Civil Law and Data Training Contribution Labour Compensation Claims in Europe

1. Introduction

“Data training contribution labour compensation claims” is an emerging legal issue concerning situations where an employee's work, personal data, professional knowledge, communications, performance records, creative output, or other work-generated information is used to train, test, fine-tune, evaluate, or improve an AI or machine-learning system.

A typical dispute may look like this:

An employee performs work for an employer → the employer collects the employee's work-related data → the data is incorporated into an AI training dataset → the AI system generates commercial value → the employee argues that the contribution should generate additional remuneration or compensation.

This area is not yet governed by a single European rule granting employees a general right to royalties for their data being used in AI training. The legal analysis currently has to be constructed from several bodies of law, including:

employment law;

remuneration law;

working-time law;

GDPR/data-protection law;

copyright and related rights;

trade-secret/confidentiality law;

contract law;

equality law;

collective labour law;

EU fundamental freedoms;

national civil-liability rules.

There is, however, useful European case law concerning remuneration for training, compensation connected with employment, employee-generated information, worker data and employer control of employee information.

A particularly important starting point is Bötel (C-360/90), where the CJEU held that compensation connected with employment-related training can fall within the concept of “pay.” (InfoCuria)

2. What Is a “Data Training Contribution”?

A data-training contribution can arise when an employee supplies information that becomes an input into an AI system.

Examples include:

Employee performance data

productivity records;

performance scores;

sales records;

customer interactions;

work schedules;

error rates.

Professional knowledge

technical solutions;

workflows;

professional classifications;

internal expertise;

problem-solving methods.

Employee-created content

reports;

photographs;

videos;

designs;

software code;

manuals;

written materials.

Personal employee data

communications;

behavioural information;

biometric information;

location;

performance information.

AI interaction data

Employees may interact with an employer's AI system and thereby generate:

prompts;

corrections;

feedback;

evaluations;

examples;

annotations;

model outputs.

That information may subsequently be used for training or improving the system.

3. Important Preliminary Point: Data Is Not Automatically “Wages”

An employee cannot generally argue:

“My employer used my personal data for AI training, therefore I automatically own part of the AI company's profits.”

European law does not presently establish such a general rule.

A compensation claim normally needs an identifiable legal basis, such as:

employment contract;

collective agreement;

statutory remuneration right;

copyright or related right;

database right;

privacy/data-protection damages;

breach of confidentiality;

unjust enrichment under national law;

discrimination/equal-pay rules;

damages caused by unlawful processing.

Therefore:

Data contribution ≠ automatic ownership of AI profits.

4. Case Law 1 — Arbeiterwohlfahrt der Stadt Berlin v Bötel, C-360/90

CJEU, 4 June 1992

This is one of the most useful European authorities for the subject.

The case concerned part-time employees who were members of staff councils and attended training courses necessary for performing their staff-council functions.

The CJEU considered whether compensation for attending those courses constituted “pay.”

It held that the concept of pay is broad and includes consideration provided to a worker, directly or indirectly, in respect of employment. (InfoCuria)

The Court also found that limiting compensation for part-time workers could raise equal-pay problems where the arrangement disproportionately affected women. (InfoCuria)

Importance for AI/data training

The principle is highly relevant:

Where an employee performs an activity connected sufficiently closely with employment and the law or employment arrangement provides compensation for that activity, the compensation can fall within the concept of employment remuneration.

This can support arguments concerning:

compulsory AI training;

employee data labelling;

model evaluation;

annotation work;

AI feedback work;

training performed outside ordinary working hours.

5. Case Law 2 — BX v Unitatea Administrativ-Teritorială D., C-909/19

CJEU, 28 October 2021

This case directly concerned mandatory vocational training requested by an employer.

The issue was whether training undertaken outside the worker's normal workplace and beyond ordinary working hours constituted “working time” under Directive 2003/88.

The CJEU treated the mandatory training as working time where the worker was required to attend it at the employer's request and was subject to the employer's instructions. (InfoCuria)

Importance

This principle is extremely useful for modern AI-training work.

Suppose an employer tells employees:

“Spend three hours every evening correcting AI outputs so that our model can learn from your responses.”

The question becomes whether that activity is:

ordinary employment;

working time;

overtime;

a separate paid assignment.

If the activity is compulsory and performed for the employer's benefit, the working-time analysis becomes highly relevant.

6. Case Law 3 — Pantuso and Others, C-616/16 and C-617/16

CJEU, 24 January 2018

This case concerned specialist medical training in Italy.

The CJEU interpreted EU legislation concerning specialist medical training and held that qualifying specialist training was subject to appropriate remuneration within the relevant EU framework. (InfoCuria)

The Court also held that the remuneration obligation was sufficiently precise and unconditional in the circumstances covered by the applicable directives. (InfoCuria)

Importance for data-training labour

Pantuso establishes an important principle:

Where EU legislation creates a remuneration entitlement connected with a worker's training activity, the worker can potentially seek monetary compensation rather than merely recognition of the training.

This does not create a general AI-data royalty.

Instead, it provides an analogy for situations where legislation, contract or collective bargaining expressly makes an AI/data contribution a compensable employment activity.

7. Case Law 4 — Presidenza del Consiglio dei Ministri and Others, C-590/20

CJEU, 3 March 2022

This case also concerned specialist medical training and remuneration.

The Court confirmed that qualifying specialist training continuing after the relevant EU-law deadline had to receive appropriate remuneration for the relevant period. It further recognised that failure by a Member State to transpose the applicable remuneration obligation could potentially give rise to State liability for damages, subject to the EU conditions for State liability. (InfoCuria)

The CJEU reiterated the three conditions for State liability:

the EU rule must confer rights on individuals;

the breach must be sufficiently serious;

there must be a direct causal link between the breach and the damage.

(InfoCuria)

Importance

This demonstrates that a failure to provide a legally required employment-related remuneration can potentially become a damages claim, rather than merely an administrative issue.

8. Case Law 5 — K.K.O. 2019:28, Finland

Finnish Supreme Court, 22 March 2019

This is particularly relevant to training-cost clauses in employment contracts.

An employee had agreed to reimburse training costs after leaving employment and joining a competitor.

The Finnish Supreme Court held that employers and employees could, in principle, agree on reimbursement of training costs, but such arrangements could not require the employee to reimburse training that the employer was legally required to provide. The terms also had to be reasonable. (curia)

Importance for data-training disputes

The case shows that employment-related training arrangements cannot simply be drafted in whatever form the employer chooses.

A similar question could arise with AI training:

Can an employer require an employee to undertake extensive AI-data annotation and then treat the activity as part of ordinary unpaid employment?

The answer will depend upon:

applicable employment legislation;

contract;

collective agreement;

working-time rules;

nature of the task;

employer's instructions;

remuneration arrangements.

9. Case Law 6 — S v État belge, C-... / Worker Traineeship Jurisprudence

European law also recognises that a person undergoing vocational training can still possess worker status where they perform genuine work.

The CJEU has repeatedly examined whether trainees performing real work can fall within EU worker protection. In the relevant jurisprudence, the Court has stressed that factors such as the legal label attached to the relationship or the source of remuneration do not necessarily determine worker status. (InfoCuria)

Importance for AI data work

This is significant where companies classify workers as:

trainees;

contributors;

annotators;

freelancers;

contractors;

“AI testers”;

volunteers.

The actual substance of the relationship can matter.

If the person is genuinely performing productive work under the direction of another party, the relationship may attract employment protections even if the employer uses a different contractual label.

10. Case Law 7 — Meta Platforms and Others v Bundeskartellamt, C-252/21

CJEU, 4 July 2023

This is not an employment-remuneration case, but it is highly relevant to the data side of the problem.

Meta combined personal information from Facebook with information obtained from other Meta services and external websites/apps.

The CJEU held that a competition authority may consider compliance with GDPR requirements when examining abuse of dominance, while respecting the division of powers between competition and data-protection authorities. (tuc.org.uk)

Relevance to employees

The case supports the proposition that:

The commercial exploitation of personal data can have consequences under multiple legal regimes simultaneously.

Thus an employee's data may potentially raise:

Employment law + GDPR + contract law + civil liability

at the same time.

11. Case Law 8 — Österreichische Post, C-300/21

CJEU, 4 May 2023

The CJEU considered compensation under Article 82 GDPR.

It held that compensation requires:

an infringement of the GDPR;

damage;

a causal connection between the infringement and the damage.

However, EU law does not impose a general requirement that non-material damage must reach a particular seriousness threshold before compensation becomes available.

Relevance to employees

Suppose an employer uses employee information to train an AI system unlawfully.

The employee could potentially pursue a data-protection damages claim if:

unlawful processing → actual damage → causal connection

can be established.

But this is different from a claim for:

“I contributed data to the AI model, therefore I deserve a percentage of its profits.”

The GDPR damages route compensates legally recognised damage; it does not automatically create an AI royalty.

12. Case Law 9 — Österreichische Post, C-154/21

CJEU, 12 January 2023

The CJEU held that where personal data have been or will be disclosed to recipients, the controller may have to provide the actual identity of recipients, where those recipients can be identified, rather than merely categories of recipients.

Importance for workplace AI

An employee could potentially need information concerning:

which AI provider received employee data;

which processor processed it;

whether data was transferred to another company;

whether data was supplied for model training;

who the relevant recipients were.

This can become important evidence in a later compensation action.

13. Case Law 10 — F.F. v Österreichische Datenschutzbehörde and CRIF, C-487/21

CJEU, 4 May 2023

The CJEU interpreted the right of access under Article 15 GDPR.

It emphasised that the controller must provide a sufficiently faithful and intelligible reproduction of the personal data so that the data subject can effectively exercise GDPR rights.

Relevance

In an employee AI-training dispute, access rights can potentially help establish:

what data was collected;

what information was processed;

whether performance data was used;

what information was transferred;

whether profiling occurred;

whether AI systems processed the information.

This can be essential before bringing a damages claim.

14. Is There Currently a European Right to AI Training Royalties?

Generally, no.

There is currently no general EU rule saying:

“Every employee whose data contributes to AI training must receive a percentage of the AI system's revenue.”

That distinction is crucial.

Possible legal claims are instead divided into separate categories.

15. Employee Data vs Employee Work

This distinction is extremely important.

Situation A — Personal data

Example:

Employee's attendance record is used to train an HR algorithm.

The main issues may be:

GDPR;

employment privacy;

proportionality;

lawful basis;

transparency.

Situation B — Creative work

Example:

Employee creates illustrations that are incorporated into an AI training dataset.

The analysis may involve:

copyright ownership;

employment-created works;

contractual allocation of rights;

licensing;

remuneration.

Situation C — Work activity

Example:

Employee spends 20 hours annotating AI outputs.

The strongest issue may be:

whether those 20 hours constitute working time;

whether remuneration/overtime is payable.

Situation D — Commercial knowledge

Example:

Employee's expert corrections are incorporated into a commercial AI model.

Potential issues include:

confidentiality;

trade secrets;

contract;

employment duties;

intellectual property.

16. Mandatory AI Data-Labelling Work

Suppose an employer requires 100 employees to classify:

images;

text;

customer complaints;

AI outputs;

product descriptions.

The employees perform this activity during normal employment.

Normally the first legal question is:

Is this simply work for which the employee is already contractually paid?

If yes, a separate royalty does not automatically arise.

But if employees are required to perform substantial additional work:

ordinary work + additional AI annotation outside normal hours

then working-time and remuneration issues become much stronger.

The C-909/19 principle concerning mandatory vocational training is particularly relevant to the classification of compulsory employer-directed activities as working time. (InfoCuria)

17. Voluntary Employee Data Contribution

A different situation exists where the employer says:

“Employees can voluntarily contribute professional data to our AI model and receive compensation.”

This resembles a licensing arrangement.

The contract should specify:

what data is contributed;

purpose of use;

duration;

territory;

AI training rights;

future model versions;

sublicensing;

commercial exploitation;

confidentiality;

withdrawal;

remuneration;

audit rights.

A clear contractual framework can substantially reduce later disputes.

18. Employee Data and Consent

Consent is complicated in employment relationships.

Employees may not always have genuine freedom to refuse an employer's data-processing request because of the power imbalance between employer and employee.

Therefore, an employer cannot simply say:

“The employee clicked ‘I agree’, so all AI training is lawful.”

The employer must identify the correct legal basis and comply with applicable employment and data-protection requirements.

19. Compensation for Privacy Harm

An employee could potentially claim damages where unlawful AI training causes legally recognised harm.

Examples might include:

disclosure of sensitive information;

unlawful profiling;

exposure of health information;

reputational harm;

loss of control over personal information;

discriminatory consequences;

unlawful international transfers.

The CJEU's C-300/21 judgment is important because it requires infringement, damage and causation for Article 82 GDPR compensation.

20. Compensation for Working Time

Consider:

An employee works 8 hours per day and is required to spend another 2 hours each evening correcting AI outputs.

The two additional hours may potentially constitute working time.

Questions include:

Was participation compulsory?

Was it employer-directed?

Was it performed for the employer's benefit?

Was the employer aware of the work?

Did national law classify it as overtime?

Was it already included in the salary?

Was there a collective agreement?

Were maximum working-time limits respected?

The C-909/19 judgment is therefore particularly relevant. (InfoCuria)

21. Compensation for Employee-Created Copyright

Suppose an employee writes 10,000 technical explanations and the employer uses them to train an AI model.

The legal question is not merely:

“Who supplied the data?”

Instead:

Who owns the copyright or other rights in the underlying material?

National copyright and employment rules may determine:

initial ownership;

employer's rights;

employee's rights;

contractual assignment;

permitted uses;

additional remuneration.

This is separate from the GDPR question.

22. Training Data Created During Employment

A useful distinction is:

Employer-owned work product

The employment contract may provide that certain works created in employment belong to the employer.

Employee personal information

The employer does not simply acquire unlimited ownership over the employee's personal data.

Third-party information

Employees may process customer or patient data that belongs neither to the employee nor automatically to the employer.

Trade secrets

Special confidentiality protections may apply.

Thus a single AI training dataset can contain four different legal categories simultaneously.

23. Data Training and Collective Bargaining

Collective bargaining can become particularly important.

A collective agreement may establish:

AI-use rules;

monitoring restrictions;

data-processing rules;

compensation for additional AI tasks;

consultation requirements;

transparency;

retraining;

job-transition protections;

employee participation.

The Bötel case is particularly useful because it demonstrates the relationship between employment-related training and remuneration/equality. (InfoCuria)

24. Equal Pay Issues

Suppose:

full-time employees receive full compensation for AI-training sessions;

part-time employees receive compensation only for their contractual hours;

part-time employees are disproportionately women.

Bötel demonstrates how apparently neutral compensation rules may create an indirect sex-discrimination/equal-pay problem if they disproportionately disadvantage part-time employees without objective justification. (InfoCuria)

Therefore, AI-training compensation policies should be reviewed for equality effects.

25. Employer's Defence: “The Data Was Generated During Work”

An employer may argue:

“The employee generated the information while being paid, so the employer owns everything.”

That proposition is too broad.

Being generated during employment does not necessarily answer:

whether the information is personal data;

whether copyright exists;

whether additional working time was required;

whether a contract transferred particular rights;

whether confidential information was involved;

whether processing is lawful.

Each legal regime has its own requirements.

26. Employer's Defence: “No Separate Work Was Performed”

This may be stronger where:

AI training is merely an ordinary part of the employee's existing job;

the employee performs the activity during paid hours;

the employment contract covers the activity;

no separate statutory or contractual compensation exists.

In that situation, a separate royalty generally does not automatically arise.

27. Employee's Argument: “AI Training Increased the Employer's Commercial Value”

An employee may argue:

“My data and professional contributions significantly improved the AI product, so I should receive additional compensation.”

Commercial value alone, however, does not automatically create a civil-law payment obligation.

The employee needs a legal basis such as:

contract;

copyright;

collective agreement;

statute;

unlawful processing;

unjust enrichment;

another recognised national-law claim.

28. Unjust Enrichment

Some national legal systems recognise unjust-enrichment claims where one person has obtained a benefit at another's expense without sufficient legal justification.

A theoretical AI-data claim could involve:

Employee contribution → employer benefit → absence of sufficient legal basis → enrichment claim.

But this is highly dependent on national law.

It should not be assumed that every profitable use of employee data produces unjust enrichment.

29. Data Training and Trade Secrets

An employee may contribute:

manufacturing methods;

technical specifications;

customer strategies;

confidential algorithms;

pricing methods.

If those materials are protected trade secrets, AI training may raise separate legal questions.

An employer may also be restricted from transferring confidential employee-created information to an external AI provider.

30. External AI Providers

The issue becomes more complicated where:

Employee → Employer → AI Provider

The employer may provide employee data to:

OpenAI-type providers;

cloud providers;

AI vendors;

analytics companies;

model-training contractors.

The legal questions then include:

controller/processor status;

lawful basis;

data-processing agreements;

international transfers;

confidentiality;

security;

employee notification;

contractual restrictions.

The CJEU cases concerning access to recipients and personal-data rights become especially relevant.

31. Cross-Border AI Training

An employee may work in Germany while:

employer is French;

AI provider is Irish;

cloud server is in the Netherlands;

training contractor is in Poland;

model-development team is in the United States.

This creates multiple legal questions:

applicable employment law;

GDPR;

international data transfer;

jurisdiction;

contractual law;

copyright;

civil damages.

The location of the AI server does not, by itself, determine every legal issue.

32. Burden of Proof

A worker seeking compensation may need to establish:

employment relationship;

contribution made;

nature of the contribution;

employer's use;

AI-training purpose;

additional work performed;

legal right to compensation;

damage;

causation.

Evidence may include:

employment contract;

timesheets;

AI-system logs;

internal policies;

emails;

data-processing records;

training documentation;

model-development records;

payroll;

collective agreements;

DPIAs;

processor agreements.

33. Right of Access as Evidence

The GDPR access jurisprudence can be particularly useful.

Under C-487/21, the right of access must allow meaningful understanding of personal-data processing.

Under C-154/21, information about actual recipients can be important where they can be identified.

Therefore:

Data access → information about processing → evidence → possible civil claim

can become an important litigation pathway.

34. Can an Employee Claim a Percentage of AI Revenue?

Generally, not automatically.

A revenue-sharing claim is strongest where there is:

an express contract;

collective agreement;

copyright licence;

royalty clause;

statutory remuneration right;

recognised employee invention/work compensation scheme.

Without such a legal basis, the fact that employee data contributed to an AI model does not by itself establish a percentage entitlement.

35. Employee Inventions and AI

A related but distinct area is employee invention law.

Some European jurisdictions provide special rules concerning inventions created by employees.

These may provide:

employer rights to inventions;

employee compensation;

notification procedures;

valuation mechanisms.

However:

employee invention law ≠ general employee-data royalty law.

The distinction should be maintained carefully.

36. Data Contribution by Creative Employees

The issue becomes particularly important for:

writers;

translators;

photographers;

designers;

musicians;

software developers;

journalists;

researchers.

Their work may contain copyright-protected expression.

An AI training dispute may therefore contain two separate claims:

Claim 1 — Labour claim

“I performed additional work and should have been paid.”

Claim 2 — Intellectual-property claim

“My protected work was used without the required authorisation or remuneration.”

The two claims should not be confused.

37. Data Training and Platform/Gig Workers

Gig workers generate huge quantities of operational data.

Examples:

drivers;

delivery workers;

couriers;

freelancers.

Their data can be used to:

optimise algorithms;

predict demand;

allocate jobs;

calculate prices;

evaluate performance;

train AI systems.

The legal issues can include:

worker status;

algorithmic management;

transparency;

personal-data processing;

remuneration;

automated decision-making.

This is an especially important developing area.

38. AI Training as a New Form of Labour

A useful conceptual distinction is:

Traditional labour

Employee creates a product or service.

Data labour

Employee creates information through work.

AI-training labour

Employee deliberately creates or labels information so that an AI model can learn.

For example:

Worker evaluates 50,000 AI-generated answers and marks each answer “correct” or “incorrect.”

That is not merely incidental data generation.

It may be direct productive labour for model development.

The stronger the connection between the activity and the employer's instructions, the stronger the employment-law analysis becomes.

39. Current Legal Gap

As of September 2026, European law does not yet provide a general rule saying:

“Workers are entitled to royalties whenever their personal or professional data is used to train AI.”

The existing legal system instead approaches the problem through several separate doctrines.

40. Six+ Case-Law Comparison

41. Most Important Legal Principle

For examination purposes, remember this formula:

Employee contribution to AI training does not automatically create a royalty.

Instead:

If it is work:

Working time → employment remuneration

If it is personal data:

Lawful processing → GDPR

If it is creative work:

Copyright → ownership/licensing/remuneration

If it is confidential knowledge:

Confidentiality/trade secrets

If it causes legally recognised harm:

Civil damages

If a contract promises compensation:

Contractual claim

42. Practical Example

Assume an employer develops an AI customer-service model.

Employees spend six months:

correcting AI answers;

classifying customer questions;

writing ideal answers;

evaluating AI performance.

The employer then commercialises the model.

Employee Claim A

“I worked 300 additional hours.”

→ Working-time/remuneration claim.

Employee Claim B

“My personal information was incorporated into the training system.”

→ GDPR analysis.

Employee Claim C

“My written training materials are copyright protected.”

→ Copyright analysis.

Employee Claim D

“My employment contract promised bonuses for AI-development contributions.”

→ Contract claim.

Employee Claim E

“Employees of one sex received less compensation for the same AI-training work.”

→ Equality/equal-pay analysis.

These claims may coexist but have different legal foundations.

43. Defences Available to Employers

Employers may argue:

AI-related work was already included in the employment contract.

Employees were already paid for the relevant hours.

No additional work was performed.

The data was necessary for employment administration.

Processing had a lawful GDPR basis.

No copyright belongs to the employee.

Rights were contractually transferred.

No actual damage occurred.

No causal connection exists.

The alleged compensation right does not exist under national law.

The employee was not required to participate.

The employee's data was anonymised or otherwise processed in a legally permissible manner.

44. Employee Arguments

Employees may argue:

AI annotation was additional work;

the activity was compulsory;

work was performed outside paid hours;

personal data was processed beyond the original employment purpose;

information was supplied to third-party AI providers;

the employer failed to provide transparency;

protected creative works were used;

contractual compensation was promised;

the employee suffered material or non-material damage;

collective agreements required consultation or compensation.

45. Remedies

Depending on the legal basis and national law, remedies may include:

Labour remedies

unpaid wages;

overtime;

holiday/pay adjustments;

contractual compensation.

GDPR remedies

access;

rectification;

erasure where applicable;

restriction;

objection;

compensation.

IP remedies

injunction;

damages;

licensing remuneration;

account of profits where national law permits.

Contract remedies

payment;

damages;

restitution.

Collective remedies

representative proceedings;

works-council remedies;

collective bargaining.

46. Conclusion

Data-training contribution labour compensation is an emerging area of European civil and employment law rather than an already-settled independent cause of action.

The most important established principles are:

Bötel (C-360/90) shows that employment-related training compensation can fall within the concept of pay. (InfoCuria)

C-909/19 shows that mandatory employer-requested training can constitute working time. (InfoCuria)

Pantuso (C-616/16 and C-617/16) establishes remuneration rights for qualifying specialist medical training under the applicable EU legislation. (InfoCuria)

C-590/20 confirms remuneration and potential State-liability consequences in the specialist-training context. (InfoCuria)

KKO 2019:28 demonstrates that employment training-cost arrangements must respect mandatory employee protections and reasonableness. (curia)

Meta C-252/21 demonstrates the growing interaction between commercial data exploitation, data protection and market regulation. (tuc.org.uk)

C-300/21 establishes the infringement + damage + causation structure for GDPR compensation.

C-154/21 and C-487/21 provide important tools for discovering and understanding the processing of employee data.

The central distinction is therefore:

An employee does not automatically become a shareholder or royalty recipient merely because their data contributes to AI training. But where AI training involves additional labour, unlawful data processing, protected intellectual property, contractual compensation rights, or legally recognised damage, European and national law can provide routes to compensation.

Ultra-Basic Keywords

Data Training – Employee Data – AI Training – Data Labour – AI Annotation – Model Training – Working Time – Remuneration – Overtime – Employee Data – GDPR – Personal Data – Copyright – Trade Secret – Contract – Collective Agreement – Training Compensation – Data Contribution – AI Worker – Algorithmic Management – Data Access – Consent – Lawful Basis – Damage – Causation – Compensation – Bötel – C-909/19 – Pantuso – C-590/20 – KKO 2019:28 – Meta – C-300/21 – C-154/21 – C-487/21.