Autonomous Vessels, Algorithmic Navigation and the Emerging Law of Unmanned Shipping
From the IMO MASS Code to the River Drone 5 Collision: How Maritime Law Is Absorbing the Crewless Ship
Introduction
On 22 May 2026, the International Maritime Organization adopted the first global safety code for autonomous and AI-enabled commercial ships.1 The Maritime Autonomous Surface Ships Code, agreed at the 111th session of the Maritime Safety Committee in London, establishes a goal-based framework for the design, construction, certification and operation of remotely controlled and autonomous vessels. It applies to large internationally-trading cargo ships and takes effect on 1 July 2026. The code is non-mandatory. A mandatory version is planned for adoption by 2030 at the latest, with entry into force on 1 January 2032.1
The adoption is a regulatory milestone, but the legal questions it does not answer are more significant than the safety requirements it establishes. Maritime liability law was constructed around the presence of a human master exercising judgment on a physical bridge. The International Regulations for Preventing Collisions at Sea presuppose a lookout maintaining a proper watch by sight and hearing. The Hague-Visby Rules allocate carrier liability on the basis of due diligence by human agents. When the bridge is empty and the navigation is algorithmic, each of these frameworks confronts a gap between its assumptions and the technology it must now govern.
The commercial deployment of autonomous vessels is no longer theoretical. The Yara Birkeland, a 120-TEU electric container vessel, has been in commercial operation between Heroya and Brevik in Norway since 2022 and completed a fully autonomous voyage under human supervision in March 2023, though regulatory constraints mean it continues to carry a small crew.2 The River Drone fleet of ten autonomous dry cargo barges is conducting commercial operations on European inland waterways.3 The Netherlands amended its Inland Waterways Police Regulations from 1 January 2025 to remove the mandatory crew requirement, enabling applications for autonomous navigation.3 The MASS Code will accelerate this trajectory. The law is catching up, but it is doing so by adapting existing frameworks rather than building new ones.
The MASS Code and Its Limits
The MASS Code addresses safety requirements across thirteen chapters, covering surveys and certificates, risk assessment, system design, software principles, safe operations management, alert management, manning and training, safety of navigation, connectivity and remote operations.1 Its central structural feature is the retention of a designated master who holds absolute legal responsibility for the vessel at all times, whether physically stationed on the bridge or monitoring from a remote operations centre ashore.1
This is a pragmatic regulatory choice, not a legal answer. It preserves the existing chain of liability under The International Convention for the Safety of Life at Sea (SOLAS) and flag-state law by ensuring that someone, somewhere, can be identified as the responsible human. But it does not resolve the harder questions. When an autonomous navigation system makes a routing decision that leads to a collision, the master at the remote operations centre may have had no practical opportunity to intervene. The MASS Code requires human oversight but does not define the standard of care for a remote operator monitoring multiple vessels through a digital interface rather than standing watch on a single bridge.
The experience-building phase from 2026 to 2030 is designed to generate the operational data that will inform the mandatory code. During this period, flag states may authorise trials and commercial operations under their own domestic regulatory frameworks. The regulatory landscape will be jurisdictionally fragmented until at least 2032.
The River Drone 5 and the First Collision Involving an Autonomy-Equipped Barge
On 5 December 2024, the autonomous barge River Drone 5 collided with a container vessel on the Scheur River near Rotterdam.3 At least four containers were spilled into the waterway. The River Drone 5 is one of ten autonomous dry cargo barges, each approximately 100 metres in length with a cargo capacity of 3,850 tonnes, operated by the Dutch maritime technology firm Seafar using remote navigation systems.3
The operator confirmed that the vessel was under the command of a captain on board at the time of the collision and was not operating in autonomous mode.3 The distinction is critical for liability allocation: if the vessel had been navigating autonomously, the liability analysis would have engaged product liability for the navigation system, the classification society’s certification of the autonomous capability and the regulatory framework under which the vessel was authorised to operate. Because a human captain was in command, the incident falls within conventional fault-based collision law.
The incident arrived three weeks before the Netherlands activated its new inland waterway regulations permitting crewless autonomous operation from 1 January 2025.3 The timing illustrates the regulatory challenge: vessels capable of autonomous operation will frequently switch between autonomous and human-commanded modes. The liability framework must accommodate both states and the transitions between them.
Collision Liability and the COLREGs Problem
The 1972 Convention on the International Regulations for Preventing Collisions at Sea was drafted for vessels commanded by human officers making real-time navigational decisions. Rule 5 requires every vessel to maintain a proper lookout “by sight and hearing as well as by all available means appropriate in the prevailing circumstances.” Rule 7 requires the use of “all available means” to determine risk of collision. Rule 8 requires action to avoid collision to be “positive, made in ample time and with due regard to the observance of good seamanship.”4
These rules assume human sensory perception and human judgment. An autonomous navigation system does not maintain a lookout “by sight and hearing” in any conventional sense. It processes sensor data, radar returns, AIS signals and camera feeds through algorithmic decision-making. Whether this constitutes compliance with Rule 5 depends on whether “all available means appropriate in the prevailing circumstances” can encompass an autonomous sensor suite. The IMO’s regulatory scoping exercise acknowledged this ambiguity but did not resolve it. The MASS Code’s approach is to require that the autonomous system achieve functional equivalence to human-commanded navigation, without specifying how that equivalence is to be assessed in a collision inquiry.1
When two autonomous vessels collide, the traditional apportionment of fault under the Brussels Collision Convention of 1910 faces a category problem. Fault in collision law is assessed against the standard of good seamanship. Good seamanship is a human standard, developed through centuries of admiralty case law and predicated on the judgment of a competent mariner. The question of whether an algorithm can exercise or fail to exercise good seamanship has no settled answer. Norwegian maritime law scholars have proposed that collision costs could be shared proportionally based on COLREGs compliance, treating technical failure as the functional equivalent of navigational error, but no court has adopted this analysis.5
Classification Society Liability
Classification societies set the standards for ship design, construction and maintenance. Hull, cargo and P&I underwriters require classification as a prerequisite to insurability. For autonomous vessels, classification societies will certify not only the physical structure of the ship but the reliability and safety of the autonomous navigation system, the sensor suite, the communications infrastructure and the software that controls the vessel.6
The liability exposure is material. Classification societies have historically disclaimed responsibility for vessel safety, fitness for purpose and seaworthiness, maintaining that their role is verification against published standards rather than warranty of operational capability. This disclaimer has been tested in conventional shipping and has generally held. The question is whether the same disclaimer will survive when the classification society has certified an AI navigation system that subsequently causes a collision. The society’s assessment of autonomous capability is not a routine structural survey. It involves evaluating software reliability, sensor integration, failure-mode analysis and cybersecurity resilience. If the certified system fails and a vessel is lost, claimants will argue that the society’s certification constituted a representation of capability that induced reliance by the shipowner, charterer and cargo interests.6
P&I Insurance and the Coverage Gap
Protection and indemnity insurance covers third-party liabilities including crew injury, cargo claims, collision damage and pollution. The P&I clubs that provide this cover operate as mutual associations, pooling risk across their membership. Their rules and coverage are designed around conventional crewed vessels.7
Autonomous vessels introduce risks that sit outside the existing P&I framework. Cyber-attack on a navigation system, software failure causing grounding, sensor malfunction leading to collision and communication loss between the vessel and its remote operations centre are all plausible casualty scenarios with no direct precedent in P&I claims history. The International Group has not yet produced a uniform market wording for autonomous vessel risk. Individual clubs have indicated willingness to cover autonomous operations on a case-by-case basis, but the premium methodology, the exclusion architecture and the allocation of liability between shipowner, software provider and remote operations company remain undefined.7
The insurance market is likely to follow the pattern seen in other AI-exposed sectors. For a period, autonomous risks will sit in “silent” coverage: neither expressly included nor expressly excluded. The ambiguity will probably be resolved by litigation after a casualty, not by proactive policy drafting.
Electronic Trade Documents and Smart Bills of Lading
The digitisation of shipping documents is proceeding on a separate but convergent track. UNCITRAL’s Model Law on Electronic Transferable Records has been enacted or adopted in legislation in at least eleven jurisdictions.8 The United Kingdom’s Electronic Trade Documents Act 2023 gave digital trade documents legal equivalence with paper.8 India’s Bills of Lading Act 2025 modernised the statutory framework for bills of lading in a major emerging market, but it should not be treated as equivalent to the United Kingdom’s Electronic Trade Documents Act 2023.8
Blockchain-based smart bills of lading promise to automate the transfer of title to goods in transit. If an autonomous vessel carries cargo under a smart bill of lading that releases title on delivery confirmation from the vessel’s own sensors, the entire transaction from loading to title transfer could in principle occur without human intervention. The legal infrastructure is being assembled. The ICC’s 2024 Digital Trade Survey found that 49.2% of respondents were already using electronic bills of lading in some capacity. That is a respondent-usage figure, not a measure of eBLs as a proportion of all bills of lading issued.8
The convergence of autonomous vessels and electronic trade documents raises liability questions that neither framework addresses in isolation. If a smart bill of lading releases title based on sensor data from an autonomous vessel and the cargo is subsequently found to be damaged, the allocation of liability between the carrier, the software provider, the sensor manufacturer and the blockchain platform operator has no precedent.
Strategic Outlook
Maritime law absorbs technological change slowly and through accretion. The MASS Code will not create a comprehensive law of autonomous shipping. It will provide a safety framework within which flag states, classification societies, P&I clubs and commercial parties must build their own liability and contractual structures.
Three gaps will define the next decade of autonomous shipping disputes. The COLREGs gap, where the collision avoidance regime has not been rewritten for algorithmic navigation and will be interpreted by admiralty courts applying human-seamanship standards to machine decisions. The insurance gap, where P&I coverage has not been restructured for autonomous risks and where the first major casualty will produce coverage litigation before it produces safety regulation. And the liability gap, where the allocation of responsibility between the remote master, the shipowner, the software provider, the classification society and the communications infrastructure operator remains contractually and doctrinally unresolved.
The shipowners and technology companies that build the first generation of commercial autonomous fleets are constructing the factual matrix from which maritime law will be remade. The direction of travel is not fixed. The law may adapt fault-based principles to autonomous operation. It may move towards strict liability for autonomous systems. Much will depend on where the first catastrophic casualty occurs and which court hears the case.
Notes
1. IMO Maritime Safety Committee, 111th session (MSC 111), 13-22 May 2026; adoption of the non-mandatory Maritime Autonomous Surface Ships (MASS) Code; entry into force 1 July 2026; mandatory code planned for adoption by 2030, entry into force 1 January 2032. DNV reporting on MSC 111 confirms thirteen-chapter structure and retention of designated master with absolute legal responsibility.
2. MV Yara Birkeland, 120-TEU electric autonomous container vessel, operating between Heroya and Brevik, Norway (approximately 7 nautical miles), since 2022; fully autonomous voyage under human supervision completed March 2023; continues to carry small crew due to regulatory constraints; built by Vard, owned by Yara International. Yara International reporting.
3. River Drone 5 collision, Scheur River near Rotterdam, 5 December 2024; approximately 100m autonomous barge, 3,850-tonne capacity, operated by Seafar; four containers spilled; operator confirmed vessel was under human command, not autonomous mode, at time of collision. Maritime Executive, Marine Insight and Breakbulk News reporting. Netherlands amended Inland Waterways Police Regulations (BPR) from 1 January 2025 removing mandatory crew requirement and enabling applications for autonomous navigation exemptions.
4. Convention on the International Regulations for Preventing Collisions at Sea 1972 (COLREGs), Rules 5 (lookout), 7 (risk of collision) and 8 (action to avoid collision).
5. Wiersholm, ‘Liability for damage caused by autonomous ships: a Norwegian perspective’, discussing fault-based collision liability under the Norwegian Maritime Code and its application to autonomous vessels; ScienceDirect, ‘Tortious liability for autonomous marine vehicle collisions: A suggestive move from fault-based to strict liability’ (2025).
6. Skuld, ‘Liability of Classification Societies’; IACS classification requirements for hull, cargo and P&I insurability; SAFETY4SEA, ‘Insurers’ Considerations for Autonomous Ships’.
7. International Group of P&I Clubs; SAFETY4SEA, ‘Insurers’ Considerations for Autonomous Ships’; Riviera Maritime Media, ‘Autonomous shipping creates liability waves’.
8. UNCITRAL Model Law on Electronic Transferable Records (MLETR), enacted or adopted in legislation in at least 11 jurisdictions per UNCITRAL status page; UK Electronic Trade Documents Act 2023; India’s Bills of Lading Act 2025 modernised India’s bills of lading framework; it does not provide the same electronic trade document equivalence model as the UK Electronic Trade Documents Act 2023; ICC 2024 Digital Trade Survey reporting 49.2% of respondents using electronic bills of lading, with actual eBL penetration as a proportion of total bills of lading issued remaining in the low single digits.