Agentic Enterprises and Distributed Autonomous Production in Europe

1. Meaning

Agentic enterprises are businesses in which software agents or AI systems can independently perform business functions that were traditionally carried out by employees or managers.

An agent may be authorised to:

identify suppliers;

negotiate prices;

place purchase orders;

allocate production capacity;

schedule machines;

monitor inventory;

detect defects;

select logistics providers;

respond to changing demand;

modify production schedules;

contract with other businesses;

initiate payments;

coordinate several autonomous production units.

Distributed autonomous production goes one step further. Instead of one centrally controlled factory, production is distributed among:

robotic factories;

3D-printing facilities;

cloud-controlled manufacturing systems;

independent suppliers;

automated warehouses;

autonomous logistics systems;

micro-factories;

software-controlled production cells.

The legal difficulty is that the enterprise's decisions may be made by multiple autonomous agents rather than one identifiable human decision-maker.

There is presently no single European legal doctrine called “agentic enterprise liability.” The existing law must therefore be applied across contract, agency, product liability, AI regulation, data protection, competition law and general civil liability.

2. Basic example

Consider an enterprise called EuroManufacture AI.

It operates 500 autonomous production units across Europe.

Its central AI system:

forecasts demand;

selects suppliers;

negotiates procurement contracts;

allocates raw materials;

sends instructions to robotic factories;

detects production defects;

changes production parameters;

selects logistics providers.

Suppose the AI decides to purchase a defective component from Supplier A because Supplier A is cheaper.

The component is incorporated into 50,000 products.

The products are distributed throughout Europe.

Later, the component fails and causes personal injury.

The legal questions become:

Who is responsible?

the enterprise?

the AI-system provider?

the software developer?

the component manufacturer?

the autonomous factory?

the supplier?

the logistics operator?

several parties jointly?

This is the central problem of distributed autonomous production liability.

3. The European legal framework

A. Contract law

Agentic enterprises may conclude contracts automatically.

The important issues include:

authority;

formation of contract;

mistake;

misrepresentation;

unauthorised transactions;

automated negotiation;

electronic signatures;

attribution of an agent's conduct to the enterprise;

contractual allocation of AI risk.

An AI agent does not ordinarily become the legal person itself.

The starting question is:

Who authorised and controlled the system?

4. AI Act

The EU AI Act provides the regulatory framework for AI systems, including obligations concerning risk management, transparency, human oversight and other requirements depending on the AI system's classification.

For an agentic enterprise, the crucial distinction is between:

AI system as a tool

and

AI system performing operational functions affecting safety, workers, consumers or production.

The more autonomous the production system becomes, the more important it is to identify:

human oversight;

monitoring;

technical documentation;

logging;

cybersecurity;

risk management;

foreseeable misuse;

post-market monitoring.

5. Product Liability Directive 2024/2853

This is particularly important.

The new Product Liability Directive expressly adapts European product liability to digital technologies.

It treats software, including AI systems, as products for the purposes of the Directive. It also addresses software updates, upgrades and AI-related defects. (Eur-Lex)

The Directive applies to products placed on the market or put into service after 8 December 2026. (Eur-Lex)

It is especially significant for distributed production because a defective product may result from interaction between:

hardware + software + AI model + update + component + production instructions.

The Directive also provides for circumstances in which multiple economic operators can be jointly and severally liable for the same damage. (Eur-Lex)

6. Digital manufacturing files

An especially important provision concerns digital manufacturing files.

A defective CAD or other digital manufacturing file that controls machinery such as a 3D printer can fall within the new product-liability framework.

Thus:

defective digital design → autonomous manufacturing → defective physical product → injury

can potentially produce product liability.

The Directive expressly recognises digital manufacturing files containing functional information necessary to produce tangible objects through automated machinery. (Eur-Lex)

This is highly relevant to distributed autonomous manufacturing.

7. Case Law

Case 1 — Veedfald v Århus Amtskommune

C-203/99

This is an important foundational product-liability decision.

A defective product was used during the provision of medical services. The Court interpreted the Product Liability Directive and rejected an overly narrow understanding of when a product is placed into circulation.

The case demonstrates that product liability can apply even in circumstances where the product is used within an integrated service environment. (curia)

Relevance to agentic production

An autonomous production enterprise may argue:

“The defective component was never separately sold to the injured person.”

That does not necessarily end the analysis.

The relevant question is whether the product falls within the product-liability regime and whether the statutory conditions are satisfied.

Principle

Integration of a product into a wider automated service or production process does not automatically eliminate product liability.

8. Case 2 — O'Byrne v Sanofi Pasteur

C-127/04

O'Byrne concerned the meaning of putting a product into circulation and the relationship between a producer and its wholly owned subsidiary.

The Court held that a product is put into circulation when it leaves the manufacturing process operated by the producer and enters a marketing process in the form in which it is offered to the public. The case also considered the significance of relationships within a distribution structure. (Infocuria)

Relevance

Imagine:

Parent AI enterprise → autonomous subsidiary → robotic factory → distributor.

The enterprise cannot simply assume that adding autonomous subsidiaries or intermediate entities automatically removes responsibility.

Corporate and distribution structures must be analysed according to the applicable liability rules.

Principle

Organisational separation and distribution architecture do not automatically determine who is legally responsible.

9. Case 3 — Boston Scientific Medizintechnik

Joined Cases C-503/13 and C-504/13

This is one of the most important authorities for autonomous production systems.

The Court considered medical devices where a significant number of products in the same series could potentially have a defect.

It held that products belonging to the same group or production series can be regarded as defective where they share the same potential safety deficiency, even if the particular individual product has not yet been shown to malfunction. (curia)

Relevance to distributed manufacturing

Suppose 200 autonomous factories use the same AI production model.

A safety defect is discovered in the model.

The enterprise might have to investigate:

Are all products produced under that model potentially affected?

This creates a major distinction between:

individual-product failure

and

systemic production defect.

Principle

Where a common production characteristic creates a significant safety risk, liability analysis may extend beyond the single product that has already failed.

10. Case 4 — Sanofi Pasteur

C-621/15

In Sanofi Pasteur, the Court considered proof of product defect and causation where scientific consensus was lacking.

The Court accepted that, subject to the conditions established by EU law, serious, specific and consistent evidence could potentially establish defect and causal connection. (curia)

Relevance to AI production

Autonomous manufacturing may create causation problems.

For example:

AI production instruction → unusual manufacturing parameter → microscopic defect → product failure → injury.

The injured person may not have direct access to the AI model or manufacturing logs.

Therefore, evidence may have to be constructed from:

production records;

system logs;

model outputs;

maintenance records;

defect patterns;

manufacturing history.

Principle

Complex technology does not eliminate the need to establish defect and causation, but courts must apply the applicable evidentiary rules realistically.

11. Case 5 — Boston Scientific and distributed defect detection

The importance of Boston Scientific becomes even greater when production is distributed.

Imagine:

Factory A operates in Germany;

Factory B in France;

Factory C in Italy;

Factory D in Spain.

All use the same autonomous production model.

A defect is discovered in Factory C.

The legal question is not merely:

“Did Factory C produce a defective item?”

It may also become:

“Does the common design, software, component or production process create a systematic safety risk throughout the network?”

That is precisely the type of systemic risk that the Boston Scientific reasoning can help analyse.

12. Case 6 — SCHUFA Holding

C-634/21

Although SCHUFA is a GDPR case rather than a manufacturing case, it is highly relevant to agentic enterprises.

The Court considered automated scoring and the use of automatically generated probability values in decision-making. (Infocuria)

Application to agentic enterprises

An agentic enterprise may generate scores for:

suppliers;

employees;

customers;

machines;

factories;

logistics operators.

For example:

Supplier A = 94% reliability
Supplier B = 72% reliability

The autonomous procurement agent automatically chooses Supplier A.

The legal question becomes whether the score is merely an internal operational tool or becomes part of an automated decision producing legally significant consequences for an individual.

Principle

Automated scoring can itself become legally significant where it determines subsequent decisions.

13. Case 7 — Dun & Bradstreet Austria

C-203/22

This is particularly relevant to agentic enterprises because the Court's 27 February 2025 judgment addressed meaningful information concerning automated decision-making and profiling.

The Court considered how a person can understand and challenge an automated decision, while also considering trade-secret limitations. (curia)

Relevance

Suppose an autonomous enterprise refuses to contract with a supplier because its agent generates a low reliability score.

The supplier asks:

“Why was I rejected?”

The enterprise cannot necessarily answer merely:

“Our AI decided.”

The applicable transparency and data-protection rules may require meaningful information, subject to the statutory protection of other interests such as trade secrets.

Principle

Automation does not automatically remove accountability or explanation requirements.

14. Case 8 — Sanofi Pasteur, C-338/24

The Court's 26 March 2026 judgment in Sanofi Pasteur, C-338/24 concerned the EU Product Liability Directive framework, including limitation and the relationship between product-liability rules and national fault-based liability. (curia)

Relevance

Distributed autonomous production frequently creates several potential causes of action.

For example:

Product liability

plus

contractual liability

plus

national fault-based civil liability.

The existence of product-liability legislation does not necessarily eliminate every other potentially applicable national claim.

15. Case-law table

These are analogical authorities rather than cases specifically deciding liability for autonomous factories. No established CJEU doctrine currently treats an AI agent as an independent legal person.

16. Who is responsible when an AI agent makes the decision?

A useful hierarchy is:

Level 1 — Enterprise

Usually the starting point.

The enterprise:

deploys the system;

defines its objectives;

provides resources;

benefits economically;

controls the production network.

Level 2 — AI provider

Potentially relevant where:

the AI itself is defective;

the provider supplied defective software;

inadequate updates caused the problem;

contractual obligations were breached.

The new Product Liability Directive expressly treats software, including AI systems, as products. (Eur-Lex)

Level 3 — Component manufacturer

If the autonomous system incorporates a defective component, the component manufacturer may have separate liability.

Level 4 — Integrator

A company integrating AI into machinery may bear responsibility where the integration itself creates the defect.

Level 5 — Operator

Human operators can potentially be responsible under applicable national law where negligent intervention, maintenance or supervision contributes to the harm.

17. The AI itself is generally not the legal person

A critical principle is:

Autonomy is not the same thing as legal personality.

An AI agent can:

negotiate;

order;

calculate;

produce;

schedule;

communicate;

optimise.

But that does not automatically make the AI itself the bearer of civil liability.

The law therefore generally looks through the autonomous system toward the relevant economic operator, manufacturer, provider, controller, contractor or other legally responsible person.

18. Distributed production and joint liability

This is one of the most difficult problems.

Suppose:

AI developer
↓
AI enterprise
↓
autonomous factory
↓
component manufacturer
↓
robot manufacturer
↓
logistics operator

All contribute to the final product.

The new Product Liability Directive expressly contains rules concerning multiple economic operators and joint and several liability where the statutory requirements are met. (Eur-Lex)

This is important because the victim should not necessarily have to identify one single software line responsible for a complex technological failure.

19. Software updates create continuing responsibility

Traditional manufacturing assumes:

Product manufactured → product sold → manufacturer's control decreases.

Autonomous production is different.

The software can continue changing after deployment.

For example:

Version 1.0 → Version 1.1 → machine-learning update → new production behaviour.

The new Product Liability Directive expressly addresses defects arising from software updates, upgrades and related services within the manufacturer's control. (Eur-Lex)

Therefore:

post-market software intervention can remain legally relevant.

20. Autonomous production and cybersecurity

An agentic enterprise can be attacked.

Suppose an attacker changes:

AI production parameter = 10 mm

to:

AI production parameter = 8 mm.

Thousands of products are then manufactured incorrectly.

Questions include:

Was the AI system adequately secured?

Was the enterprise negligent?

Was the software defective?

Was there an inadequate update?

Was the attack reasonably foreseeable?

Did a third-party cyberattack break causation?

Who had control over security?

The new Product Liability Directive specifically recognises cybersecurity vulnerabilities and software updates as relevant to continuing product safety. (Eur-Lex)

21. Autonomous procurement

An agent can automatically enter procurement arrangements.

Example:

Agent identifies cheapest supplier → negotiates → accepts terms → places order.

Legal issues include:

Authority

Was the agent authorised to conclude the contract?

Scope

Did it exceed its purchasing limits?

Error

Did the agent select the wrong supplier?

Misrepresentation

Did the agent communicate incorrect information?

Contract terms

Which version of the terms was accepted?

Evidence

Can the parties reconstruct exactly what happened?

This makes audit logs and machine-readable contractual records extremely important.

22. Distributed production and contract allocation

A sophisticated enterprise may allocate risk contractually.

For example:

AI provider accepts software defects.

Factory accepts manufacturing defects.

Component supplier accepts component defects.

Enterprise accepts integration defects.

But contractual allocation does not necessarily eliminate statutory liability owed to consumers or other protected persons.

The parties may subsequently pursue contribution or recourse between themselves.

23. Autonomous agents and competition law

Agentic enterprises can also create competition concerns.

Suppose autonomous procurement agents communicate with thousands of suppliers.

They may independently:

observe competitor prices;

alter purchasing;

reduce orders;

coordinate inventory;

choose exclusive suppliers.

If independent enterprises use autonomous agents in ways that result in coordination, Article 101 TFEU may become relevant.

If a dominant enterprise uses autonomous agents to exclude rivals, Article 102 may become relevant.

Potential conduct includes:

discriminatory supplier access;

exclusive dealing;

self-preferencing;

refusal to deal;

algorithmic foreclosure;

predatory pricing;

discriminatory procurement.

Thus:

Autonomous behaviour does not create immunity from competition law.

24. Autonomous production and labour law

Distributed autonomous factories may dramatically reduce direct human involvement.

Nevertheless, workers may remain involved in:

supervision;

maintenance;

quality control;

emergency intervention;

system training;

warehouse operations.

Potential issues include:

algorithmic work allocation;

automated performance assessment;

workplace surveillance;

occupational safety;

discrimination;

consultation rights;

responsibility for machine-related accidents.

The fact that a machine made the immediate decision does not necessarily eliminate the employer's obligations.

25. Evidence and black-box production

A major litigation problem is the black-box problem.

Imagine:

20 AI agents
↓
100 factories
↓
10,000 production decisions per minute.

After an accident, the claimant asks:

“Which decision caused the defect?”

The enterprise may have to reconstruct:

input data;

model version;

agent objective;

instructions;

supplier information;

production parameters;

human interventions;

software updates;

machine logs;

final product characteristics.

This makes traceability a central element of modern autonomous production governance.

26. Causation

Causation can be represented as:

AI decision

↓

production instruction

↓

machine execution

↓

physical defect

↓

product failure

↓

injury

↓

financial loss

Every link may be contested.

For example, the enterprise may argue:

“The AI recommendation was correct; the machine malfunctioned.”

The machine manufacturer may argue:

“The machine operated correctly; the AI supplied an incorrect command.”

The AI provider may argue:

“The model was modified by the enterprise.”

The component manufacturer may argue:

“The component was damaged during integration.”

Distributed autonomy therefore creates distributed causation.

27. Counterfactual analysis

Courts may need to ask:

What would have happened if the autonomous system had operated correctly?

Possible counterfactuals include:

correct supplier selection;

correct machine parameter;

correct software update;

human approval;

alternative component;

manual production.

The counterfactual helps identify whether the alleged defect actually caused the damage.

28. Hypothetical case

Facts

EuroAgent Manufacturing SE operates autonomous factories throughout Europe.

Its AI agent selects components based upon:

price;

reliability;

delivery time;

predicted failure rate.

The agent chooses Supplier X.

Supplier X's component contains a microscopic defect.

The component is incorporated into 100,000 autonomous products.

Six months later, 2,000 products fail.

Possible claims

Against Supplier X

Defective component.

Against EuroAgent

Defective integration, inadequate supervision or defective final product.

Against AI provider

Potential software/product-liability issues if the AI itself was defective.

Against machine manufacturer

Potential machinery defect.

Against logistics provider

Only if transportation contributed to the damage.

Evidence

AI logs;

supplier-selection history;

component specifications;

model version;

production logs;

quality-control records;

software-update records.

29. Important distinction: autonomy vs responsibility

30. Practical legal test

For an agentic enterprise, a court or regulator can analyse the problem through ten questions:

1. Who owns the enterprise?

Identify the legal person.

2. Who deployed the AI agent?

Enterprise, software provider or another operator?

3. What authority was given to the agent?

Procurement only, or complete contractual and production authority?

4. What did the agent actually do?

Separate prediction from actual execution.

5. Who controlled the agent?

Control is important for allocating responsibility.

6. Was the AI/software defective?

Consider software errors, design defects and inadequate updates.

7. Was the physical product defective?

Apply the applicable product-liability regime.

8. Were multiple operators involved?

Identify component manufacturers, integrators and providers.

9. Can causation be established?

Connect algorithmic action to physical damage.

10. Are there additional claims?

Consider:

contract;

tort/delict;

GDPR;

AI Act;

product liability;

competition law;

consumer protection.

31. Direct versus analogical case law

At present, there is no established CJEU judgment specifically holding an “agentic enterprise” liable for decisions made by autonomous AI production agents.

Therefore:

Strongest direct product-liability foundations

Veedfald

O'Byrne

Boston Scientific

Sanofi Pasteur

Strongest automated-decision analogies

SCHUFA

Dun & Bradstreet

These cases should not be cited as if they already decided the precise question of autonomous AI factories.

Their value is that they provide established principles which can be extended to the emerging technology.

32. Future direction of European law

The new Product Liability Directive is particularly significant because European legislation is moving away from a purely physical concept of a “product.”

The framework expressly recognises:

software;

AI systems;

software updates;

upgrades;

digital manufacturing files;

interconnected services;

multiple economic operators. (Eur-Lex)

This means that future litigation is likely to focus less on:

“Was the machine defective?”

and increasingly on:

“Which combination of software, model, component, update, data, instruction and physical process produced the defect?”

33. Conclusion

Agentic enterprises and distributed autonomous production fundamentally change the traditional structure of civil liability.

The enterprise may no longer have a simple chain:

Human manager → factory → product.

Instead, the chain may be:

AI agent → autonomous procurement → autonomous supplier selection → distributed factory → robotic production → AI quality control → autonomous logistics → consumer.

European law does not presently treat this entire network as a separate legal category. Instead, liability is allocated through existing doctrines concerning producers, economic operators, contracts, defective products, automated decisions, data processing and causation.

The most important development is the new Product Liability Directive 2024/2853: it expressly brings software and AI systems within the product-liability concept and recognises software updates, digital manufacturing files and multiple potentially liable economic operators. (Eur-Lex)

Exam Keywords

Agentic enterprise — autonomous enterprise — AI agents — distributed manufacturing — autonomous factory — robotic production — machine-to-machine contracting — algorithmic procurement — AI attribution — legal agency — product liability — software defect — AI defect — digital manufacturing file — systemic defect — joint and several liability — causation — cybersecurity — software updates — human oversight — SCHUFA — Dun & Bradstreet — Boston Scientific — Veedfald — O'Byrne — Sanofi Pasteur.