Civil Law and Autonomous Greenhouse System Failure Liability in Europe

1. Introduction

An autonomous greenhouse system is a technologically integrated agricultural system in which software, sensors, AI and automated machinery control important greenhouse functions with limited continuous human intervention.

Such a system may automatically control:

temperature;

humidity;

ventilation;

irrigation;

fertilisation;

lighting;

CO₂ levels;

pest detection;

shading;

heating and cooling;

energy consumption;

crop monitoring;

robotic harvesting;

nutrient delivery;

emergency shutdowns.

A failure can therefore produce substantial losses.

For example:

AI incorrectly detects adequate soil moisture → irrigation is stopped → greenhouse temperature rises → plants die → farmer suffers crop and contractual losses.

Or:

Sensor fails → AI does not detect excessive temperature → ventilation remains closed → entire commercial crop is destroyed.

European civil liability therefore has to deal with product defects, software defects, contractual liability, negligence, AI-related risks, causation, evidence and agricultural losses.

There is currently no substantial European appellate case law specifically concerning an autonomous AI greenhouse failure. The appropriate approach is therefore to apply established European product-liability, technological-liability and environmental-liability principles by analogy. The European Commission itself has identified autonomy, opacity, connectivity, data dependency, software updates and complex value chains as challenges for AI-related product liability. (Eur-Lex)

2. Basic Legal Structure

An autonomous greenhouse can be viewed as a system, rather than merely one machine.

For example:

Sensors → Internet connection → AI software → controller → irrigation/heating system → greenhouse → crops

A failure at any point may cause the final damage.

Potentially responsible actors include:

greenhouse manufacturer;

AI/software developer;

sensor manufacturer;

irrigation-equipment manufacturer;

installer;

maintenance company;

cloud/AI service provider;

farmer/operator;

supplier of defective inputs;

insurer.

The central legal question is:

Which actor's legally relevant failure caused the damage?

3. Main Types of Damage

An autonomous greenhouse failure can cause several different forms of damage.

A. Property damage

destruction of greenhouse equipment;

damaged irrigation systems;

destroyed heating systems;

damaged crops.

B. Economic loss

lost harvest;

lost future profits;

contractual penalties;

wasted electricity;

replacement costs.

C. Personal injury

A malfunctioning robotic or automated system could injure:

workers;

contractors;

visitors;

maintenance personnel.

D. Environmental damage

A malfunction could cause:

excessive water consumption;

chemical discharge;

fertiliser leakage;

contamination;

uncontrolled emissions.

E. Contractual losses

A farmer may be unable to supply crops to:

supermarkets;

food processors;

wholesalers;

export customers.

4. European Product-Liability Framework

The traditional European product-liability regime is based on strict liability for defective products.

Under the former Product Liability Directive, the injured person generally had to establish:

damage;

defect;

causal relationship between defect and damage.

The CJEU has described this as a strict-liability system that does not require proof that the producer was personally at fault. (curia)

The EU product-liability framework has subsequently been modernised, including to deal more explicitly with digital technologies and software.

For an autonomous greenhouse, this is significant because the defective element might be:

hardware;

embedded software;

control software;

an AI model;

an update;

cybersecurity;

interaction between components.

5. Case 1 — Commission v United Kingdom, C-300/95

Court

CJEU

Year

1997

Issue

The case concerned the development-risk defence under the Product Liability Directive.

Principle

The CJEU examined when the state of scientific and technical knowledge could prevent a producer from discovering a defect at the relevant time. (curia)

Importance for autonomous greenhouse systems

Suppose an AI greenhouse contains a previously unknown software vulnerability.

The manufacturer might argue:

“The defect could not reasonably have been discovered when the system was placed on the market.”

The development-risk defence may therefore become relevant, subject to the applicable product-liability legislation.

However, autonomous systems create a special complication:

The system may continue changing after deployment.

Therefore, courts may have to distinguish between:

a defect existing at initial deployment;

a defect introduced through an update;

a defect resulting from machine learning;

a defect caused by inadequate maintenance.

6. Case 2 — Boston Scientific Medizintechnik, Joined Cases C-503/13 and C-504/13

Court

CJEU

Year

2015

Facts

The cases concerned potentially defective medical devices, including pacemakers and implantable cardioverter defibrillators.

Principle

The CJEU explained that a product is defective where it does not provide the safety that a person is entitled to expect, taking all circumstances into account.

It also recognised that products belonging to the same production series may present a higher risk of failure even when an individual product has not yet malfunctioned. (Infocuria)

Relevance

This principle can be important for autonomous greenhouse systems.

Imagine that a manufacturer discovers:

2,000 greenhouse controllers contain a software/hardware defect capable of causing overheating.

Even if only one controller has failed so far, the safety expectations concerning the relevant series may become important.

Possible responses could include:

software update;

recall;

replacement;

warning;

inspection.

Key principle

Safety expectations can extend beyond the particular unit that has already failed.

7. Case 3 — N.W., L.W. and C.W. v Sanofi Pasteur, C-621/15

Court

CJEU

Year

2017

Issue

The case concerned proof of defect and causation in product-liability litigation where scientific evidence was uncertain.

Principle

The CJEU accepted that, where there is no scientific consensus, national courts may in appropriate circumstances consider a body of serious, specific and consistent evidence in determining whether defect and causation have been established. (curia)

Relevance to autonomous greenhouse failure

AI systems can create difficult causation questions.

Suppose:

Sensor malfunction → AI miscalculation → irrigation failure → crop destruction.

The farmer may not have access to the internal AI logs.

Evidence may instead consist of:

sensor records;

temperature records;

irrigation records;

system alerts;

maintenance records;

timing of crop damage;

expert evidence.

The case demonstrates the importance of carefully analysing available circumstantial evidence when direct proof of the defect or causal mechanism is difficult.

Important qualification

This was a vaccine case, not an agricultural or AI case. Its evidentiary principles are therefore analogical.

8. Case 4 — O'Byrne v Sanofi Pasteur MSD, C-127/04

Court

CJEU

Year

2006

Issue

The case concerned when a product is regarded as having been put into circulation and the identity of the responsible producer within a distribution structure. (Infocuria)

Principle

The CJEU examined when the product leaves the manufacturer's production process and enters the marketing process.

Relevance to autonomous greenhouse systems

An autonomous greenhouse is often assembled from multiple components:

sensor manufacturer → controller manufacturer → AI developer → integrator → greenhouse supplier → farmer.

Suppose the final system fails.

A legal question may be:

At what point did the relevant product or component enter circulation, and which entity qualifies as the legally relevant producer?

This becomes especially important where:

software is supplied separately;

components are integrated later;

updates occur after installation;

the system is modified by an integrator.

Key lesson

Complex technological supply chains require careful identification of the responsible producer and relevant point of circulation.

9. Case 5 — González Sánchez v Medicina Asturiana, C-183/00

Court

CJEU

Year

2002

Issue

The case concerned the relationship between the harmonised product-liability regime and national civil-liability systems.

Principle

The CJEU explained that the Product Liability Directive establishes a harmonised regime within its scope, while other contractual or non-contractual liability regimes based on different legal grounds may continue to operate in appropriate circumstances. (curia)

Relevance

This is highly important for autonomous greenhouse failures.

A farmer may potentially have several legal routes:

Product liability

“The greenhouse control system was defective.”

Contract

“The supplier promised a particular level of automated irrigation.”

Negligence

“The installer failed to configure the system correctly.”

Maintenance contract

“The service provider failed to install a critical update.”

Warranty

“The system did not perform according to the contractual guarantee.”

Thus, failure of a product-liability claim does not necessarily mean that every other possible civil claim disappears.

10. Case 6 — Commission v France, C-52/00

Court

CJEU

Year

2002

Issue

The case concerned the harmonised character of the Product Liability Directive.

Principle

The CJEU held that the Directive established a harmonised framework in the areas it covered and rejected certain national deviations from that regime. (Infocuria)

Relevance

This matters because autonomous greenhouse disputes may involve businesses operating across multiple EU Member States.

A manufacturer might be based in:

Germany

while:

AI developer = Netherlands
greenhouse = Spain
farmer = Italy
damage = Italy.

The harmonised EU product-liability rules help establish a common baseline, while national law continues to matter for issues outside the harmonised scope and for procedural and other civil-law questions.

11. Case 7 — Ford Italia, C-157/23

Court

CJEU

Judgment

19 December 2024

Issue

The case concerned the meaning of “producer” under the Product Liability Directive, including circumstances where another business presents itself in a manner associated with the actual manufacturer.

Principle

The CJEU examined when a supplier may fall within the producer concept where it presents itself as the producer. (Infocuria)

Relevance

This can be significant in autonomous greenhouse systems.

Suppose a company sells a complete system under its own brand:

“AgriAI Smart Greenhouse.”

But the system actually contains:

third-party sensors;

another company's AI software;

another manufacturer's irrigation controller.

The farmer may reasonably deal primarily with the branded system provider.

Ford Italia is therefore useful when analysing whether a party other than the original component manufacturer can fall within the relevant producer concept.

12. Case 8 — Sanofi Pasteur, C-338/24

Court

CJEU

Judgment

26 March 2026

This recent case concerned the relationship between the EU defective-product regime and national fault-based liability, as well as limitation periods and access to court.

The Court confirmed that the harmonised defective-product regime does not necessarily eliminate national liability systems based on a different foundation, such as fault. (Infocuria)

Relevance

This is particularly useful for autonomous greenhouse disputes.

A farmer might argue:

“The AI system was defective.”

But alternatively:

“The supplier was negligent because it knew about repeated failures and failed to install an available safety update.”

Those are conceptually different bases of liability.

13. AI-Specific Liability Problem

An autonomous greenhouse differs from an ordinary machine because its behaviour may depend on:

machine learning;

changing environmental data;

software updates;

remote configuration;

cloud services;

interconnected sensors;

cybersecurity;

autonomous decision-making.

The European Commission has specifically identified autonomy, opacity, connectivity, data dependency, continuous adaptation and limited predictability as challenges for traditional liability regimes. (Eur-Lex)

14. Defective Hardware

The simplest situation is:

Irrigation valve physically breaks.

If the valve was defective and caused damage, traditional product-liability principles may apply.

Possible evidence:

manufacturing records;

inspection;

engineering reports;

failed component;

product specifications;

testing records.

15. Defective Software

More difficult:

Hardware works perfectly, but AI software incorrectly shuts down irrigation.

Questions include:

Is the software part of the product?

Was the software defective?

Was the software properly updated?

Was the error foreseeable?

Did the defect exist when supplied?

Did a later update cause the problem?

Was the software supplied by a separate provider?

Modern EU product-liability rules are particularly important here because the EU has moved toward expressly addressing software and digital components in the product-liability framework.

16. AI Hallucination and Greenhouse Failure

Suppose the AI system interprets sensor information incorrectly:

Sensor says humidity = 80%.

AI incorrectly interprets it as:

“Humidity safe; no ventilation necessary.”

The resulting humidity damages the crop.

The legal question is not whether the AI was “intelligent.”

The important questions are:

Was the system designed appropriately?

Was the sensor functioning correctly?

Was the software reasonably safe?

Was there an adequate warning?

Was the system appropriately monitored?

Was an update available?

Was the operator trained?

Was the greenhouse used as reasonably expected?

17. Sensor Failure

Autonomous greenhouse systems are heavily dependent on sensors.

Examples:

temperature sensor;

soil-moisture sensor;

humidity sensor;

CO₂ sensor;

light sensor;

nutrient sensor.

A faulty sensor can cause a chain reaction:

Sensor error → AI error → actuator error → crop damage

This creates a multi-causal liability problem.

18. Multi-Actor Liability

Consider:

The court may have to determine whether:

one actor caused the loss;

multiple actors contributed;

the farmer was contributorily negligent;

liability should be apportioned under national law.

19. Cyberattack and Autonomous Greenhouse Failure

Cybersecurity is particularly important.

Suppose:

Hacker enters greenhouse network → changes temperature settings → ventilation disabled → crop destroyed.

Potential liability questions include:

Was the software adequately secured?

Was the vulnerability known?

Were security updates installed?

Did the manufacturer provide security patches?

Was the farmer negligent in maintaining security?

Did a third party cause the damage?

Was the cyberattack foreseeable?

Modern product-liability policy discussions specifically identify cybersecurity weaknesses as a challenge for product liability. (Eur-Lex)

20. Software Update Liability

Imagine:

Manufacturer releases critical security update.

Farmer does not install it.

Two months later:

cyberattack → irrigation system fails → crops destroyed.

Potential arguments include:

Farmer's argument

“The original product was unsafe.”

Manufacturer's argument

“The necessary update was provided.”

Court's questions

Was the update communicated properly?

Was installation automatic?

Was installation technically possible?

Was the farmer warned?

Was the update necessary for safety?

Did the manufacturer continue to owe monitoring duties?

Autonomous systems make the traditional concept of “putting the product into circulation” more complicated because the system can materially change after sale.

The European Commission has specifically identified this issue in connection with self-learning and update-dependent products. (Eur-Lex)

21. Contractual Liability

A greenhouse supplier may contractually promise:

“The system will automatically maintain temperature between 20°C and 25°C.”

If the system repeatedly fails, the farmer may have a contractual claim.

Possible contractual terms include:

performance guarantees;

uptime;

maintenance obligations;

service-level agreements;

software updates;

response times;

warranties;

exclusions;

limitation of liability.

Contract law may therefore provide a separate route from strict product liability.

22. Negligence

A fault-based claim may arise where an actor failed to exercise reasonable care.

Examples:

Manufacturer

Failed to test the AI under realistic conditions.

Installer

Configured the sensors incorrectly.

Maintenance provider

Failed to respond to repeated alarms.

Operator

Ignored warnings and deliberately disabled safety controls.

The applicable national law determines the precise negligence standard.

23. Duty to Warn

A manufacturer may need to provide appropriate warnings concerning:

extreme temperatures;

sensor limitations;

network failure;

AI uncertainty;

maintenance;

emergency shutdown;

software updates.

A warning such as:

“System may occasionally malfunction”

may not necessarily be adequate where a known failure can destroy an entire crop.

The adequacy of the warning must be evaluated under the applicable law and circumstances.

24. Foreseeability

Foreseeability is particularly important.

Suppose an AI system has repeatedly experienced:

overheating → ventilation failure.

If the manufacturer continues deploying the same system without correction, a future failure may become increasingly foreseeable.

This can affect:

negligence;

defect assessment;

causation;

contractual obligations;

maintenance duties.

25. Burden of Proof

Traditional product liability generally requires the injured person to prove:

Damage + Defect + Causation

The difficulty with autonomous systems is that the farmer may not know:

what the AI actually did.

The system may contain:

proprietary algorithms;

encrypted logs;

remote servers;

machine-learning models;

inaccessible telemetry.

This creates an information asymmetry between the victim and manufacturer.

EU policy work has recognised that AI's opacity and complexity can make it difficult for injured persons to obtain the evidence necessary to establish liability. (Eur-Lex)

26. Causation in Autonomous Greenhouse Cases

Consider:

Crop destroyed.

Possible causes:

AI error;

sensor error;

power failure;

extreme weather;

poor seed quality;

pest infestation;

operator error;

irrigation defect.

The claimant must connect the legally relevant defect or wrongful conduct to the damage.

A technical expert may need to reconstruct:

input → algorithm → decision → actuator → environmental condition → crop damage

27. Expert Evidence

Expert evidence can be crucial.

Experts may examine:

Software

algorithm design;

update history;

logs;

decision rules.

Hardware

sensor accuracy;

controller operation;

valve operation.

Agriculture

crop sensitivity;

expected yield;

temperature thresholds.

Cybersecurity

intrusion;

vulnerability;

patching.

Economics

lost crop value;

lost profits;

replacement costs.

28. Environmental Liability

An autonomous greenhouse could also cause environmental damage.

For example:

AI irrigation algorithm fails → water continuously pumped → reservoir depleted.

Or:

Fertiliser-control algorithm malfunctions → excessive nutrients discharged into nearby water.

Or:

Chemical application system incorrectly activates → contamination occurs.

Where EU environmental-liability legislation applies, environmental remediation and prevention mechanisms may become relevant.

However, environmental liability should not automatically be equated with private compensation for crop losses. Different statutory regimes can protect different interests.

29. State Liability

A further scenario is a government-operated autonomous agricultural facility.

If a public authority deploys an automated greenhouse system and an unlawful administrative decision or failure causes damage, questions of State liability may arise under applicable EU and national law.

But ordinary malfunction of a private greenhouse does not automatically become State liability.

30. Contributory Negligence

Suppose:

AI detects dangerous temperature → warning issued → farmer manually disables alarm → crop destroyed.

The farmer's conduct may affect liability.

Potential questions include:

Did the farmer act reasonably?

Was the warning sufficiently clear?

Was the alarm itself defective?

Was the farmer trained?

Was emergency intervention reasonably possible?

The exact effect depends upon the applicable national law.

31. Force Majeure

Extreme circumstances can complicate liability.

Examples:

unprecedented storm;

lightning strike;

extraordinary power-grid failure;

natural disaster;

major cyberattack.

But:

“Unexpected event”

does not automatically equal force majeure.

The applicable contractual and national civil-law rules determine whether the event qualifies and what consequences follow.

32. Autonomous Greenhouse and Product Recall

Suppose a manufacturer discovers:

AI controller has a dangerous defect.

The manufacturer may need to:

notify customers;

stop distribution;

provide updates;

recall affected systems;

replace defective components;

provide warnings.

This is particularly important where the defect can cause physical injury or significant property damage.

The EU's modern product-safety framework increasingly emphasises product safety throughout the product's life cycle.

33. Liability of the AI Developer

The AI developer may potentially be relevant where:

the software itself is defective;

the developer supplied incorrect specifications;

the developer failed to address known vulnerabilities;

an update caused the malfunction;

the software was unsuitable for the intended task.

But liability does not automatically transfer to the AI developer merely because AI was involved.

The claimant must identify an applicable legal basis and satisfy its requirements.

34. Liability of the System Integrator

The integrator can be particularly important.

Suppose:

Sensor A + AI B + irrigation controller C

are combined by Company D.

Company D configures the components incorrectly.

The individual components may all function correctly.

The complete system nevertheless fails.

This creates a potential integration defect.

35. Liability of the Farmer/Operator

The farmer is not automatically protected simply because the system is autonomous.

Potential operator failures include:

ignoring alarms;

failing to maintain equipment;

disabling safety mechanisms;

refusing critical updates;

operating outside specifications;

modifying software without authorisation.

But the operator should not automatically bear liability merely because a machine-learning system behaves unpredictably.

36. Autonomous Greenhouse Example

Facts

A commercial greenhouse uses an AI system to control:

temperature;

humidity;

irrigation;

lighting.

At 2:00 a.m.:

Temperature sensor reports 21°C instead of the actual 39°C.

AI therefore keeps ventilation closed.

At 6:00 a.m.:

temperature reaches 48°C.

The entire tomato crop is destroyed.

Legal analysis

Step 1 — Damage

Crop worth €500,000 is destroyed.

Step 2 — Defect

The sensor or control system may have failed to provide expected safety.

Step 3 — Causation

Sensor failure → incorrect AI decision → ventilation failure → crop destruction.

Step 4 — Responsible actor

Possible defendants:

sensor manufacturer;

AI provider;

system integrator;

greenhouse supplier.

Step 5 — Contract

The supplier may have guaranteed automatic temperature control.

Step 6 — Evidence

Relevant evidence includes:

sensor logs;

AI decision logs;

temperature records;

maintenance records;

software versions.

Step 7 — Defences

Potential arguments include:

misuse;

failure to maintain;

unforeseeable event;

third-party interference;

force majeure.

Step 8 — Remedy

Possible remedies depend on the applicable law and may include:

repair;

replacement;

damages;

contractual compensation;

crop-loss compensation.

37. Multi-Layer Liability Model

A useful model is:

Layer 1 — Hardware

Was the physical component defective?

Layer 2 — Software

Was the software defective?

Layer 3 — Data

Were the sensor inputs accurate?

Layer 4 — AI

Did the AI make an unreasonable or unsafe decision?

Layer 5 — Integration

Were components properly connected?

Layer 6 — Operation

Was the system properly used?

Layer 7 — Maintenance

Were updates and repairs properly performed?

Layer 8 — Causation

Did the failure cause the claimed loss?

38. Case-Law Summary

39. Important Legal Principles

1. Automation does not eliminate product liability

A product does not become legally risk-free because software controls it.

2. AI can become part of the defect

A malfunctioning algorithm can be relevant to whether the overall system provides the safety legitimately expected.

3. The complete system matters

Hardware + software + sensors + updates may need to be examined together.

4. Causation is critical

The claimant must establish a legally sufficient connection between the defect/wrongful conduct and the damage.

5. Evidence is a major problem

AI opacity can make it difficult for farmers to reconstruct what happened.

6. Multiple actors can contribute

Manufacturer, developer, integrator, installer and operator may all have different roles.

7. Contract and tort may coexist with product liability

The product-liability regime does not necessarily eliminate other liability bases. (Curia)

8. Updates create continuing safety questions

Autonomous products may change after being placed on the market.

9. Cybersecurity can become a safety issue

A remotely controlled greenhouse can suffer physical damage from a digital attack.

10. Agriculture-specific losses require careful proof

Crop value and lost profits must be established under the applicable national law.

40. Future Legal Issue: Self-Learning Greenhouses

The most difficult future scenario is a self-learning greenhouse.

Imagine:

AI initially operates safely → learns from thousands of growing cycles → modifies irrigation strategy → gradually develops an unsafe pattern → crop fails.

The manufacturer may argue:

“The system was safe when sold.”

The farmer may argue:

“The system became unsafe through its autonomous operation.”

This creates a fundamental modern product-liability question:

When an autonomous system changes its own behaviour after deployment, when does the resulting unsafe behaviour become a product defect, a maintenance issue, an operator issue, or a new event of liability?

The European Commission has specifically recognised that self-learning and post-market modification create difficulties for traditional concepts of product safety and liability. (Eur-Lex)

41. Examination-Friendly Formula

For an exam, remember:

Autonomous Greenhouse Liability =

**Defect

Damage

Causation

Product/Software

AI Autonomy

Evidence

Contract

Negligence

Operator Conduct

Remedy**

Key words

Defective product

Software defect

Sensor failure

AI malfunction

Causation

Strict liability

Fault

Contributory negligence

Updates

Cybersecurity

42. Conclusion

Autonomous greenhouse failure represents a particularly interesting application of modern European civil liability because a single agricultural loss can result from the interaction of hardware, software, AI, sensors, data, connectivity and human operation.

The established European product-liability cases remain highly relevant. Boston Scientific provides an important safety-expectation framework; O'Byrne addresses the producer and circulation concepts; W and Others demonstrates how defect and causation can be approached when direct scientific proof is difficult; González Sánchez confirms the possibility of different liability regimes operating on different legal bases; and Ford Italia is relevant to identifying the legally responsible producer. (Infocuria)

For autonomous greenhouse systems, the central legal difficulty is causal attribution:

Sensor failure → AI interpretation → automated decision → physical system response → agricultural damage.

European law increasingly recognises that AI's autonomy, opacity, connectivity, continuous adaptation and dependence on software updates create difficulties that traditional liability rules must address. (Eur-Lex)

Thus, the most important future question is not simply “Who made the greenhouse?”, but rather:

Which legally responsible actor controlled, designed, supplied, integrated, updated or failed to maintain the autonomous system whose decision ultimately caused the crop, property, personal or environmental damage?