WHO IS AT FAULT ON A CAPTAINLESS SHIP?

WHO IS AT FAULT ON A CAPTAINLESS SHIP?

Autonomous Maritime Trade, Artificial Intelligence, and the Reconstruction of Liability in 2043

A Future Law Analysis from the Law 2043 Perspective

Author: Abdulkadir Çiçek President of the Global Lawyers Association – Law 2043

Abstract

Assuming that, by 2043, a significant portion of the international merchant fleet is operated by highly or fully autonomous systems, one of the fundamental concepts of maritime commercial law—“fault”—will have to be redefined. Where the route of a merchant vessel is determined by artificial intelligence rather than a human captain, the AI evaluates technical failures aboard the vessel and risks along the intended route, makes a navigational decision, and the vessel subsequently sinks as a result of that decision, traditional models of liability will face a serious challenge.

This study examines the liability of the ship operator or carrier, the developer of the artificial intelligence system, and insurance companies under the current provisions of the Turkish Commercial Code, contemporary international regulations, and a 2043 perspective. The central thesis of this study is as follows: the transfer of decision-making authority to artificial intelligence by an actor benefiting economically from an activity should not result in the disappearance of legal responsibility. A legal order that permits such an outcome would create an “autonomy paradox,” in which legal responsibility diminishes as technological autonomy increases.

Accordingly, this study advances the concepts of algorithmic voyage-worthiness, the Principle of Non-Disappearance of Liability, the Autonomous Navigation Decision Ledger, and the Layered Liability Model within the framework of the Law 2043 vision.

Keywords: Artificial intelligence, autonomous ship, MASS, maritime commercial law, insurance law, algorithmic liability, carrier liability, autonomous systems, Law 2043.

⸻

I. 2043, 03:17

In 2043, M/V Nomos, carrying cargo from Mersin to the Far East, is a crewless and fully autonomous merchant vessel.

The vessel’s navigation, energy management, monitoring of machinery systems, meteorological assessments, and route optimization are performed by an artificial intelligence system called ARGOS.

ARGOS is not merely an automated system controlling the rudder. It simultaneously evaluates satellite data, maritime traffic, weather forecasts, the vessel’s fuel status, machinery sensors, port congestion, navigational restrictions, and commercial delivery deadlines in order to determine which route the vessel should take.

During the voyage, the system detects a loss of performance in the cooling unit.

At the same time, meteorological conditions along the route ahead are deteriorating, and strong currents are present.

ARGOS has two options.

The first option is to alter the vessel’s course and take a longer but safer route.

The second option is to continue along the existing route, which preserves the delivery deadline and is shorter, but carries a significantly higher risk in light of the vessel’s current technical condition.

The artificial intelligence selects the second option.

At 03:17, the problem in the machinery system worsens. The vessel’s manoeuvring capacity decreases. Driven out of the safe navigation corridor by the current, the vessel grounds on submerged rocks, its hull is breached, and it sinks within several hours.

The vessel is lost.

Cargo worth hundreds of millions of dollars sinks beneath the sea.

Environmental damage occurs.

And one of the oldest questions in legal history is asked again in an entirely new form:

Who is at fault?

The vessel?

The artificial intelligence?

The company that developed the artificial intelligence?

The company operating the vessel?

The insurance company?

Or, in this new world in which the human being has been removed from the immediate decision-making process, is the law relying on the wrong concepts to define responsibility?

This will be one of the fundamental legal questions of 2043.

⸻

II. WHERE THE LAW STANDS TODAY: 2043 IS NO LONGER SCIENCE FICTION

Although this scenario may appear distant today, the legal transformation has already begun.

The International Maritime Organization (IMO) adopted the Maritime Autonomous Surface Ships – MASS Code in May 2026, and the Code entered into effect on 1 July 2026. Although not yet binding, the Code establishes a broad safety framework covering matters ranging from the navigation of autonomous merchant vessels and connectivity systems to remote operations centres and cybersecurity. According to the IMO’s current roadmap, the binding MASS Code is intended to be adopted by 2030 and to enter into force on 1 January 2032. (International Maritime Organization)

Accordingly, by 2043, international maritime law may already have accumulated at least a decade of experience with binding regulation of autonomous vessels.

Nevertheless, the present approach remains substantially human-centred.

The 2026 MASS Code preserves the principle that a human master remains ultimately responsible for the vessel, even where the ship is highly autonomous. The master does not necessarily have to be physically present on board and may perform the role remotely. IMO work also recognises that the master should have the ability to intervene in the system where necessary. (International Maritime Organization)

This approach is understandable during a transitional period.

However, in a genuinely fully autonomous system, the following question becomes unavoidable:

How sustainable is it to continue holding a human being responsible for an incident in which that human being was not the actual decision-maker, merely because the law needs someone to hold responsible?

This is where the legal problem of 2043 begins.

⸻

III. HOW WOULD THE TURKISH COMMERCIAL CODE VIEW THIS ACCIDENT TODAY?

The existing provisions of the Turkish Commercial Code contain important principles that may form the starting point for future debate.

Under Article 1141 of the Turkish Commercial Code, the carrier is required to ensure that the vessel is seaworthy, voyage-worthy, and cargo-worthy. As a rule, the carrier is liable for damage resulting from the vessel’s unfitness. However, the carrier may avoid liability if it can prove that the deficiency could not have been discovered before the beginning of the voyage despite the exercise of the care and diligence expected from a prudent carrier. (TBMM CDN)

Article 1178 of the Turkish Commercial Code also requires the carrier to exercise the care and diligence expected from a prudent carrier in relation to the carriage, preservation, and custody of the goods. (TBMM CDN)

For artificial intelligence, however, Article 1180 is even more interesting.

Under this provision, where the damage results from an act relating to the navigation or other technical management of the vessel, the carrier is liable only for its own fault. (TBMM CDN)

This is precisely where the problem will emerge in 2043.

If, instead of a human captain making an incorrect route choice, a fully autonomous artificial intelligence system makes the wrong navigational decision, how should that conduct be legally classified?

A navigational fault?

A technical management fault?

A case of unseaworthiness or unfitness for the voyage?

An organisational fault of the operating company?

Or a product defect attributable to the software manufacturer?

Existing legal categories can explain different parts of the same incident. Yet none of them, standing alone, fully captures the legal problem created by autonomous decision-making.

⸻

IV. THE AUTONOMY PARADOX

At this point, a concept that I consider particularly important from the perspective of Law 2043 emerges:

The Autonomy Paradox

Suppose a company can be held responsible for the decision of a human captain. Yet when that same company delegates the same decision entirely to an autonomous algorithm, it can argue:

“The company did not make this decision; the system did.”

If such a defence succeeds, an unacceptable result follows.

As technological autonomy increases, legal responsibility decreases.

In other words, by transferring decision-making authority to artificial intelligence, the company begins to externalise not only its operating costs, but also its legal risks.

The law should not permit this.

For that reason, one of the fundamental principles of future law should be:

The Principle of Non-Disappearance of Liability

The delegation of decision-making authority to an autonomous system within an economic activity must not, by itself, eliminate the liability toward third parties of the actors who derive economic benefit from that activity.

When a company removes the human captain and replaces that person with artificial intelligence, a “liability gap” should not arise in law.

Technology may change.

The decision-maker may change.

But there must continue to be a legally accountable bearer of the risk.

⸻

V. LIABILITY OF THE SHIPPING COMPANY

In the 2043 scenario, the first potentially liable actor is the company operating the vessel for commercial purposes.

Under Turkish law, a shipowner is the owner of a vessel who operates it for the purpose of obtaining economic benefit. Article 1062 of the Turkish Commercial Code currently regulates the shipowner’s liability for damage caused to third parties by crew members and certain pilots while performing their duties. (TBMM CDN)

In 2043, however, the question asked by the law should no longer be limited to:

“Was the crew member at fault?”

Instead, the law should ask:

“Was the autonomous decision-making system used in operating the vessel safe, and did the operator exercise the required level of care in using that system?”

If the company operating the vessel:

selected the AI system,

integrated it into the vessel,

determined the conditions under which it was permitted to make autonomous decisions,

was responsible for maintenance and software updates,

and derived economic benefit from the system’s operation,

then the company should not be removed from the chain of liability merely because the final route command was not issued by a human being.

⸻

VI. A NEW CONCEPT: ALGORITHMIC VOYAGE-WORTHINESS

To carry Article 1141 of the Turkish Commercial Code into the future, the law should require not only the physical vessel, but also the algorithmic system controlling it, to be fit for the voyage.

This may be described as algorithmic voyage-worthiness.

In 2043, it will not be sufficient for a vessel’s engine, hull, and navigation equipment merely to function properly in order for the vessel to be considered fit for the voyage.

Its autonomous system must also be:

up to date,

validated,

appropriate for the relevant navigation area,

sufficiently resilient against cyberattacks,

capable of detecting critical sensor failures,

capable of recognising its own safety limits,

and capable of entering a safe mode where necessary.

The current MASS Code already provides indications in this direction. It requires the conditions under which autonomous vessels may operate safely to be defined in advance; limitations such as wind, visibility, sea conditions, water depth, and weather must be taken into account; and the response of the system when it leaves its defined operational boundaries must be predetermined. Risk assessment, system resilience, and cybersecurity are also treated as central elements. (International Maritime Organization)

The law of 2043 should go one step further.

Under certain circumstances, the inadequacy of the artificial intelligence system itself should directly render the vessel legally unfit for the voyage.

In this way, software and vessel could no longer be separated from one another for legal purposes.

Because aboard an autonomous vessel, the algorithm is no longer an ordinary tool installed on the ship.

The algorithm is one of the vessel’s decision-making organs.

⸻

VII. CAN THE ARTIFICIAL INTELLIGENCE COMPANY ESCAPE LIABILITY?

The second actor is the company that developed the autonomous navigation system.

It would not be appropriate to hold the software developer responsible for every accident involving artificial intelligence.

For example, if the system was properly designed and correctly assessed the relevant risks, but the ship operator failed to maintain a critical sensor, responsibility may rest primarily with the operator.

By contrast, liability of the AI provider should arise where there is:

a design defect in the algorithm,

a defective model update,

failure to remedy a known security vulnerability,

incorrect acceptance of unreliable sensor data,

continued decision-making beyond the system’s defined safety limits,

or failure by the manufacturer to inform the operator of a critical system defect known to it.

The law is already slowly moving in this direction. The European Union’s new 2024 Product Liability Directive expressly recognises that software may constitute a “product” for the purposes of product liability. This regime does not directly resolve every commercial loss arising in maritime trade, but it is significant because it demonstrates that the law is moving away from treating software as something wholly separate from physical products. (EUR-Lex)

If, in 2043, the manufacturer of a ship’s engine can face liability for a defective engine, it would be difficult to justify entirely excluding from liability the producer of defective software that actually determines the vessel’s route.

⸻

VIII. WHAT ABOUT THE INSURANCE COMPANY?

An important distinction must be made here.

An insurance company is not automatically “at fault” merely because the accident occurred.

Its obligations arise primarily from the insurance contract.

Under Article 1409 of the Turkish Commercial Code, the insurer is obliged to compensate losses arising from the occurrence of a risk covered by the insurance contract. (TBMM CDN)

More importantly, under Article 1429, unless otherwise agreed, the insurer must also cover losses resulting from the negligence of the policyholder, the insured, and persons for whose acts they are legally responsible. A different regime applies in cases of intentional conduct. (TBMM CDN)

Accordingly, from the standpoint of legal policy, an insurer in the future should not simply be able to say:

“The accident was caused by artificial intelligence rather than a human being, so I will not pay.”

The real question is:

Is the risk created by the autonomous system covered by the policy?

By 2043, maritime insurance may need to expressly classify risks such as “AI-related navigational risk,” “algorithmic failure,” “model error,” and “cyber manipulation.”

Moreover, there will not be only one insurance relationship.

The physical loss of the vessel may trigger hull and machinery insurance;

the loss of cargo may trigger cargo insurance;

while damage to third parties and certain operational liabilities may trigger liability insurance.

Article 1473 of the Turkish Commercial Code regulates the insurer’s obligation under liability insurance to compensate the injured party, up to the contractually agreed amount, for losses arising from the insured’s legal liability. Article 1478 also allows the injured party, where the relevant conditions are satisfied, to bring a direct claim against the insurer. (TBMM CDN)

⸻

IX. THE FILE DOES NOT CLOSE AFTER THE INSURER PAYS

This may be where the most interesting part of the 2043 scenario begins.

After paying hundreds of millions of dollars in compensation, the insurer will ask:

Who actually caused the accident?

Under the principle of subrogation in Turkish law, an insurer that pays compensation may, to the extent of that payment, succeed to the insured’s rights against the responsible parties. Article 1472 regulates this mechanism in property insurance, while Article 1481 addresses it in liability insurance. (TBMM CDN)

The consequences of this will be extremely important in 2043.

Suppose an investigation establishes that:

the ARGOS system systematically underestimated risk levels in certain types of technical malfunction,

the manufacturer knew this before the accident,

but failed to release the necessary update.

In that case, the insurance companies that paid the losses may seek recovery from the AI manufacturer.

As a result, litigation concerning autonomous vessel accidents will no longer remain within the traditional “ship–cargo–insurance” triangle.

Maritime commercial law, software liability, product liability, cybersecurity law, and artificial intelligence law will converge in the same case file.

⸻

X. THE GREATEST PROBLEM: HOW WILL WE PROVE FAULT?

When reviewing the decision of a human captain, a court may ask:

Why did you choose this route?

Were you aware of the weather conditions?

Did you know about the machinery malfunction?

Why did you not turn back?

In 2043, those same questions will be directed toward artificial intelligence.

But artificial intelligence will not walk into the courtroom and testify.

Accordingly, before future law can solve the problem of liability, it must first solve a problem of evidence.

For this reason, my second institutional proposal from a Law 2043 perspective is:

The Autonomous Navigation Decision Ledger

Just as conventional vessels have Voyage Data Recorder systems, fully autonomous systems should be required to maintain a substantially more comprehensive AI Decision Ledger.

At critical decision points, such a system should preserve, in an immutable form, at least the following information:

* sensor data received by the artificial intelligence, * the model and software version in use, * detected technical failures, * meteorological and navigational risks, * alternative routes considered, * risk values calculated for each alternative, * the route selected, * the system’s confidence level, * the safety limits of the system, * whether human intervention occurred or was possible, * and new information that emerged after the decision.

After an accident, these records would effectively become the algorithm’s navigation logbook.

The court would no longer ask only:

“Why did the artificial intelligence make this decision?”

It could instead ask:

“Should a reasonably and safely designed autonomous system, possessing the same information at the relevant moment, have made this decision?”

In this way, alongside the traditional “prudent carrier” standard, a future standard of the “prudent autonomous system” may emerge.

⸻

XI. GRANTING LEGAL PERSONALITY TO ARTIFICIAL INTELLIGENCE IS NOT THE SOLUTION

At this point, there is an apparently easy but dangerous solution:

to recognise artificial intelligence as a separate legal person and assign the loss to it.

I do not consider this approach appropriate.

Imposing liability on an algorithm that possesses no meaningful pool of economic assets would, from the perspective of the injured party, often create liability that exists only in theory.

If a vessel causes USD 500 million in losses and the law closes the case by declaring:

“The artificial intelligence was at fault,”

then, in practical terms, nobody has actually been held accountable.

For this reason, the future of law should focus less on the independent legal personality of AI and more on the distribution of risk among the economic and technical actors who deploy and benefit from artificial intelligence.

The task of law is not to find a machine guilty.

Its task is to determine who created the risk, who controlled it, and who converted that risk into economic benefit.

⸻

XII. THE PROPOSED LIABILITY ARCHITECTURE FOR LAW 2043

A single liability model will not be sufficient for fully autonomous merchant vessels.

The structure I propose is a Layered Liability Model.

First Layer: Operational Liability

The shipping company that benefits economically from the operation of the vessel should not be permitted to leave the primary chain of liability toward third parties merely because it delegated decision-making authority to artificial intelligence.

Second Layer: Technological Liability

Where the damage results from the design, training, updating, or a known safety defect of the system, the AI developer or relevant technology provider should enter the chain of liability.

Third Layer: Distribution of Risk Through Insurance

Given the potentially enormous scale of losses, a special form of compulsory autonomous-system liability insurance could be considered for autonomous commercial systems above a specified risk threshold.

Fourth Layer: Algorithmic Voyage-Worthiness

The autonomous navigation system should be regarded as an integral component of the vessel’s legal fitness for the voyage.

Fifth Layer: Mandatory Decision Records

Immutable AI decision records should be maintained so that critical decisions can subsequently be reviewed.

The fundamental logic of this model is simple:

Liability should not disappear because artificial intelligence exists; it should be redistributed because artificial intelligence has entered the chain of decision-making.

⸻

XIII. WHO WOULD BE LIABLE IN THE M/V NOMOS CASE?

Let us return to the initial accident in 2043.

The court should first determine whether the sinking resulted from the technical malfunction, the incorrect route selection, or a combination of both.

Even today, the Turkish Commercial Code recognises that multiple causes may jointly contribute to a loss, and Article 1183 allows liability to be separated where causes combine. (TBMM CDN)

Accordingly:

If the shipping company failed to remedy serious maintenance problems reported by the system, the operator’s liability would increase.

If the AI was aware of the safer alternative but selected an unacceptably risky route solely to optimise the delivery deadline, and the system had been designed to behave in that manner, the liability of the AI developer would have to be examined.

If both the company’s maintenance negligence and the algorithm’s defective risk assessment contributed to the accident, responsibility could be apportioned.

Insurance companies would cover the losses falling within their respective policies and could subsequently exercise rights of recourse against legally responsible parties.

However, the most important conclusion for future law should be:

No one should be permitted to escape the chain of liability by arguing that “the decision was not made by a human.”

⸻

XIV. THE ROLE OF THE STATE AND INTERNATIONAL LAW

The emergence of autonomous vessels will transform public law as well as private law.

Under the existing system of the United Nations Convention on the Law of the Sea, flag States are required to exercise effective jurisdiction and control over ships flying their flag in administrative, technical, and social matters. The protection of maritime safety lies at the heart of this framework. (United Nations)

In 2043, the supervisory responsibility of the flag State cannot remain limited to:

the vessel’s hull,

machinery,

personnel,

and conventional navigational equipment.

The State will also have to supervise the reliability of the vessel’s digital captain.

Future flag-State inspections should therefore include:

AI model certification,

autonomous-system safety testing,

cybersecurity,

software update records,

algorithmic voyage-worthiness,

and the integrity of decision-recording systems.

In 2043, port States will have to inspect not only rusting hulls, but also outdated algorithms.

⸻

XV. THE FUTURE ROLE OF LAW IS NOT TO STOP TECHNOLOGY

Autonomous vessels may reduce human error.

They may optimise fuel consumption.

They may reduce accidents.

They may make maritime trade more efficient.

The law should therefore not respond to technology with fear.

However, encouraging innovation and eliminating responsibility are not the same thing.

The successful legal system of the future will be one that:

does not obstruct technology, yet does not leave the risks created by technological power without an accountable bearer.

The law should not attempt to dictate in every detail how artificial intelligence must “think.”

Instead, it should define in advance the levels of risk at which such systems may operate, who must supervise them, how their decisions must be recorded, and who will bear responsibility when damage occurs.

Because economic systems do not tolerate uncertainty well.

Insurance companies cannot properly price uncertainty.

Investors do not want unlimited legal exposure.

And injured parties want to know that there is someone legally answerable for their losses.

A strong body of AI liability law would therefore not be the enemy of innovation.

On the contrary, it would constitute the trust infrastructure of the autonomous economy.

⸻

CONCLUSION: THE CAPTAIN MAY DISAPPEAR, BUT LIABILITY CANNOT

For centuries, human beings stood at the centre of maritime commercial law.

The captain made decisions.

The crew implemented them.

The shipowner operated the vessel.

The insurer assumed the risk.

And the court evaluated human conduct.

In the world of 2043, a new decision-maker will enter this chain:

Artificial intelligence.

Yet the emergence of artificial intelligence does not mean the end of law.

On the contrary, it increases the need for law.

For most of history, law has regulated decisions made by human beings.

The task of law in 2043 will increasingly be to regulate decisions created through systems designed by humans but not made by humans in real time.

The fundamental legal principle of autonomous maritime trade can therefore be summarised in one sentence:

Decision-making authority may be delegated to a machine; legal responsibility cannot be left without an accountable bearer.

In the future, the captain may leave the vessel.

The bridge may stand empty.

Routes may be drawn by algorithms.

Ships may cross oceans without crews.

But when a ship sinks, the law will still be there.

The real question is not whether artificial intelligence will be capable of operating a vessel.

Technology will probably achieve that.

The real question is:

When machines begin making decisions, will the law evolve quickly enough to determine where responsibility belongs?

That is the question Law 2043 must answer.

And waiting until 2043 to begin answering it will already be too late.

We must debate the law of the future before the future arrives.

⸻

Sources and Legislative Framework

This study has taken into account, in particular, the provisions of Turkish Commercial Code No. 6102 concerning shipowner liability, seaworthiness/voyage-worthiness/cargo-worthiness, the carrier’s duty of care, technical fault, concurrent causes, insurance coverage, liability insurance, direct claims, and subrogation. (TBMM CDN)

With regard to international developments, the study relies on the MASS Code adopted by the IMO in May 2026, the IMO roadmap toward a binding regime extending to 2032, and the approaches concerning the MASS master and the Remote Operations Centre. (International Maritime Organization)

From a comparative-law perspective, the study also draws upon the approach adopted by the European Union’s Product Liability Directive (EU) 2024/2853, particularly its treatment of software as a product. (EUR-Lex)

Original normative concepts advanced within the Law 2043 framework:

The Autonomy Paradox – The Principle of Non-Disappearance of Liability – Algorithmic Voyage-Worthiness – The Autonomous Navigation Decision Ledger – The Layered Autonomous-System Liability Model.