Rethinking the matter of autonomous vessels in International Maritime Law
In the spring of 2022, a vessel slipped quietly out of a Norwegian port carrying a cargo of mineral fertilizer. No captain stood at the helm. No crew walked the decks. The Yara Birkeland, the world’s first fully electric, zero emissions, autonomous container ship, had entered into operation. Developed by Yara International and Kongsberg Maritime, the vessel was designed to eventually operate entirely without crew, replacing 40,000 annual diesel truck journeys.
This ship is not a singular instance of such inquisitiveness. Alongside it, many research projects have emerged, such as the MUNIN (Maritime Unmanned Navigation Through Intelligence in Networks), DNV GL’s Revolt, and Nippon Yusen Kaisha’s autonomous vessels trials, which test the technical feasibility of autonomous ships. They signal a shift towards how goods and people are going to move across the world and across oceans. The question is no longer whether autonomous ships can sail internationally, it is whether international law is prepared for them when they do.
The current legal landscape
International Maritime Law is one of the oldest and most developed bodies of Public International Law. It has several instruments which make up the current architecture of governance, these include: United Nations Convention on the Law of the Sea (UNCLOS), the Safety of Life at Sea (SOLAS) Convention, the International Regulations for Preventing Collisions at Sea (COLREGs), the Standards of Training, Certification and Watchkeeping for Seafarers (STCW) Convention, and the Maritime Labour Convention (MLC).
Despite the development in this field, there is a stark challenge that maritime law faces with autonomous vessels. Every one of these instruments was built on a single assumption: that a human being would always be on board and would bear ultimate responsibility for the vessel.
The Mass Code: A roadmap not a rulebook yet
Maritime Autonomous Surface Ships (“MASS”) are vessels that operate independent of human action to varying degrees. The International Maritime Organization (IMO), has a taxonomy of degrees of autonomy for MASS vessels. This framework is essential to establish the ambit of the existing regulatory framework and includes 4 degrees of automation, with degree 1 having the most human involvement and degree 4 being fully autonomous.
An IMO scoping exercise, concluded in May 2021, aimed at mapping the adequacy of existing conventions, the need for amendments, and the conventions/provisions, which had gaps, in dealing with autonomous vessels. The diagnosis of this exercise was sobering. Most major conventions need significant amendments to accommodate these developments.
Subsequently, a joint working group developed a goal-aligned MASS code. At the 109th session of the Maritime Safety Committee (MSC 109), the IMO agreed to a revised roadmap which involved a non-mandatory MASS code to be finalized and adopted by May 2026. This would be followed by an experience building phase till 2028, leading to a finalized mandatory code to be adopted by 1st July 2030 that will enter into force on January 1st, 2032.
This roadmap is significant, but pertinent in understanding what the MASS code will and will not do. The code will be goal-based instead of prescriptive. It will provide a framework “against which a vessel using autonomous technologies can be assessed and certified for commercial operations within IMO’s structure,” but will not mandate specific technical methods. The responsibility for developing a compliance stratagem falls largely on industry, flag states, and classification societies.
It is pertinent to note that autonomous shipping is currently not fully regulated by the International Association of Classification Societies (IACS) or the IMO. This means that individual flag states must decide how to regulate MASS within their own jurisdictions, creating a fragmented, uneven global picture.
Where the law falls overboard
The scoping exercise identified gaps across virtually every major international maritime instrument; they cluster into five critical areas, each having legal challenges. Liability and accountability are the largest challenges. Thus the question remains: Who is responsible when an autonomous vessel causes a collision; the owner, the software developer, the remote operator, or the algorithm itself? Collision regulations (COLREGS compliance) require human judgement. Thus, we extend our question further, asking if artificial intelligence will be able to exercise the ‘ordinary practice of seamen?’
The status of seafarers is a further question as well. SOLAS, STCW and MLC all presuppose crew on board. Remote operators have no clear legal status under existing conventions. UNCLOS requires a genuine link between a vessel and its flag state, developing this concept around the presence of human nationals.
Fully networked autonomous vessels face new threat vectors. A possible hijacked navigation system could become a weapon scaled at the volume of a ship with no crew to intervene. Furthermore, search and rescue and salvaging are heavily impacted by the duty to render assistance at sea, as enshrined in UNCLOS Article 98 and SOLAS Reg. V/33. How does an unmanned ship fulfil this duty of helping other people in peril at sea?
The liability labyrinth
There is a particularly troubling dimension: the “black box” nature of AI decision-making. If a neural-network navigates a vessel into a collision, recovering the sequence of decisions that led to the incident and attributing fault becomes technically impossible. Evidence-gathering in maritime litigation already depends on voyage data recorders (VDRs); the opacity of the machine learning and decision making process significantly compounds the difficulty in tackling this challenge enormously.
Traditional maritime law assigns ultimate responsibility to the master, who is the human being commanding the vessel. When something goes wrong at sea, the legal chain involves the master, the shipowner, and the insurer. Autonomous vessels introduce multiple new parties into play: manufacturers, AI software developers, remote operators, and algorithmic decision-making systems that may act in ways their creators did not anticipate them to.
In the United States, this problem has been recognized to have taken a legal form by the identification of the “Pennsylvania Rule.” The Pennsylvania rule is a federal maritime doctrine under which a party in a collision, who has violated a regulation intended to prevent that type of collision, is presumed liable. The rule presents unique challenges for cases involving autonomous ships because causation for these collisions or other incidents may be difficult to attribute. Principles like the Pennsylvania Rule, are now increasingly pushed to be incorporated in maritime liability allocation as seen in recent allisions and collisions cases.
The COLREGs problem
The 1972 Convention on the International Regulations for Preventing Collisions at Sea (COLREGs) is the “road code” of the ocean. Every collision case before an admiralty court is ultimately measured against it. Rule 2(a) holds that no vessel, owner, master, or crew can be absolved from the consequences of neglecting the rules. Rule 5 mandates a “proper look-out by sight and hearing.” Rule 8 requires that any action taken to avoid collision be “large enough to be readily apparent to another vessel observing visually.”
These rules presuppose human perception, judgment, and moral responsibility. The COLREGs contain subjective requirements and deviations from the written rules may be required by “the ordinary practice of seamen, or by the special circumstances of the case.” This rule sets a standard for professional wisdom and contextual reasoning. We can conclude that the 1972 COLREGs require amending, for the rules that are presently laid forth are not compatible with the operation of fully autonomous vessels.
Remote operators in the legal web
Whether the ship is operated from the shores or operates independently without any crew, as is the case for Degree 3 and Degree 4 MASS ships, the question of the legal position of the “master” is spectral. SOLAS defines a master as a person in command of a ship. STCW requires specific training, certification, and watchkeeping standards for officers on board. The MLC defines a seafarer as “any person who is employed or engaged or works in any capacity on board a ship.” None of these instruments contemplate someone sitting in an office in Oslo, overseeing the movement of a container ship passing through the Malacca Strait.
There are suggestions that remote operators should be treated as masters, “by perpetuating the notion of the master, retaining their status as the final entity of responsibility for the ship,” but these are merely suggestions and are not the law just yet. IMO’s own working groups have begun differentiating between “MASS crew/seafarers” and “remote operators,” but no clear legal regime for remote operator qualifications, duties, or liabilities currently exists under international law.
Cybersecurity: another crew member
“Ships are now interconnected with many devices and the shore. The communication and devices used have cyber vulnerabilities.” For a conventionally crewed vessel that experiences a cyberattack, a crew can respond, improvise, and override the attack. For a degree 4 autonomous vessel with no crew on board, the same attack could result in total loss of control with no human recourse available.
The IMO’s MSC resolution 428 on Maritime Cyber Risk Management in Safety Management Systems provides a starting point to addressing cybersecurity concerns, but it is noted that the IACS has identified “the human element and careless habit of employees” as the weakest link in cyber security, This forms a factor that shifts dramatically in character when there is no human on board at all.
Frameworks for a hybrid future
The question of how to close these legal gaps is not merely technical. It sits at the intersection of commercial interests, labour rights, environmental imperatives, and geopolitical competition. Any framework that hopes to succeed must “ensure that the uniform and consistent enforcement of UNCLOS provisions” is maintained while making space for technology that did not exist when these provisions were written.
The move by the IMO to adopt a goal-based approach for the development of the MASS Code, is in part an experience learnt from aviation regulations. The Chicago Convention on international civil aviation, and the standards issued by the International Civil Aviation Organization (ICAO), have successfully regulated unmanned aircraft systems (UAS) using a similar framework. Maritime lawyers and academics have pointed to the aviation precedent as an instructive, yet imperfect, model.
A few goal-based suggestions
There are a few mechanisms that can be adopted and would aim at bridging the gap that exists at the moment between current regulations and prospective regulations that would comprehensively accommodate autonomous vessels.
First, COLREGs regulations need amending to replace anthropocentric language (“ordinary practice of seamen”) with outcome-based substitutes, such as collision avoidance that is “effective, and verifiable.” This preserves the safety objective while allowing AI navigation systems to meet the standard by any technically adequate means.
Second, ensuring increased algorithmic transparency. Existing VDR requirements should be extended to mandate logging of AI decision-making processes in a standardised, interpretable format. Without this safeguard, post-incident liability allocation is arbitrary and practically guesswork.
Third, mandating the IMO’s existing MSC resolution on cyber risk management. The currently non-binding nature of these resolutions is evidently inadequate for fully networked autonomous vessels. A new SOLAS provision should establish minimum cybersecurity standards for MASS, including requirements for intrusion detection and fallback systems.
Charting a Course Forward
The human cost of autonomy is not a secondary concern, the question of the maritime workforce cannot be set aside as a downstream consequence of automation; it must be addressed as a central feature of any regulatory framework. Aspects of the seafaring profession, like situational awareness, improvised engineering, and diplomatic negotiation cannot be replicated by AI yet.
The challenge before regulatory and educational bodies is not to mourn the demise of shipping careers but to plan for a shift in focus such that the vast reservoir of knowledge within the shipping sector is utilized to fulfil new roles.
Autonomous maritime vehicles are no longer a theoretical problem: they are a commercial reality. Vessels are already in service, more are under development, and the legal framework has not yet fully adapted. The IMO deserves credit for recognizing the regulatory lacunae and taking serious steps to remedy it. But the roadmap to a mandatory code by 2032 is a decade away, and commercial pressure to deploy autonomous shipping will not wait for legal surety.
What the MASS Code must deliver is a framework that is technologically neutral in its standards, human-centred in its values, and international in its reach. It should determine whether the voyage is safe, the environment is protected, the workers are treated fairly, and someone can be held accountable when things go wrong.
The boat still needs a captain. The question is where that captain sits.
Featured/Headline Image Caption and Citation: “35th anniversary of the International Maritime Organization (IMO) International Maritime Law Institute (IMLI) at IMO Headquarters in London, England on 22 April 2024,” Image Sourced from Wikimedia Commons | CC License, no changes made