applicable framework.

Relevance

Algorithmic propaganda is inherently transnational.

A political campaign can disseminate the same material:

in France;

Germany;

Italy;

Spain;

Poland;

simultaneously.

The case illustrates that online information regulation must confront the territorial scope of algorithmic dissemination.

Case 6 — Satakunnan Markkinapörssi Oy and Satamedia Oy v Finland

ECtHR Grand Chamber, Application No. 931/13, 27 June 2017

Facts

The case concerned large-scale processing and publication of personal taxation information.

Principle

The Court balanced:

freedom of expression;

public-interest journalism;

privacy;

personal-data protection.

Relevance

Algorithmic propaganda often involves enormous quantities of personal information.

The case is important because it demonstrates that data processing and expression rights cannot be considered separately.

A propagandistic campaign may claim political-expression protection while simultaneously engaging in unlawful or disproportionate personal-data processing.

Case 7 — Big Brother Watch and Others v United Kingdom

ECtHR Grand Chamber, Applications Nos. 58170/13, 62322/14 and 24960/15, 25 May 2021

Principle

The Court examined large-scale interception and surveillance systems and emphasised the importance of safeguards against abuse.

The judgment considered:

necessity;

proportionality;

safeguards;

authorisation;

oversight;

selection of communications.

Relevance

Algorithmic propaganda can depend upon large-scale data collection.

For example:

surveillance/data collection → political profiling → algorithmic classification → targeted propaganda.

Big Brother Watch therefore provides an important framework for examining the safeguards surrounding large-scale technological systems.

Case 8 — S. and Marper v United Kingdom

ECtHR Grand Chamber, Applications Nos. 30562/04 and 30566/04, 4 December 2008

Principle

The Court found that indiscriminate retention of biometric information raised serious Article 8 concerns.

Relevance

Although not a propaganda case, it establishes an important technological principle:

The fact that technology can collect and process information does not mean that unlimited collection and retention is legally permissible.

This becomes relevant where political-profiling systems build extensive datasets used for targeted propaganda.

Case 9 — Magyar Helsinki Bizottság v Hungary

ECtHR Grand Chamber, Application No. 18030/11, 8 November 2016

Principle

The Court recognised that Article 10 can, in appropriate circumstances, protect access to information held by public authorities.

Relevance

Algorithmic propaganda disputes may involve government-generated information, automated public communication, or algorithmic information environments.

The case supports the broader principle that access to information is an important component of democratic participation.

Case 10 — Glukhin v Russia

ECtHR, Application No. 11519/20, 4 July 2023

Facts

The case involved facial-recognition technology used by Russian authorities to identify a person.

Principle

The Court found serious Article 8 implications arising from the use of facial-recognition technology in the circumstances of the case.

Relevance

Political propaganda can be combined with:

facial recognition;

biometric identification;

political profiling;

surveillance;

targeted messaging.

Glukhin demonstrates that technologically sophisticated identification systems remain subject to privacy and proportionality requirements.

8. Case-Law Comparison

9. Liability of Different Actors

A. AI developer

The developer may face liability where it:

deliberately designs a manipulation system;

knowingly enables unlawful targeting;

negligently fails to implement foreseeable safeguards;

violates contractual obligations;

provides a defective product.

However, merely creating a general-purpose AI model does not automatically make the developer liable for every subsequent political misuse.

B. Platform provider

The platform may face regulatory or civil exposure where it:

deliberately amplifies unlawful propaganda;

performs unlawful profiling;

fails applicable risk-management obligations;

unlawfully processes personal data;

ignores legally relevant risks;

fails to implement required safeguards.

C. Political campaign

Political organisations may be responsible where they:

unlawfully process personal information;

use prohibited targeting techniques;

commission deceptive advertisements;

disseminate defamatory material;

manipulate voters through unlawful means.

D. Data broker

A data broker may be responsible for unlawful collection or disclosure of:

political preferences;

behavioural profiles;

demographic characteristics;

sensitive personal information.

E. Government authority

A public authority can incur responsibility when it:

unlawfully deploys propaganda technology;

conducts disproportionate surveillance;

manipulates public information;

unlawfully profiles citizens;

violates freedom of expression;

fails to provide effective remedies.

10. Algorithmic Amplification and Causation

One of the most difficult questions is proving:

Would the claimant have suffered the harm without algorithmic amplification?

Consider:

Political post → recommender system → 2,000 users → 2 million users → targeted dissemination → reputational/electoral harm.

The claimant must distinguish between:

Ordinary publication

The defendant merely made content available.

Algorithmic amplification

The defendant's system actively:

selected;

ranked;

recommended;

targeted;

repeated;

optimised

the content.

The second situation can provide stronger evidence concerning causation and foreseeability, depending on the applicable legal regime.

11. Algorithmic Propaganda and Freedom of Expression

This is one of the most difficult areas.

European law generally cannot simply say:

“Propaganda is harmful, therefore remove it.”

Some propaganda is political expression protected by Article 10 ECHR and Article 11 Charter.

The legal analysis should instead ask:

Is the material protected expression?

Is it unlawful?

Is it political expression?

Is the restriction prescribed by law?

What legitimate objective is pursued?

Is the restriction necessary?

Is it proportionate?

Is there a less restrictive alternative?

Was the algorithmic intervention transparent?

Is there an effective appeal mechanism?

12. Political Manipulation vs Political Persuasion

This distinction is extremely important.

Legitimate political persuasion

Examples:

political advertisement;

campaign speech;

targeted communication based on lawful information;

political criticism.

Potentially unlawful manipulation

Examples:

exploitation of sensitive personal data;

deceptive synthetic political material;

discriminatory political targeting;

covert foreign influence;

fraudulent impersonation;

unlawful surveillance-based targeting.

Therefore:

Algorithmic personalisation is not automatically unlawful propaganda.

The legality depends on the method, data, content, purpose, safeguards and applicable law.

13. Discrimination and Algorithmic Propaganda

Propaganda algorithms can produce discriminatory outcomes.

For example, an algorithm might:

exclude certain ethnic groups from political information;

target particular religious communities;

suppress advertisements for certain populations;

exploit disability or economic vulnerability;

send different political claims to different protected groups.

Relevant European equality principles can therefore interact with data-protection and expression law.

A platform cannot necessarily avoid discrimination law simply because the discriminatory outcome was produced by an algorithm.

14. Privacy and Political Profiling

Political opinions receive particularly strong protection under European data-protection law.

An algorithm that predicts:

“This person is probably a supporter of Party X”

may itself create legally sensitive personal information.

The legal analysis can therefore include:

source of data;

accuracy;

lawful basis;

special-category data;

profiling;

transparency;

retention;

sharing;

automated decision-making;

objection rights.

15. AI-Generated Propaganda

Generative AI substantially increases the scale of propaganda.

A single operator can potentially produce:

thousands of political messages;

multiple languages;

synthetic photographs;

fake videos;

fake interviews;

fabricated quotations;

automated social-media responses.

This creates a shift from:

human-scale propaganda → machine-scale propaganda.

The principal legal concern is not merely whether AI generated the material, but whether the deployment creates an unlawful mechanism of deception, manipulation, privacy infringement or other legally cognisable harm.

16. Defences

A defendant may rely upon:

Freedom of expression

Political expression receives particularly strong protection.

Lack of causation

The defendant may argue that the algorithm did not materially cause the alleged harm.

Lack of knowledge

A platform may argue it had no sufficient knowledge of unlawful content.

Intermediary protections

Applicable EU and national intermediary rules may limit liability in appropriate circumstances.

User responsibility

The platform may argue that the individual user, rather than the platform, created the unlawful content.

Legitimate algorithmic operation

Recommendation or ranking may have been based on legitimate technical criteria rather than an intention to manipulate.

Proportionality

A defendant may argue that imposing liability would excessively interfere with expression, innovation or business freedom.

17. Remedies

Depending on the applicable legal regime, possible remedies include:

Individual remedies

compensation;

correction;

deletion;

de-referencing;

cessation of unlawful processing;

access to information;

human review;

rectification.

Injunctive remedies

Courts may order:

removal of unlawful content;

cessation of particular targeting;

suspension of unlawful profiling;

restriction of data processing.

Regulatory remedies

Authorities may impose:

administrative fines;

compliance orders;

risk-management requirements;

transparency obligations;

corrective measures.

Systemic remedies

In serious cases:

independent audits;

recommender-system modifications;

risk assessments;

monitoring;

transparency reporting;

improved content-moderation systems.

18. Evidentiary Issues

Algorithmic propaganda cases are frequently difficult because the claimant may not know why particular content was shown.

Important evidence includes:

algorithmic ranking records;

recommender-system documentation;

advertising records;

targeting criteria;

profiling information;

model documentation;

content-moderation logs;

risk assessments;

internal communications;

platform policies;

audit reports;

engagement statistics;

A/B testing;

user-exposure data;

political-advertising records.

The claimant may therefore need disclosure or regulatory access to information that is largely controlled by the platform.

19. Algorithmic Propaganda and the AI Act

The EU AI Act is relevant to the governance of AI systems, particularly where AI is used in contexts involving:

manipulation;

prohibited practices;

biometric categorisation;

high-risk decision-making;

transparency;

human oversight.

An important distinction is:

Regulatory non-compliance and civil damages are not automatically the same thing.

A breach of an AI regulatory requirement may be powerful evidence of unlawful conduct or failure to meet a statutory duty, but the claimant may still need to establish the requirements of the particular damages action.

20. Practical Liability Test

A European court analysing an algorithmic propaganda dispute can effectively be approached through the following sequence:

1. Identify the content

Was it political propaganda, ordinary opinion, commercial advertising, misinformation, hate speech, defamatory material or something else?

↓

2. Identify the technology

Was AI used for:

creation;

targeting;

recommendation;

ranking;

amplification;

suppression?

↓

3. Identify the actor

Who controlled the relevant system?

↓

4. Identify the legal duty

GDPR?
ECHR?
Charter?
DSA?
Political-advertising law?
Equality law?
Consumer law?
Tort law?

↓

5. Examine the data

Was political or other sensitive personal data processed?

↓

6. Examine algorithmic behaviour

Did the system merely host the material or actively amplify it?

↓

7. Examine safeguards

Were appropriate controls, transparency and risk-management measures implemented?

↓

8. Establish causation

Did the algorithm materially contribute to the alleged harm?

↓

9. Establish damage/right infringement

Was there:

financial loss;

privacy harm;

discrimination;

reputational injury;

electoral interference;

freedom-of-expression infringement?

↓

10. Determine remedy

Compensation, injunction, deletion, de-referencing, correction, regulatory sanctions or systemic measures.

21. Key Legal Principle

The emerging European position can be summarised as follows:

An algorithm is not itself a legal shield.

But the opposite proposition is equally important:

The fact that an algorithm amplified propaganda does not automatically make its operator legally liable.

Liability normally depends upon the combination of:

algorithmic activity + applicable legal duty + unlawful conduct/defect + causation + legally recognised harm, together with any applicable intermediary, expression, proportionality or statutory defences.

22. Conclusion

Algorithmic propaganda liability in Europe is an emerging, cross-disciplinary field rather than a standalone cause of action. Its most important legal tensions are between democratic free expression and protection against technological manipulation.

The strongest European authorities currently provide different pieces of the legal framework:

Delfi — intermediary responsibility;

MTE and Index.hu — proportionality in intermediary liability;

Glawischnig-Piesczek — removal of unlawful online content;

Google Spain — responsibility surrounding algorithmic processing and presentation;

Google v CNIL — territorial limits of online information remedies;

Satakunnan — privacy/data processing versus expression;

Big Brother Watch — technological surveillance safeguards;

S. and Marper — limits on technologically enabled data retention;

Magyar Helsinki Bizottság — information and democratic participation;

Glukhin — technologically enhanced identification and privacy.

Accordingly, a future European algorithmic propaganda liability claim is most likely to succeed where the claimant can demonstrate a clear chain:

unlawful data collection/profiling or unlawful content → algorithmic targeting/amplification → foreseeable exposure or manipulation → legally relevant interference or damage → causation → responsibility of the actor controlling the relevant system.

The most difficult issues will remain causation, attribution, platform responsibility, political-expression protection, algorithmic opacity, and proving that automated amplification materially contributed to the harm.