Navigating without a Helmsman vis-à-vis liability for Maritime autonomous surface ships under Singapore Law

Singapore is one of the world's leading maritime hubs, with a handling capacity of over 37 million TEUs annually. It holds the Herculean task of administering one of the largest open-registry shipping communities. It is clear that the legal challenges posed by Maritime Autonomous Surface Ships (MASS) are not merely theoretical; with rising crust, they are becoming imminent. This research article examines the robustness of Singapore's existing maritime liability framework for MASS-related collision claims. It also asserts that fault-based liability, built over centuries on the premise of human agency, is structurally unsuited to vessels where navigational decisions are made by self-learning artificial intelligence. Drawing on Singapore admiralty law, English common law precedent, comparative Norwegian and EU instruments, IMO regulatory outputs, and leading tort scholarship, the article proposes a channelled strict liability regime for Degree Four autonomous vessels, paired with the existing limitation architecture under the Convention on Limitation of Liability for Maritime Claims 1976. It further argues that the doctrine of res ipsa loquitur offers an interim bridging mechanism pending legislative reform, and that Singapore's regulatory institutions are well-placed to lead international norm-setting in this domain.

I. Introduction

As per the report, Singapore's port handles and manages over 130,000 vessel calls per year, and the Maritime and Port Authority of Singapore (MPA) has actively stimulated MASS trial operations within its waters as part of the Sea Transport Industry Transformation Map.^[1] The legal apparatus governing these operations, however, remains substantially unchanged from the mid-twentieth century. Collision liability is still determined by fault, vicarious liability is still harboured in the law of agency, and the standard of care remains that of the prudent mariner, that is, a fictional person whose competence is measured against centuries of seafaring tradition rather than lines of software code.

In such a time, the arrival of MASS fundamentally disrupts these assumptions. When a collision occurs involving a vessel navigating at IMO Degree Four, where the ship's operating system makes decisions independently, without human intervention, the question of fault arises and cannot be resolved by asking which officer failed to give the correct helm order. There may be no officer, and an algorithm made the concerned decision; that algorithm may have self-learned its navigational behaviour through exposure to maritime data, such that even its designers cannot explain, after the fact, why a particular manoeuvre was executed.^[2] This is the juridical problem this article addresses.

Part II surveys the IMO's MASS autonomy framework and the status of Singapore's regulatory response. Part III analyses the existing collision liability doctrine under Singapore law and its adequacy for autonomous vessels. Part IV examines the structural problems created by artificial intelligence, principally the black-box problem and the collapse of conventional agency chains for fault attribution. Part V considers the available alternatives, with particular attention to strict liability, res ipsa loquitur, and comparative developments in Norway and the European Union. Part VI proposes a structured reform framework for Singapore and concludes.

II. The MASS Framework and Singapore's Regulatory Position

The IMO's regulatory scoping exercise, concluded in 2021, organised autonomous vessel operations into four degrees of autonomy.^[3] At one end of the spectrum, Degree One vessels carry a crew that retains full operational control, with automation assisting but not displacing human judgment. At the other end, Degree Four vessels operate entirely autonomously, with the ship's system determining all navigational decisions without reference to any human operator, whether aboard or ashore. It is these latter vessels that raise the sharpest liability questions. However, Degree Three, remotely controlled and uncrewed, presents significant issues when communication links are severed, and the remote operator loses situational awareness.

A non-mandatory MASS Code is being developed by the IMO, with a mandatory instrument anticipated by 2028.^[4] Singapore has signalled its intention to engage constructively with this process. The MPA has issued trial advisories for MASS operations in Singapore waters and participated in IMO working group sessions on the scoping exercise.^[5] Domestically, Singapore's primary maritime statute is the Merchant Shipping Act 1995, and the Collision Regulations incorporated thereunder were drafted with crewed vessels exclusively in mind.^[6] Neither statute contains provisions adapted to autonomous navigation, nor has dedicated MASS liability legislation been proposed.

Singapore is a contracting party to the COLREGs, which apply to all vessels in Singapore's territorial waters and to all Singapore-registered vessels worldwide.^[7] The COLREGs assume throughout that a vessel is under the control of a human navigator who can exercise judgment, modify course in response to developing situations, and invoke the general prudential rule under Rule 2(b) where strict adherence to the Regulations would create rather than avert danger.^[8] A vessel navigating at Degree Four that cannot recognise the exceptional circumstances Rule 2(b) addresses or that slavishly follows prescribed rules without adaptive judgment, may be rendered unseaworthy. As a consequence, the UK Supreme Court has confirmed that this flows from deficiencies in navigational decision-making systems.^[9]

III. Fault-Based Collision Liability and Its Foundational Assumptions

Liability in collision cases in Singapore, as in England, is determined in negligence. Duty of care is readily established under the neighbour principle, as restated in Singapore by the Court of Appeal's two-stage Spandeck test, requiring factual foreseeability, legal proximity, and the absence of overriding policy considerations that negate the duty.^[10][11] The standard of care is objective; that is, navigating officers are held to the standard of reasonable competence measured against the customs and regulatory framework of maritime navigation.^[12] This framework operates efficiently where human navigators are in control. It begins to fracture where the relevant navigational decisions are made entirely by code.

Sir William Scott articulated the foundational propositions of collision liability in The Woodrop-Sims, which established the four canonical outcomes: inevitable accident (loss lies where it falls), mutual fault (damages apportioned proportionately), sole fault of the suffering vessel (no recovery), and sole fault of the other vessel (full recovery).¹³ These propositions were incorporated into Singapore law through the proportionate liability regime of the Merchant Shipping Act 1995, itself tracing its lineage to the Maritime Conventions Act 1911 and the Collision Convention 1910.^[13] Underlying all four propositions is the axiomatic assumption that fault or its absence can be traced to a human actor whose conduct is assessable against a normative standard.

Where negligence is attributable not to the ship owner’s own crew but to a third party, vicarious liability governs. The controlling test derives from the House of Lords' decision in Mersey Docks and Harbour Board v. Coggins & Griffiths (Liverpool) Ltd and has been adopted in Singapore: liability attaches to the party that retained control over the manner in which the relevant work was performed, not merely over the outcome.¹⁵ This test was applied in the towage context in The Panther and The Ericbank, where the court held that the tug's master remained the servant of the tug owner for navigational manoeuvres notwithstanding the contractual structure of the towage engagement.¹⁶

The Coggins & Griffiths control test faces a fundamental difficulty when applied to MASS. Control in the legally relevant sense requires the capacity to direct the manner in which a specific operation is carried out. A ship-owner that has purchased a Degree Four autonomous navigation system from a software developer exercises no such control over the system's navigational decisions. Those decisions are made through machine-learning processes that the ship-owner cannot monitor in real time, override during a voyage, or audit after a collision in any technically meaningful way. The control that the law demands does not reside in the ship-owner, yet it does not clearly reside elsewhere either.

IV. The Black-Box Problem and the Collapse of Agency Chains

The most structurally destabilising feature of advanced AI navigation systems for maritime tort law is the black-box problem, that is, the inability of current machine-learning systems to generate human-legible explanations for specific decisions.^[14] Where a collision occurs involving a crewed vessel, the post-incident inquiry can examine bridge recordings, the navigator's testimony, and the communications log to reconstruct the decision-making sequence. Where an AI system made the relevant decisions, no comparable reconstruction may be possible. The system cannot be subpoenaed. Its decision pathway may be opaque even to its developers. Researchers are working on explainable AI, but no such capability is currently commercially standard in maritime navigation systems.

The implications for fault attribution are severe, as fault in negligence requires a standard against which conduct is measured. For human navigators, that standard is good seamanship, a body of customary practice crystallised in the COLREGs and refined through admiralty judgments across jurisdictions. For an AI system, no equivalent normative standard has been formally established. The IMO's proposed MASS Code may supply some goal-based benchmarks, and industry codes such as the MASS UK Conduct Principles already signal that autonomous operations demand heightened collision-avoidance awareness precisely because of the absence of onboard human judgment.^[15] But these instruments do not resolve the legal question of to whom the fault, once established, is to be allocated.

The agency analysis presents equal difficulty. Ship-owners operating autonomous vessels may engage specialist MASS operators under a ship management agreement modelled on BIMCO Shipman 2009.^[16] Under such agreements, the manager acts as the ship owner’s agent, and the ship-owner is vicariously liable for the manager's employees' negligent acts in the course of their management functions. But where the navigational decision causing the collision was made not by any employee of the manager, but by the autonomous system itself, supplied by a software developer who is neither an agent nor a servant of the ship-owner, the agency chain breaks. Extending the concept of agency to artificial intelligence is both analytically incoherent and practically useless, as fiduciary obligations cannot bind an algorithm, hold an algorithm personally liable, or compensate its principal for a breach.

The software developer sits outside the established liability structures. Under a fault-based regime, the ship owner’s exposure to a claimant would be limited to its failure to exercise due diligence in selecting the software supplier, an attenuated and uncertain basis of liability that may leave the collision victim without practical recourse. If the claimant were instead to pursue the developer in product liability, this would circumvent the LLMC limitations framework entirely, potentially exposing technology developers to uncapped claims and thereby creating severe disincentives to investment in MASS technology. The EU Artificial Intelligence Act classifies high-autonomy maritime navigation systems as high-risk AI and imposes pre-market conformity obligations on developers. Still, it does not resolve post-collision liability allocation between developers and ship-owners operating in Singapore waters.

V. Alternative Liability Frameworks

The inadequacy of fault-based liability for MASS collision claims does not, by itself, determine what ought to replace it. Several candidate frameworks present themselves, each carrying a different distributional logic and a different set of institutional consequences. Strict liability would shift the burden entirely to the ship-owner, irrespective of fault, on the theory that those who deploy inherently hazardous autonomous systems in shared waters ought to bear the cost of harm those systems cause. The doctrine of res ipsa loquitur offers a softer intervention, remaining within the fault framework but inverting the evidentiary burden in ways that acknowledge the structural opacity of AI decision-making. Comparative law, particularly the Norwegian and EU experience, provides a further reference point that jurisdictions that are already living with commercial MASS operations have had to confront these questions in regulatory practice, not merely in theory, and their choices illuminate both the available options and the costs of deferral. Each of these frameworks is considered in turn.

A. Strict Liability: The Case For and Against

Strict liability attaches regardless of whether the defendant was at fault. Its policy foundation is that those who profit from activities that carry inherent risks to third parties ought to bear the cost of the harm those activities cause, irrespective of the care taken.^[17] In the maritime context, strict liability is not a doctrinal novelty. The CLC 1992, implemented in Singapore through the Merchant Shipping (Civil Liability and Compensation for Oil Pollution) Act 1998, channels strict liability to the registered shipowner for oil pollution damage.^[18] The 2002 Athens Protocol similarly creates a presumption of liability for shipping incidents affecting passengers, which the carrier can rebut only by establishing enumerated defences.^[19] These precedents demonstrate that maritime law is not doctrinally closed to strict liability as it has been accepted where the nature of the risk, the difficulty of proving fault, and the availability of limitation combine to make it the most coherent solution.

Carey draws an analogy between the opacity of AI decision-making and the behaviour of an animal, as in neither case can the responsible party be asked to explain the decision that caused the harm, yet in neither case does that opacity justify leaving the victim without recourse.^[20] The analogy is illuminating. Singapore's common law, drawing on Rylands v. Fletcher, recognises that those who introduce inherently dangerous things bear the cost of escapes that cause damage, regardless of fault.^[21] The extension of this principle to those who deploy autonomous vessels in shared navigational waters is analytically coherent, though legislative confirmation would be required in Singapore's statutory maritime environment.

The statistical argument reinforces the case, as approximately 75% of maritime accidents are attributable to human error as a primary factor,^[22] and MASS advocates invoke this figure to argue that autonomous vessels will be safer than crewed ships. That may ultimately prove true. But until empirical evidence from commercial MASS operations demonstrates a superior safety profile, there is no basis for presuming that Degree Four vessels are less dangerous than their crewed counterparts.³³ The presumption ought to run in the opposite direction, that is, a vessel introducing a novel, incomprehensible and empirically untested decision-making process into shared waterways is a hazardous actor, and its operator ought to bear strict liability for harm caused.

Collin identifies the principal objection to strict liability as being that it may be unjust to hold a shipowner strictly liable for a collision that a crewed vessel in identical circumstances would equally have been unable to prevent.^[23] This is a genuine concern, but not a dispositive one. First, the shipowner captures the commercial benefits of operating an autonomous vessel, reduced crewing costs, 24-hour operations, and expanded cargo capacity. It is not obviously unjust that it also bears the residual risk. Second, strict liability under the LLMC is subject to limitation, which caps the shipowner's exposure at a level calibrated to the vessel's tonnage. Third, the shipowner retains full contractual recourse against the software developer where the collision results from a demonstrable defect in the system as supplied.^[24]

B. Res Ipsa Loquitur as a Transitional Mechanism

Even within the existing fault-based framework, the doctrine of res ipsa loquitur offers a partial solution to the evidentiary difficulties of MASS collision claims. The doctrine, stated by Erle CJ in Scott v London and St Katherine's Dock Co, permits an inference of negligence from the fact of the accident itself where the thing causing harm was under the management of the defendant and the accident is of a kind that does not ordinarily occur in the absence of negligence.^[25] The Singapore Court of Appeal confirmed in Grace Electrical Engineering Pte Ltd v Te Deum Engineering Pte Ltd that res ipsa loquitur shifts the evidential burden to the defendant to furnish an explanation for the accident.

Applied to MASS collisions, the doctrine would operate as follows: the mere fact of a collision involving a Degree Four vessel raises a presumption that the autonomous system was not performing to the standard of a prudent seafarer, and the burden falls on the shipowner to explain why this presumption is wrong. This is an improvement over the ordinary fault regime, where the claimant carries the full burden of proving the specific navigational error. It does not, however, fully resolve the problem as even with the burden reversed, the shipowner's inability to audit the AI's decision-making may render exculpation practically impossible, a result that comes close to the functional effect of strict liability without its formal clarity or legislative authority.

C. The Norwegian and EU Comparative Perspective

Norway, which is at the vanguard of commercial MASS deployment through the Yara Birkeland and ASKO Maritime autonomous ferry programmes, has thus far maintained the proportionate fault liability regime of the Norwegian Maritime Code for collision claims. At the same time, its legislature considers bespoke MASS provisions.^[26] The Norwegian approach reflects a cautious incrementalism as it preserves the existing framework, observes the technology as it matures, and legislates when empirical data is available. This is defensible in a jurisdiction with a small shipping community and robust state oversight of trial programmes. It is less well-suited to Singapore's position as a global trading hub through which millions of tonnes of third-party cargo transit annually.

The EU AI Act, now in force, classifies autonomous navigation systems operating at high degrees of independence as high-risk AI, requiring conformity assessments, technical documentation, and human oversight mechanisms before market placement.^[27] While Singapore is not an EU member, the supply-chain implications are significant as MASS navigation software developed for the European market and deployed in Singapore-registered vessels will increasingly be designed to the AI Act's standards. Singapore regulators would be well-advised to develop interoperable liability rules that do not conflict with the compliance architectures that AI Act certification will create, particularly given that the Act's product liability implications for developers may otherwise incentivise limiting the commercial availability of MASS systems in Asian markets.

VI. Proposals for Reform and Conclusion

This article proposes three reforms to Singapore's maritime liability framework for MASS.

First, the Merchant Shipping Act 1995 should be amended to introduce a statutory strict liability regime for collisions involving Degree Three and Degree Four MASS, channelled to the registered owner or demise charterer of the autonomous vessel. Liability should be subject to the existing LLMC limitation framework,^[28] which already extends to claims "whatever the basis of liability may be," ensuring that strict liability does not upset the foundational balance between ship-owner exposure and claimant recovery that the international maritime liability system has carefully crafted. The LLMC's "whatever the basis of liability" formulation is no drafting accident: it anticipates precisely this kind of doctrinal development, and Singapore's implementation of the 1996 Protocol places it within the most current generation of limitation regimes.

Second, the MPA should issue guidance pending legislation clarifying that res ipsa loquitur applies as a default evidentiary rule in MASS collision proceedings, placing the burden on the shipowner to explain the autonomous system's navigational conduct.³¹ This interim measure works within existing common law doctrine and requires no statutory amendment. It would reduce the asymmetry of evidentiary difficulty that currently places MASS collision victims at a structural disadvantage without requiring the legislature to resolve, in advance, every doctrinal question that autonomous navigation raises.

Third, the MPA, in coordination with the IMO working group on the MASS Code, should advocate for product liability gateways in the mandatory MASS Code provisions, allowing a ship-owner held strictly liable for an autonomous system's navigational failure to claim contribution from the software developer where the failure results from a defect in the system as supplied.^[29] This mirrors the product liability logic of the EU AI Act and the recourse provisions of the CLC 1992, without collapsing the LLMC structure by exposing developers to direct, unlimited claims from third-party collision victims.

The introduction of autonomous vessels into shared maritime waters is occurring now, in Singapore's own port.^[30] Tort law's function, as the scholarship has long recognised, is not merely to compensate past injury but to distribute risk in ways that reflect the moral economy of a given era.³² The moral economy of autonomous maritime navigation is one in which the human capacity for error has been traded for algorithmic decision-making whose error profile is unknown, whose choices cannot be explained, and whose operation generates substantial commercial profit for the ship-owner. The case for imposing the cost of collision damage, subject to sensible limitations, on the owner of that profit is not an argument against technological progress. It is an argument that progress ought to be underwritten by those who capture its benefits.

Singapore is uniquely positioned to lead on this question. As the jurisdiction that brought the Spandeck framework to bear on duty of care, that has developed a sophisticated admiralty jurisprudence responsive to commercial realities, and that hosts significant IMO liaison functions in the region, Singapore has both the institutional authority and the doctrinal flexibility to forge the liability framework that MASS demands. The window for principled legislative initiative remains open before the first significant autonomous vessel collision claim reaches the Singapore courts and forces an improvised answer from the bench. It will not do so indefinitely.

Four designates fully autonomous ships capable of independent navigational decision-making without any human intervention, whether onboard or remote.

References

1. 2. International Maritime Organisation, Outcome of the Regulatory Scoping Exercise for the Use of Maritime Autonomous Surface Ships (MASS), MSC.1/Circ.1638 (3 June 2021) [hereinafter IMO RSE]. ↩
2. 3. IMO RSE (n 2). Degree One encompasses automated decision-support vessels with crew aboard; Degree Four designates fully autonomous ships capable of independent navigational decision-making without any human intervention, whether onboard or remote. ↩
3. 4. IMO, Development of a MASS Code: Report of the Maritime Safety Committee, MSC 107/16 (2023). A non-mandatory interim Code is anticipated before a mandatory instrument enters into force, provisionally targeted for 2028. ↩
4. 5. MPA, Singapore MASS Trials Advisory (MPA 2022); see also the MPA's participation in IMO correspondence groups on MASS regulatory development, summarised in IMO RSE (n 2) Annex 1. ↩
5. 6. Merchant Shipping Act 1995 (Cap 179, 2020 Rev Ed) [MSA 1995]. The Collision Regulations are incorporated through the Merchant Shipping (Prevention of Collisions at Sea) Regulations (Cap 179, Rg 10), which give domestic effect to the COLREGs for vessels in Singapore waters and Singapore-registered vessels worldwide. ↩
6. 7. Convention on the International Regulations for Preventing Collisions at Sea 1972 (COLREGs), art VI(4) (tacit acceptance procedure). Singapore is a contracting party: IMO, Status of IMO Conventions (IMO 2024). ↩
7. 8. COLREGs, r 2(b). The general prudential rule was considered in The Nordlake and The Seaeagle [2015] EWHC 3605 (Admlty); [2016] 1 Lloyd's Rep 656, [190], and applied in Singapore in The Dream Star [2017] SGHC 200; [2017] 2 Lloyd's Rep 538, [47]. ↩
8. 9. Alize 1954 v Allianz Elementar Versicherungs AG (The CMA CGM Libra) [2021] UKSC 51; [2021] 2 Lloyd's Rep 613. The Supreme Court confirmed that a defective passage plan can render a vessel unseaworthy. Singapore admiralty courts follow English precedent by virtue of the Application of English Law Act 1993 (Cap 7A). ↩
9. 10. Donoghue v Stevenson [1932] AC 562 (HL), 580 (Lord Atkin). The neighbour principle frames the duty of care inquiry in Singapore maritime tort proceedings. ↩
10. 11. Spandeck Engineering (S) Pte Ltd v Defence Science & Technology Agency [2007] 4 SLR(R) 100 (CA), [77]. The Court of Appeal adopted a universal two-stage test requiring (i) factual foreseeability, (ii) legal proximity, subject to (iii) policy considerations, as the governing framework for duty of care in Singapore. ↩
11. 12. Blyth v Birmingham Waterworks Co (1856) 11 Ex 781, 784 (Alderson B). The objective standard of the reasonable person in maritime proceedings is calibrated against the customs and regulatory framework of seafaring: The Tasmania(1890) 15 App Cas 223 (HL), 226 (Lord Herschell). ↩
12. 13. The Woodrop-Sims (1815) 2 Dods 83; 165 ER 1422 (Sir William Scott). The four propositions: inevitable accident, mutual fault, sole fault of the suffering vessel, sole fault of the striking vessel remain foundational in Singapore admiralty law. ↩
13. 14. The proportionate fault rule, introduced in England by the Maritime Conventions Act 1911 giving effect to the Collision Convention 1910, is implemented in Singapore through MSA 1995 (Cap 179), s 10. Proportionate liability means each vessel contributes to the damages of the other in proportion to its degree of fault. ↩
14. 15. MASS UK Industry Conduct Principles and Code of Practice 2022 (V6) (Maritime UK 2022), [6.10.2]. Though a UK instrument, it represents the most developed industry-level standard presently available and informs MPA guidance development, analogously to how IMO non-mandatory codes have historically preceded Singapore domestic regulation. ↩
15. 16. BIMCO Shipman 2009, cl 17(2)(b)(ii). The clause excludes the ship manager's liability for crew acts or omissions save where the default results from the manager's failure to discharge its crew management obligations. Its application to an AI-operated vessel, where there is no "crew" in the conventional sense, raises unresolved drafting questions. ↩
16. 17. Baris Soyer, "Autonomous Vessels and Third-Party Liabilities: The Elephant in the Room" in B Soyer and A Tettenborn (eds), New Technologies, Artificial Intelligence and Shipping Law in the 21st Century (Informa 2019) 106–107. Soyer identifies five structural reasons why fault-based liability is inapt for fully autonomous vessels, including the opacity of self-learning systems and the ethical values embedded in algorithmic design. ↩
17. 18. International Convention on Civil Liability for Oil Pollution Damage 1992 (CLC 1992), art III(1). Implemented in Singapore through the Merchant Shipping (Civil Liability and Compensation for Oil Pollution) Act 1998 (Cap 180). Strict liability is channelled to the registered owner and paired with compulsory insurance under art VII. ↩
18. 19. Protocol of 2002 to the Athens Convention relating to the Carriage of Passengers and their Luggage by Sea 1974, art 3(1). The carrier is presumptively liable for "shipping incidents" and may escape liability only by establishing one of the enumerated defences set out in art 3(1)(a)–(b). ↩
19. 20. Luci Carey, "Contractual and Tortious Maritime Liability Regimes and the Introduction of Autonomous Vessels" NUS Law Working Paper 2023/012; NUS Centre for Maritime Law Working Paper 23/03, 27–28. Carey draws an analogy between an AI system's inexplicable navigational choices and an animal's unexplained behaviour, noting that in neither case should opacity insulate the owner from liability. ↩
20. 21. Rylands v Fletcher (1868) LR 3 HL 330. The rule that one who brings onto land something likely to do mischief if it escapes bears strict liability for damage caused by the escape was received into Singapore law through the Application of English Law Act 1993 (Cap 7A). Its extension to autonomous vessels operating in shared navigational waters is analytically plausible, though legislative confirmation would be required. ↩
21. 22. Carine Dominguez-Péry et al, "Reducing Maritime Accidents in Ships by Tackling Human Error: A Bibliometric Review and Research Agenda" (2021) 6 Journal of Shipping and Trade 4. The 75% figure is widely cited in MASS regulatory debates but must be assessed against the equally unquantified error profile of autonomous navigation systems, for which no comparable empirical dataset yet exists. ↩
22. 23. Felix Collin, "Unmanned Ships and Fault as the Basis of Shipowner's Liability" in H Ringbom et al (eds), Autonomous Ships and the Law (Routledge 2021) ch 5. Collin raises the objection that it may be unjust to hold a shipowner strictly liable for a collision that a crewed vessel in identical circumstances would also have been unable to prevent. ↩
23. 24. LLMC, art 2.1. The phrase "whatever the basis of liability may be" means that limitation applies whether liability sounds in negligence, breach of contract, or strict liability: Francis Berlingieri, Berlingieri on Arrest of Ships (6th edn, Informa 2016) para 60.04. ↩
24. 25. Scott v London and St Katherine's Dock Co (1865) 3 H & C 596; 159 ER 665 (Erle CJ). Applied in Singapore in Grace Electrical Engineering Pte Ltd v Te Deum Engineering Pte Ltd [2018] 1 SLR 76 (CA), [39], where the Court of Appeal confirmed that res ipsa loquitur shifts the evidential burden to the defendant to furnish an explanation for the accident. ↩
25. 26. Norwegian Maritime Code 1994 (Sjøloven), ch 7. Norway retains proportionate fault liability for collision claims while commercial MASS operations proceed aboard the Yara Birkeland and ASKO Maritime vessels: Lena Grete Mjelde, "Autonomous Ships in Norwegian Law" (2021) 36 IJSL 112, 118–120. ↩
26. 27. EU Artificial Intelligence Act (Regulation (EU) 2024/1689), art 6 and Annex III. Autonomous maritime navigation systems operating at high degrees of independence are likely to fall within the "high-risk AI systems" category, triggering conformity assessment, technical documentation, and human-oversight obligations before market placement. ↩
27. 28. Convention on Limitation of Liability for Maritime Claims 1976 (LLMC), art 1(4). Singapore implemented the LLMC and its 1996 Protocol through MSA 1995 (Cap 179), Pt VIII and the Third Schedule: The Sunrise Crane [2004] 2 SLR(R) 692 (CA). ↩
28. 29. M Tsimplis, "Carriage of Goods on Autonomous Ships" in S Girvin and V Ulfbeck (eds), Maritime Organisation, Management and Liability: A Legal Analysis of New Challenges in the Maritime Industry (Hart 2022) ch 12. Tsimplis examines the interaction between MASS liability, general average, and salvage obligations, areas in which Singapore's position as a leading maritime arbitration seat makes doctrinal clarity especially urgent. ↩
29. 30. MPA, Singapore MASS Trials Advisory (MPA 2022); see also the MPA's participation in IMO correspondence groups on MASS regulatory development, summarised in IMO RSE (n 2) Annex 1. ↩