Civil Law and Air Traffic Control AI Decision Error Claims in Europe

1. Introduction

Air Traffic Control (ATC) AI decision-error liability concerns civil claims arising when an AI-assisted or AI-controlled air-traffic-management system makes an erroneous recommendation, prediction, classification, sequencing decision, conflict alert, routing decision, or other operational decision and that error causes damage.

Examples include:

AI incorrectly predicting an aircraft conflict;

AI failing to identify a potential collision;

AI generating a false collision warning;

AI incorrectly assigning aircraft sequencing;

AI recommending an unsafe route;

AI incorrectly allocating departure or arrival slots;

AI misinterpreting radar/ADS-B/GNSS data;

AI failing to detect abnormal aircraft behaviour;

AI making an unsafe recommendation that an air-traffic controller accepts;

AI automation bias causing a controller to disregard contradictory information;

an AI system malfunctioning because of defective software, training data, sensors or communications infrastructure.

This is a developing area. There is not yet a substantial body of European judgments specifically deciding that an AI-powered ATC system caused an aviation accident and determining civil damages. The most important current authority is, however, exceptionally close to the problem: Republik Österreich v Austrian Airlines, C-408/24 (2026), concerning negligent failure of an air-navigation service provider and material loss suffered by an airline. (EUR-Lex)

The legal analysis therefore combines:

ATC liability + aviation safety + AI regulation + product liability + ordinary national civil liability + causation.

2. What Is an AI Decision Error?

An AI decision error can arise at several levels.

A. Input error

The system receives incorrect:

radar data;

ADS-B data;

weather information;

aircraft-position data;

flight-plan information;

communication data.

B. Processing error

The algorithm incorrectly processes accurate information.

C. Prediction error

The AI predicts that two aircraft are sufficiently separated when they are not.

D. Recommendation error

The system correctly understands the situation but recommends an unsafe course of action.

E. Human-overreliance error

The AI gives an incorrect recommendation and the controller follows it without adequate verification.

F. System-design error

The AI was poorly designed for the operational environment.

G. Updating/training error

A software update or machine-learning model changes system behaviour and creates an unexpected safety problem.

3. Basic Civil-Liability Formula

An AI-ATC negligence claim can be reduced to:

DUTY → AI SYSTEM → ERROR → HUMAN/OPERATIONAL DECISION → CAUSATION → DAMAGE → LIABILITY

More traditionally:

DUTY + BREACH + CAUSATION + DAMAGE + REMEDY

The difficult issue is determining who committed the legally relevant breach.

4. Possible Defendants

An AI-ATC accident may involve several entities.

The court must distinguish AI error from human error and from systemic organisational failure.

5. The Most Important Current Case: Austrian Airlines, C-408/24

Case 1 — Republik Österreich v Austrian Airlines AG, C-408/24

CJEU, 12 February 2026

This is the most directly relevant European authority presently available.

The case arose after an aeronautical telecommunications server operated by Austro Control failed. The failure substantially reduced the arrival and departure rate at Vienna-Schwechat Airport, resulting in cancellation of Austrian Airlines flights and material financial loss.

The issue was whether EU legislation governing air-navigation services was intended to protect airspace users against material damage caused by a culpable failure by the air-traffic service provider.

The CJEU answered yes: Article 2(4) of Regulation 549/2004 and Article 8 of Regulation 550/2004, together with the relevant air-navigation-service and charging provisions, are intended to protect airspace users against material damage caused by a culpable failure of an air-traffic service provider to comply with its obligations. (EUR-Lex)

Why this case is extremely important for AI-ATC litigation

Although the case did not involve AI, it establishes a highly relevant proposition:

Air-navigation-service obligations can protect airspace users against material economic damage caused by culpable failures in the provision of those services.

Therefore, if an AI-assisted ATC system causes damage because the responsible service provider negligently:

deploys it;

supervises it;

maintains it;

validates it;

fails to respond to known defects;

the case provides an important European-law foundation for analysing the protected interests of airspace users.

Important limitation

The CJEU did not itself award damages to Austrian Airlines. It left the national court to determine whether Austria could actually be held liable on the facts under national law. (EUR-Lex)

6. Case 2 — SELEX Sistemi Integrati v Commission, T-155/04

General Court, 12 December 2006

SELEX concerned air-traffic-management equipment and systems and the role of Eurocontrol in standardisation.

The General Court examined whether Eurocontrol's standardisation activities in the field of air-traffic-management equipment constituted an economic activity for EU competition-law purposes. (Infocuria)

Relevance to AI-ATC

AI-based ATC systems depend on:

technical standards;

interoperability;

equipment;

software;

operational procedures.

The case demonstrates that air-traffic-management standardisation is legally significant and cannot simply be treated as an ordinary commercial software market.

Civil-law significance

For an AI error claim, the relevant question may include:

Was the AI system designed, certified and operated consistently with the applicable aviation/ATM standards?

Failure to comply with applicable technical standards can become important evidence of breach, although standard non-compliance does not automatically establish civil liability.

7. Case 3 — SELEX Sistemi Integrati v Commission, C-481/07 P

CJEU, 16 July 2009

This was the appeal in the SELEX litigation.

The CJEU considered the General Court's treatment of Eurocontrol's activities and the distinction between activities involving the exercise of public authority and economic activities. (Infocuria)

Relevance

It reinforces an important distinction:

An entity involved in air-traffic management may perform different functions that have different legal characteristics.

For AI liability, this matters because the legal regime can differ according to whether the relevant activity concerns:

public ATC;

technical standardisation;

procurement;

commercial software supply;

private aviation services.

The identity and legal function of the defendant must therefore be determined before selecting the liability regime.

8. Case 4 — Wallentin-Hermann v Alitalia, C-549/07

CJEU, 22 December 2008

This case concerned cancellation caused by aircraft technical problems.

The CJEU held that a technical problem does not automatically constitute an “extraordinary circumstance”; the relevant event must be outside the normal exercise of the carrier's activity and beyond its actual control. The carrier must also show that it took reasonable measures. (Infocuria)

AI-ATC relevance

Suppose an airline argues:

“The AI system unexpectedly malfunctioned, so the resulting cancellation should be treated as beyond our control.”

Wallentin-Hermann shows why courts distinguish between:

ordinary operational/technical problems; and

genuinely external events.

This is particularly useful when an AI failure results in:

flight cancellation;

delay;

diversion;

operational disruption.

Important limitation

Wallentin-Hermann concerns passenger compensation under Regulation 261/2004, not civil damages for an AI-ATC accident.

It is therefore analogical authority.

9. Case 5 — Airhelp v Scandinavian Airlines, C-28/20

CJEU Grand Chamber, 23 March 2021

The CJEU considered the meaning of an “extraordinary circumstance” under Regulation 261/2004.

It distinguished events arising internally within an air carrier's activity from external events beyond the carrier's control. (Infocuria)

AI-ATC relevance

This distinction becomes useful where the AI error originates from:

Internal cause

airline's own AI system;

airline's own software;

airline's own maintenance;

airline's own operational decision.

versus

External cause

independent ATC provider;

government air-navigation decision;

third-party telecommunications failure;

external infrastructure failure.

This distinction may affect the allocation of responsibility.

10. Case 6 — D (Air Traffic Management Decision), T-134/25

General Court, 21 January 2026

This is another highly relevant current case.

The case concerned an air-traffic-management decision affecting delayed departure slots following adverse weather conditions and the application of Regulation 261/2004. The General Court held that an air-traffic-management decision was not inherently part of the airline's ordinary activity and, where the airline had not contributed to it, was beyond the airline's actual control. (Infocuria)

But there is a crucial 2026 development

The CJEU subsequently decided that the judgment should be reviewed because of a potential issue concerning the unity or consistency of EU law.

In C-108/26 RX, the Reviewing Chamber ordered review on 12 March 2026. (Curia)

Therefore:

T-134/25 should not presently be treated as the final word on the legal status of air-traffic-management decisions.

AI relevance

The case illustrates an important attribution question:

If an AI system is operated by the ATC provider, can the airline treat the resulting decision as external to its own operations?

The answer depends upon who controls the system and who legally operates the relevant ATC function.

11. Case 7 — SCHUFA, C-634/21

CJEU, 7 December 2023

This is not an aviation case, but it is one of the leading European automated-decision authorities.

The CJEU examined automated credit scoring under GDPR Article 22 and recognised the legal significance of automated decision-making where an automated score effectively determines a person's outcome. (EUR-Lex)

Relevance to AI-ATC

The analogy is:

AI score/recommendation → human decision → legal/physical consequence

An ATC AI may produce:

collision-risk score;

separation recommendation;

route recommendation;

sequencing prediction.

The legal question becomes:

How much independent human judgment actually remains?

If the controller merely rubber-stamps an AI output, the factual distinction between AI recommendation and AI decision becomes less significant for liability analysis.

Important limitation

GDPR Article 22 does not directly govern ordinary aircraft-separation decisions. SCHUFA is therefore analogical AI-law authority, not an ATC-liability precedent.

12. Case 8 — Dun & Bradstreet Austria, C-203/22

CJEU, 27 February 2025

The CJEU held that, in applicable automated decision-making situations, a person can require meaningful information explaining the logic involved, in a form that allows the decision to be understood and challenged. (EUR-Lex)

Relevance to AI-ATC

Suppose an AI system generates:

“Aircraft A should descend immediately.”

After an accident, the claimant asks:

Why did the AI produce that recommendation?

What inputs did it use?

What thresholds applied?

Was contradictory data considered?

Did the model malfunction?

Was the recommendation within the system's validated operating range?

Dun & Bradstreet provides a useful conceptual basis for understanding why algorithmic opacity can become legally significant.

But again:

The case concerns GDPR automated decision-making, not aviation safety.

It cannot simply be transplanted to every ATC AI system.

13. Current AI Act Framework

The EU AI Act is relevant, but one must be precise.

The AI Act establishes obligations for certain high-risk AI systems, including requirements concerning:

risk management;

data governance;

technical documentation;

logging;

transparency;

human oversight;

accuracy;

robustness;

cybersecurity;

post-market monitoring.

Article 14 requires high-risk AI systems to be designed so that natural persons can effectively oversee them, with oversight measures intended to prevent or minimise risks to health, safety and fundamental rights. (EUR-Lex)

Aviation connection

The AI Act's high-risk classification can interact with EU aviation product legislation. Annex I includes Regulation 2018/1139 concerning civil aviation, although the precise AI classification depends on the system, product and applicable conditions. (AI Act Service Desk)

Therefore, an AI component integrated into aviation safety equipment requires a specific classification analysis rather than simply assuming:

“Every ATC AI = high-risk AI.”

14. AI Act Compliance Is Not the Same as Civil Liability

This distinction is extremely important.

Suppose an ATC AI complies with all AI Act requirements.

Then:

AI Act compliance does not automatically eliminate civil liability.

Conversely:

A regulatory violation does not automatically establish every element of a civil damages claim.

The claimant must still establish the applicable civil-law elements, especially:

breach → causation → damage.

15. Who Is Responsible for an AI Error?

There are several possible responsibility models.

Model 1 — Developer error

The developer creates an unsafe algorithm.

Example:

AI systematically underestimates aircraft separation in certain weather conditions.

Potential issues:

defective design;

inadequate testing;

inadequate risk assessment;

software defect;

failure to warn.

Model 2 — Provider error

The AI provider supplies an inadequately validated system.

Potential issues:

inadequate documentation;

inadequate safety testing;

insufficient cybersecurity;

failure to disclose known limitations.

Model 3 — ATC operator error

The air-navigation provider deploys the AI without adequate:

validation;

training;

monitoring;

human oversight;

maintenance.

This is particularly important following C-408/24 Austrian Airlines, because EU air-navigation provisions can protect airspace users against material damage resulting from culpable failure by an air-traffic service provider. (EUR-Lex)

Model 4 — Controller error

The AI provides a correct warning but the controller:

ignores it;

misunderstands it;

fails to intervene;

enters an incorrect command.

This creates a human-AI interaction problem rather than a pure AI defect.

Model 5 — Combined liability

The most realistic difficult cases may involve:

defective AI + inadequate training + poor human oversight + controller error.

The court may need to determine how each factor contributed to the final damage.

16. Causation in AI-ATC Claims

Causation will often be the hardest issue.

Example:

AI incorrectly predicts safe separation

↓

Controller accepts recommendation

↓

Aircraft enters unsafe trajectory

↓

Controller attempts correction

↓

Pilot receives delayed instruction

↓

Collision occurs

↓

Passenger injuries

The claimant has to establish the causal chain.

17. AI Error Alone Does Not Establish Liability

A critical principle:

An incorrect AI output is not automatically a legally actionable breach.

The claimant may need to show that:

the system was defective;

the system was improperly deployed;

the system was inadequately supervised;

the operator breached a professional/technical duty;

the AI output should reasonably have been detected;

the error caused the damage.

18. Human Oversight

Human oversight is central to AI-ATC liability.

A system might be designed as:

“AI assists controller”

rather than:

“AI controls aircraft.”

This distinction affects the allocation of responsibility.

The AI Act requires effective human oversight for applicable high-risk systems, with the oversight function intended to enable detection of anomalies and intervention where necessary. (EUR-Lex)

Therefore, litigation may ask:

Was the controller actually capable of understanding and overriding the AI?

19. Automation Bias

Automation bias occurs when humans place excessive reliance on automated recommendations.

Example:

AI says “no conflict.”

The controller sees no immediate reason to disagree.

Later evidence shows the AI incorrectly interpreted radar data.

The legal question becomes:

Should the system have been designed so that the controller could reasonably identify the error?

This can involve:

interface design;

warnings;

confidence indicators;

independent verification;

training;

workload management.

20. Black-Box AI

A claimant may encounter a major evidentiary difficulty:

“The AI generated the wrong recommendation, but we do not know why.”

This creates questions concerning:

source code;

model architecture;

training data;

logs;

system inputs;

model version;

update history;

confidence scores;

safety validation.

The logic of Dun & Bradstreet makes algorithmic explainability legally significant in its GDPR context, although that judgment does not itself establish a general aviation right to obtain source code. (EUR-Lex)

21. AI Training Data

An ATC AI may be trained using:

historical flight data;

weather data;

radar data;

airspace patterns;

incident data;

simulated aircraft trajectories.

Errors may arise because training data:

are incomplete;

contain historical anomalies;

fail to represent rare events;

are geographically biased;

do not adequately cover extreme weather.

The civil-law issue becomes:

Was the training and validation methodology appropriate for a safety-critical aviation system?

22. Software Updates

AI systems may change following:

software updates;

model retraining;

parameter changes;

cybersecurity patches;

new data.

Suppose:

Version 4.1 works correctly, but Version 4.2 introduces an unsafe prediction error.

The court may have to identify:

who authorised the update;

whether it was tested;

whether the provider knew of the risk;

whether the system should have been rolled back.

This is especially important for modern product-liability analysis because software can remain subject to continuing modification after initial deployment.

23. Product Liability

An AI-powered ATC component can potentially raise product-liability questions where the relevant software or integrated equipment qualifies as a product under the applicable EU regime.

Possible defect categories include:

Design defect

The algorithm is inherently unsafe.

Manufacturing/deployment defect

The particular installed version is defective.

Information defect

Users were not adequately warned about limitations.

Cybersecurity defect

The system was insufficiently protected against manipulation.

Update defect

A later software update introduces an unreasonable safety risk.

24. Air-Navigation Service Provider Liability

This is where C-408/24 Austrian Airlines becomes particularly powerful.

The CJEU specifically dealt with:

failure of an aeronautical telecommunications server → reduction in airport traffic capacity → cancelled flights → material financial damage.

The Court concluded that the EU air-navigation rules were intended to protect airspace users against material damage caused by a culpable failure of the air-traffic service provider. (EUR-Lex)

An AI decision error could fit conceptually into the same structure:

AI malfunction → ATC service failure → operational disruption → economic/physical damage.

The precise civil claim still depends on national law and the facts.

25. Economic Damage

AI-ATC errors can cause:

cancelled flights;

delays;

fuel consumption;

diversions;

crew costs;

passenger compensation;

airport congestion;

cargo losses;

missed connections;

loss of contracts.

C-408/24 is especially relevant because the underlying dispute concerned material financial loss, rather than personal injury alone. (EUR-Lex)

26. Personal Injury and Death

The stakes become substantially greater when an AI error causes:

collision;

runway incursion;

loss of separation;

controlled-flight-into-terrain event;

emergency landing;

passenger injury or death.

Potential claims may involve:

national tort/delict law;

aviation liability;

Montreal Convention;

product liability;

employer/operator liability;

State liability.

27. AI + Aviation Accident

Consider:

AI incorrectly classifies two aircraft as sufficiently separated.

The controller follows the recommendation.

The aircraft collide.

Potential defendants include:

AI developer;

AI provider;

ATC service provider;

State;

aircraft manufacturer;

avionics manufacturer;

maintenance contractor;

potentially the airline or controller.

The court must separate each causal contribution.

28. Contributory Negligence

A defendant might argue:

“The controller should have independently checked the AI recommendation.”

The claimant may respond:

“The system was specifically designed and certified to assist the controller and gave no warning that its output was unreliable.”

The court would need evidence concerning:

operational procedures;

controller training;

system design;

warning mechanisms;

foreseeable use;

human-machine interaction.

29. Force Majeure / External Event

An AI system might fail because of:

satellite disruption;

cyberattack;

extreme weather;

telecommunications outage;

power failure.

Whether that excuses liability depends on the applicable legal regime.

Wallentin-Hermann and Airhelp demonstrate that EU aviation law carefully distinguishes circumstances that are internal to the carrier's ordinary operations from genuinely external events beyond its control. (Infocuria)

But these cases concern passenger compensation, not general tort liability.

30. AI Decision vs ATC Decision

An important legal distinction is:

AI output

“Aircraft A should turn left.”

Human decision

Controller accepts the recommendation.

Operational act

Instruction is transmitted to aircraft.

Physical consequence

Aircraft changes trajectory.

Civil liability should not automatically stop at the AI output.

The court must analyse the whole decision chain.

31. Evidence

AI-ATC litigation may require extensive technical evidence.

Important records include:

radar recordings;

ADS-B data;

flight plans;

ATC voice recordings;

controller logs;

AI input logs;

AI output logs;

model version;

software version;

system-update records;

training records;

system validation documents;

cybersecurity logs;

maintenance records;

incident reports.

32. Preservation of AI Logs

AI systems can create an unusual evidentiary problem.

If the system overwrites logs after 30 days and litigation begins months later:

Can the claimant prove what the AI actually did?

This may lead to disputes about:

evidence preservation;

disclosure;

technical documentation;

audit trails;

adverse inferences;

confidentiality;

trade secrets.

The AI Act's logging and technical-documentation requirements for applicable high-risk systems become potentially important evidence sources. (EUR-Lex)

33. Confidentiality and Trade Secrets

AI providers may argue:

“The model architecture and algorithm are trade secrets.”

But confidentiality does not necessarily mean that the system is immune from judicial scrutiny.

Dun & Bradstreet illustrates the need to balance:

explanation/access;

trade secrets;

third-party rights;

effective judicial protection.

(EUR-Lex)

In aviation litigation, a court could therefore need to balance safety evidence against legitimate confidentiality interests.

34. Cross-Border Claims

An ATC incident can involve:

Country A: AI developer
Country B: ATC provider
Country C: aircraft operator
Country D: accident location
Country E: injured passenger.

The court must determine:

jurisdiction;

applicable substantive law;

contractual relationships;

aviation conventions;

EU regulations;

national public-authority liability.

The complexity increases where the ATC provider is a public entity.

35. Public Authority Liability

Many air-navigation services are connected to public authorities or State-designated service providers.

Therefore, a claimant may have to determine whether the action is:

ordinary private-law negligence;

public-authority liability;

State liability;

EU-law liability.

C-408/24 is particularly significant because the underlying proceedings concerned the Republic of Austria's potential liability for the conduct of Austro Control. The CJEU confirmed that the relevant EU air-navigation rules are intended to protect airspace users from the relevant material damage, while leaving the actual national-law liability determination to the referring court. (EUR-Lex)

36. Important Case-Law Table

37. Important Status of T-134/25

For an exam or research paper, write:

T-134/25, D (Air Traffic Management Decision), judgment of 21 January 2026, is subject to review in C-108/26 RX.

The CJEU's Reviewing Chamber ordered review on 12 March 2026. (Curia)

Therefore, do not present T-134/25 as an unqualified final CJEU precedent.

38. Six Most Important Cases to Memorise

1. Austrian Airlines — C-408/24

ATM service failure → material damage protection.

2. SELEX — T-155/04

ATM standardisation → legal character of air-traffic-management functions.

3. SELEX — C-481/07 P

Eurocontrol/ATM public functions → economic-activity distinction.

4. Wallentin-Hermann — C-549/07

Technical/operational event → actual control and reasonable measures.

5. Airhelp — C-28/20

External event → control and aviation liability analysis.

6. D — T-134/25

Air-traffic-management decision → external to carrier; currently under CJEU review.

For the AI-specific component, add:

7. SCHUFA — C-634/21

Automated decision-making.

8. Dun & Bradstreet — C-203/22

Explainability of automated decision-making.

39. Key Legal Issues in an AI-ATC Claim

Issue 1 — Was the AI defective?

If yes:

Product/software liability may become relevant.

Issue 2 — Was the AI negligently deployed?

If yes:

ATC-service-provider liability may arise.

Issue 3 — Did the controller rely improperly on AI?

If yes:

Human negligence may contribute.

Issue 4 — Was the AI properly supervised?

If no:

Human-oversight obligations become relevant.

Issue 5 — Did the AI actually cause the damage?

This is the central causation question.

Issue 6 — Who controlled the system?

This determines attribution.

40. Example

Imagine:

A European air-navigation authority deploys an AI conflict-detection system.

The AI receives radar data showing two aircraft approaching one another.

The AI incorrectly classifies the aircraft as safely separated.

The controller accepts the recommendation.

The aircraft collide.

Possible legal analysis

AI developer

→ defective algorithm?

ATC provider

→ inadequate testing or supervision?

Controller

→ unreasonable reliance?

Aircraft operator

→ inadequate response?

Equipment provider

→ defective radar data?

State

→ public-authority liability?

The court would reconstruct the entire technological and human chain.

41. Another Example — Economic Loss

Suppose the AI does not cause an accident but wrongly allocates arrival slots.

Result:

100 flights delayed;

20 flights cancelled;

airline loses €1 million.

C-408/24 becomes particularly important because the CJEU expressly recognised the protective purpose of EU air-navigation rules concerning material damage suffered by airspace users following culpable failure of the air-traffic service provider. (EUR-Lex)

42. Difference Between Passenger Claims and Airline Claims

This distinction is essential.

Passenger

May have rights under:

Regulation 261/2004;

Montreal Convention;

national law.

Airline

May claim against:

ATC service provider;

State;

technology provider;

contractor.

The legal basis may be entirely different.

C-408/24 is especially significant because it concerns the airspace user's material damage, rather than merely standard passenger compensation. (EUR-Lex)

43. AI Liability Formula

A useful legal formula is:

AI-ATC LIABILITY = SYSTEM DESIGN + DATA + VALIDATION + DEPLOYMENT + HUMAN OVERSIGHT + MONITORING + UPDATE + CYBERSECURITY + ERROR + CAUSATION + DAMAGE

For an exam:

AI ERROR → ATC DECISION → HUMAN RESPONSE → AVIATION CONSEQUENCE → CAUSATION → CIVIL LIABILITY

44. Ultra-Basic Revision Notes

Meaning

AI-ATC decision error means an AI system used in air-traffic management produces an incorrect output or recommendation that contributes to damage.

Main legal areas

Aviation law

Air-navigation-service regulation

National tort/delict law

Contract law

Product liability

AI Act

Data protection where personal data are involved

State liability

Cross-border private international law

Main defendants

Developer → Provider → ATC Operator → Controller → Airline → State

Main questions

Duty → Error → Breach → Causation → Damage → Remedy

45. Conclusion

AI-based air-traffic-control decision-error liability in Europe is an emerging field in which traditional aviation and civil-liability principles are being applied to increasingly autonomous decision-support systems. There is not yet a large body of reported European judgments directly deciding an accident caused by an AI-controlled ATC decision.

The most important current authority is Republik Österreich v Austrian Airlines, C-408/24, decided by the CJEU on 12 February 2026. The Court held that EU air-navigation legislation is intended to protect airspace users against material damage caused by a culpable failure of an air-traffic service provider to comply with its obligations. (EUR-Lex)

The AI component then adds questions concerning algorithmic design, validation, human oversight, explainability, logging, software updates, cybersecurity and product defects. The AI Act's human-oversight requirements for applicable high-risk systems reinforce the importance of keeping meaningful human control over safety-critical AI. (EUR-Lex)

At the same time, T-134/25 (D — Air Traffic Management Decision) is a significant 2026 authority but must be treated cautiously because the CJEU ordered review of that judgment in C-108/26 RX. (Curia)

One-line exam answer

In Europe, civil liability for an AI-based ATC decision error depends on the interaction of air-navigation-service duties, national negligence and State-liability principles, aviation and product-liability rules, and AI governance requirements, with the decisive questions being whether the AI system or its deployment was legally defective, who controlled the relevant decision, whether adequate human oversight existed, and whether the error caused legally recoverable damage.