RESEARCH DOSSIER
Autonomous Vessels and Insurance Responsibility
Abstract. Maritime autonomous surface ships relocate perception, navigation decisions, monitoring, emergency response, and operational records among onboard software, remote operations centers, shore management, and human masters. That redistribution makes responsibility and marine insurance increasingly dependent on evidence of who could control a particular function, when control was available, and which legal or contractual rule governs the resulting loss. This monograph reconstructs the current international architecture from the adopted 2026 International Maritime Organization MASS Code, existing law of the sea and collision rules, national remote-operation guidance, statutory marine insurance principles, compensation conventions, mutual marine insurance practice, classification cyber standards, and current legal research. It separates safety responsibility, tort or statutory liability, contractual insurance indemnity, and subsequent recourse. It evaluates system failures, failed interventions, collision, cargo loss, pollution, wreck removal, cyber attacks, software changes, and common-mode fleet exposure through explicit causal and coverage analysis. The central finding is that autonomy does not create a single new insurance liability category. It changes the factual pathways and evidentiary dependencies through which existing maritime duties and insured losses are determined. Insurability depends on specified operations, verifiable control and fallback capabilities, disclosure and wording, and calibrated accumulation exposure. A source-appraised comparison establishes present law, regulatory guidance, and proposals for reform without conflating them. No proprietary theory or new doctrine is advanced.
Keywords: maritime autonomous surface ships; MASS Code; remote operations centers; marine insurance; hull and machinery; protection and indemnity; collision liability; proximate cause; human oversight; flag-state control; cyber insurance; marine cargo; seaworthiness; limitation of liability; forensic evidence; accumulation risk
Introduction
A marine collision can produce physical hull damage, damage to another ship, pollution, loss of cargo, injury, damage to fixed property, a salvage obligation, and a subsequent limitation proceeding. Autonomous control adds distinct technical events: an invalid sensor fusion, a compromised position source, a stale machine-learning model, a missed remote-control handoff, or a communication outage. The same occurrence can generate multiple liabilities and multiple insurance responses. The analytical unit therefore must be the actual insured operation and event chain, not an undifferentiated category called an autonomous vessel.
The adopted 2026 MASS Code provides a contemporary regulatory benchmark. Resolution MSC.595(111) was adopted in May 2026 and invites application from July 1, 2026. It is a non-mandatory, goal-based instrument supplementing other international rules, principally for cargo vessels covered by SOLAS Chapter I and associated remote centers where other requirements are impracticable or insufficient. The IMO regulatory explanation makes the voluntary status and limited scope explicit. A later mandatory instrument is contemplated, with anticipated milestones through 2032, but those projections cannot be treated as law presently in force.
Methodologically, the study compares original instruments and official interpretations, a representative domestic guidance regime, operative insurance legislation where English law governs, marine liability conventions, and professional underwriting sources. Each legal inference is conditioned on the applicable flag, voyage, coastal jurisdiction, contract, treaty membership, and version date. The source-based analysis tests contested claims against alternatives and explains why safety obligations, insurance coverage, and economic compensation are different legal questions. Mathematical examples are constructed, not measurements of commercial accident frequency.
Scientific Concepts and Historical Origins
What constitutes a maritime autonomous surface ship
The scope of autonomous shipping begins with functions. A ship may use automated route planning while a human remains on the bridge; a remotely operated uncrewed workboat may transfer maneuvering and lookout to shore; another cargo vessel may employ onboard algorithms that initiate helm commands subject to specified human oversight. These architectures expose different failure modes. The MASS Code distinguishes meaningful autonomous or remote performance of functions otherwise carried out by onboard crew from ordinary enhanced automation. The IMO definition of MASS also specifies an approval process and MASS Safety Certificate within its described regulatory framework; automation alone is insufficient. These statements inform assessment without implying that every national regime has already implemented the voluntary code in identical form.
The evolution is traceable in institutional documents. The 2019 MASS trial guidelines placed experimental operations within the responsibilities of flag administrations and other relevant coastal and port authorities; risk reduction and preservation of conventional safety levels were central concerns. The 2021 IMO scoping exercises then identified interactions with existing instruments. The May 2026 adoption announcement records the transition from pre-Code experimentation to a consolidated voluntary safety framework. This is regulatory development, not proof that accident rates or insurance premiums have fallen.
Marine operations as a distributed service
A vessel’s effective operational system now may encompass its hull and machinery, onboard navigation and detection stack, shore-based power and network infrastructure, licensed and third-party communication links, remote operations centers, shared software services, weather-data providers, classification surveys, and the organizational Safety Management System. The ship remains the physical carrier of insured values and maritime hazards. Operational performance increasingly depends on geographically dispersed, contractually divided dependencies. Determining which element was necessary for the harmful event is therefore a forensic and contractual exercise.
Insurance usually attaches to recognized interests and liabilities rather than to the location of human decision-making. A hull policy addresses insured physical ship damage subject to perils, conditions and exclusions. Protection and indemnity, or P&I, addresses specified third-party liabilities under club rules. Cargo cover generally follows cargo interests and transit conditions. Equipment and technology errors-and-omissions programs may answer for different obligations of a supplier or service contractor. Where a casualty implicates all these layers, their connection arises from insured subject matter and legal duties, not from the fact that algorithms participated.
The enduring architecture of maritime obligations
International law creates continuity. Under UNCLOS Article 94 duties, flag states must effectively exercise jurisdiction and control over their ships and adopt safety measures concerning equipment, ship fitness, master qualifications, collisions and communications. The obligations are addressed to states, implemented through administrative and domestic legal arrangements. They do not, by themselves, establish that a ship's software developer is liable for a casualty or that an insurer must indemnify the owner.
The COLREG lookout and collision rules apply to navigation irrespective of whether the lookout is organized by onboard humans, remote operators, or verified sensor technology. Rule 5's requirement for proper lookout by sight, hearing, and other appropriate available means confronts systems that rely principally on cameras or radar. The legal adequacy of technological equivalence requires an examination of the applicable legal interpretation and approval, prevailing visibility and traffic, and actual system performance. A video camera installed on a ship does not establish that a proper lookout was maintained at the time of a near miss.
The 2026 Code specifies concepts of operations, operational envelopes, operational design domains, modes, fallback states, human oversight, system robustness, data management, logging, navigation, connectivity, and remote operations. These features matter economically because each can become a condition of underwriting or evidence of diligence. Their presence in a voluntary code is documentary guidance, while an owner's legally binding obligations still derive from applicable conventions, national instruments, administrative approvals and contractual promises.
Research boundary and legal pluralism
Autonomous fishing craft, small inland vehicles, experimental research craft, passenger vessels, government ships, and internationally trading cargo vessels may fall into materially different regimes. The UK Workboat Code Edition 3 is a national route for specified workboats and cannot be substituted for the 2026 SOLAS-focused international instrument. The -gross-tonnage boundary mentioned in MASS instruments is a scope reference, not an assumption that a vessel below the threshold carries no liability. Insurance must be analyzed by vessel type, operation and jurisdiction.
Control Attribution
The legal master and the functional controller
Autonomy separates several senses of control. Command authority identifies the person or organization legally charged with making navigational decisions; operational control identifies who actually sets the helm, speed, route and emergency response; technical control identifies who can change the decision logic, network permissions, sensors or software; economic control identifies which entity selects and finances the operation. Each may belong to different actors. Insurance attribution improves when these functions are separately recorded instead of assigned to a single unqualified label of operator.
The MASS Code human-oversight provisions retain the master's overall responsibility even when the master is located ashore. Human oversight includes monitoring, an ability to respond, and properly engineered authority to intervene; the same Code addresses overriding navigation functions. The formal master's responsibility remains significant, while effective command depends on whether alerts reach that person, whether communications allow intervention within the hazard's time scale, and whether the operating mode permits safe override. A contractual designation of master cannot manufacture a functioning intervention path.
The UK MGN 703 definitions distinguish the remote operator, the remote operations center, the owner or operator, and the master in the Workboat Code context. They also place competence and coordination duties on an owner/operator for the covered remotely operated uncrewed vessels. This illustrates how a national authority translates functional innovations into recognized legal roles; it does not authorize one to assert identical definitions in every flag state.
Control at the moment of transition
Handovers are critical because two participants can have plausible evidence that the other was responsible. Consider an autonomous ship approaching a narrow channel when a traffic-density alert recommends remote takeover. The onboard autonomy system requests a change, the remote center acknowledges the request but has a latent video delay, and the vessel continues to follow its previous route. A subsequent collision requires reconstruction of the time-stamped request, acknowledgment, actual switch of authority, control messages, and physical ship response. The actor with nominal authority may differ from the actor with effective capacity to prevent the event.
The causal question is event-specific: who could have issued an effective course or speed change before the final feasible avoidance point? The legal question follows: whose conduct or organizational omission breached the governing duty, and through which legal rule is it attributable to the owner, carrier, operator or supplier? The coverage question follows later: which policy, insured and trigger captures that established liability or damage? A single log entry may be decisive for one question and inadequate for the others.
Human-machine coordination and competency
The 2026 Code's attention to human-centered design, the human-machine interface and remote-center management reflects a general systems problem. A person designated as controller can experience automation surprise when the software behavior differs from the operating assumptions on which intervention procedures were trained. Excessive alerts may delay recognition; quiet failures can remain invisible. Staffing, fatigue, simulator training, alternative communications and documented decision rights determine whether human oversight is real. The Code provides a design vocabulary; actual competence remains a measured operational matter.
A remote operator may supervise one vessel, several vessels under narrow conditions, or a different subset of functions. Each allocation has a capacity constraint. If the same remote center handles simultaneous emergencies, the loss of practical control can be common across the portfolio. Underwriting must therefore look beyond nominal crew numbers to duty rosters, safe operator workload, cross-ship handoff rules, crisis escalation, training records and operational limits. Those documents offer clearer evidence than broad marketing descriptions such as fully autonomous or human-in-the-loop.
Flag state, coastal state, operator, and provider
Flag-state obligations under UNCLOS coexist with coastal and port controls, particularly in territorial waters, access channels and trial areas. The trial guidelines emphasize authorization from the relevant maritime administrations and risk controls appropriate to the planned route. Offshore remote centers may be located in a different country from the flag state and from the insured's domicile. Conflict-of-laws and jurisdiction clauses therefore become relevant in service agreements and policies; location of a server alone does not determine the proper forum for every maritime claim.
The remote center can be owned by the shipowner, a manager, or a separate technology vendor. A supplier may have contractual maintenance duties and a product-liability exposure; the shipowner may owe nondelegable or statutory obligations to third parties. Contractual indemnities and liability caps allocate ultimate recourse between commercial entities, often after the claimant's primary rights are established. The law of an affected jurisdiction may constrain contractual attempts to shift third-party or public-law responsibility.
Control documentation and forensic traceability
A credible attribution file should preserve vessel movement data, synchronized onboard and shore clocks, the autonomy build and model version, sensor provenance, system confidence or quality flags, shore display and latency history, remote user identity, input and output commands, control-mode transitions, alarms, maintenance changes, override attempts and communications failures. The MASS Code logging requirements give an internationally intelligible regulatory foundation for making these records available. Detailed retention, access and admissibility remain matters for applicable law and an enforceable operational contract.
Evidence must establish a sequence rather than a mere collection of available data. GPS time can itself be compromised; logs may be overwritten; shore clocks may drift; software builds may be updated after the casualty; incident reconstruction may use simulated behavior rather than the actual failed configuration. The standard should be reproducibility: sufficient immutable provenance to identify what version ran, what data it received, what actions it commanded, and what physical response ensued. A missing chain materially reduces confidence in both defenses and subrogation.
Autonomy Incidents
Causal pathways in collision cases
A conventional collision analysis asks about lookout, safe speed, risk appraisal, maneuver, causation and the conduct of each involved vessel. An autonomous-vessel investigation retains those questions and adds sensor quality, object classification, decision logic, training and validation, network integrity, remote override, and the timing of fallback. The COLREG navigation duties supply the governing behavioral benchmark. Compliance by one part of a system cannot excuse a later failure elsewhere when the composite operation remained unsafe.
Suppose a sensor suite correctly identifies a crossing vessel but a path planner gives excessive priority to its original arrival-time objective. Even with continuous communications and an attentive remote operator, the collision can follow from inadequate escalation if the system suppresses the warning. Conversely, an adequate planner can be defeated by an AIS spoofing attack or radar interference, leaving the system's decisions reasonable given corrupted inputs but unsafe in the physical environment. These cases have different technical causes and potentially different contract and liability consequences.
There is no inference that a collision involving an autonomous ship must have been caused by automation. Traditional hazards include weather, traffic density, channel design, deficient maintenance and maneuvers by another ship. The investigator must compare counterfactuals: with the same weather and nearby traffic, would a compliant lookout or a timely fallback have changed the outcome? This requires identification of a plausible intervention, not simply noting that a software error occurred before the event.
Connectivity, fallback, and the operational envelope
Connectivity is an operational resource with measurable outages, latency and integrity risks. The MASS Code operational-envelope provisions call for operating conditions and fallback states in which a ship remains acceptably safe when expected functions cannot operate. Weather and sea state, traffic, water depth, visibility, daylight and available communications can define permissions for different modes. A route acceptable under normal communications may be unsafe when a vessel must continue autonomously through congested waters.
A safe fallback can be a controlled speed reduction, route hold, stopping procedure, predefined maneuver, anchoring or human takeover, depending on the environment. The mere command to stop may create a collision hazard in a traffic separation scheme; an instruction to maintain course may be unsafe near a navigational limit. Fallback must therefore be judged in its actual geographic and traffic context. Underwriting should seek results from representative tests of transition failure, not assume redundancy is synonymous with safety.
A constructed event tree illustrates the causal logic. Let L denote loss of a necessary link during one voyage, D a hazardous degraded navigation situation given L, F failure of the designed fallback given L and D, and C a collision given the preceding events. The chain probability for that one specified sequence is P(L) × P(D|L) × P(F|L,D) × P(C|L,D,F). If hypothetical probabilities are respectively 0.02, 0.25, 0.10 and 0.40, the chain probability is 0.0002 per voyage, or two occurrences per 10,000 otherwise comparable voyages. These numbers are constructed for demonstration, and the total collision probability would also include all other pathways. The example requires no unjustified independence assumption because each factor is conditional.
Sensitivity is large: if remote communications degrade more often in severe weather, P(D|L) can change with route and season; if fallback depends on the same compromised sensor used by primary navigation, the failures are dependent. The simplistic multiplication of unconditional component failure rates would understate the combined risk. A serious pricing study must estimate conditional dependencies and validate them prospectively.
Software defects and the cyber boundary
Autonomy intensifies the importance of version control, patch governance, access control, and cybersecurity testing. An attacker may corrupt position, manipulate command privileges, block ship-to-shore traffic or compromise a vendor update path. A coding defect may produce the same observable steering behavior without a hostile act. The forensic distinction matters because marine cyber exclusions or endorsements may define insured events by software process, cyber operation, malicious action, or physical consequences.
The IACS revised cyber requirements address ship-level integration and onboard equipment assurance, applying within their technical and class scope. They are evidence of design and compliance expectations, not proof that a particular attack could not have happened. The LMA marine cyber wordings illustrate how materially different insurance outcomes can result from a cyber exclusion and an affirmative endorsement. The presence of an industry model clause in a catalog is never proof that it governs a particular marine policy.
Fire, machinery, environment, and an uncrewed ship
A remotely supervised vessel can suffer machinery breakdown, thermal runaway, fire, flooding, refrigerant failure, cargo instability or a damaged propeller. Crew absence changes detection, access, first response, boundary cooling and emergency repairs. Sensor alarms may provide earlier warning in certain conditions, but an insurer should distinguish timely diagnosis from physical capacity to contain the loss. The relevant question is whether available automated suppression, isolation, shore-directed systems, rescue support and port contingency measures meet the actual hazard profile.
The resulting insured loss may be damage to machinery, a constructive or actual total loss, a casualty expense, collision liability or cargo damage. Whether software malfunction is an insured peril in a particular hull policy is a coverage construction question. A software error can create a physical marine peril that damages the ship; some contracts will cover specified ensuing physical damage and others exclude it through their wording. No cross-market rule should be inferred from the presence of autonomy alone.
Common-mode and accumulation incidents
Autonomy generates a fleet accumulation mechanism: many ships may share a common operating-system release, cloud dependency, authentication service, navigational data feed, cyber vulnerability, or remote center. These dependencies mean that nominally separate hulls are not necessarily independent insured exposures. The physical vessels can be geographically dispersed while operational causation originates from one system. A large marine underwriting portfolio may therefore be concentrated in supplier identity and software version even when spread across oceans.
Formally, if I_i is an indicator of a particular fleet loss for ship i, the variance of the fleet count is ΣVar(I_i) + 2Σ_{i<j}Cov(I_i,I_j). The covariance terms rise when common software or communications failures affect ships together. If a Poisson event count N has a random common rate Λ, then E[N]=E[Λ] and Var(N)=E[Λ]+Var(Λ), illustrating extra variance from shared uncertainty. These identities are mathematical demonstrations; neither asserts an observed maritime loss distribution.
Lessons from evidence availability
The IMO deployment-status statement reports that, as of the current regulatory account, internationally trading SOLAS cargo ships have not yet entered routine commercial autonomous or remote operation. Demonstrations and short coastal trials are informative for feasibility but do not supply mature actuarial loss frequencies for transoceanic trade. Any public claim that autonomous cargo shipping has already proved materially safer in commercial global fleet experience therefore needs evidence beyond the trial descriptions reviewed here.
Coverage Liability
Separating duty, liability, and indemnity
The fact of a loss passes through several distinct determinations. First, a statutory, treaty, contractual or tort duty may attach to an owner, carrier, master, operator or other actor. Second, the relevant decision-maker establishes breach or strict-liability conditions, causation, damages and defenses. Third, the insured must show an insurable interest, a covered occurrence or peril, and compliance with policy conditions. Fourth, exclusions, deductibles, limits, aggregation and other insurance may affect payment. Finally, subrogation, contribution and contractual indemnities may redistribute costs. Treating these as one automatic liability of artificial intelligence obscures the controlling law.
The Marine Insurance Act section 55 states the English-law default that an insurer is liable for losses proximately caused by insured perils, subject to the Act and the contract. Its treatment of negligence by master or crew, the assured's willful misconduct, delay and certain mechanical injury is highly relevant but not automatically exhaustive for every contemporary autonomy incident. Policy wordings can depart from defaults within applicable legal limits. For an event involving a remote operator, the legal characterization of that operator and the specific policy's crew or employee language must be determined rather than assumed.
Hull and machinery coverage
Hull and machinery cover concerns the insured hull, propulsion and associated insured equipment, subject to agreed wording and covered perils. An insurer assessing a remotely operated ship will seek construction and class details, maintenance and testing evidence, operational limits, interface redundancy and cyber language. A fire started by a defective autonomous component may create insured physical loss, while the cost to correct faulty software without insured physical damage may fall outside a particular hull insuring clause. The decisive distinction comes from wording, facts and proximate causation rather than a general technological category.
Underwriting may specify warranted control and operational conditions: approved routes, staffing, emergency connectivity, software-change reporting, certified remote-center operation or permitted autonomous functions. Where English insurance law applies, the Insurance Act section 11 limits reliance on breach of some risk-reducing terms where the insured demonstrates the noncompliance could not have increased the actual loss risk. The provision has scope limits and excludes terms defining the risk as a whole. Consequently, describing every control as a warranty does not alone decide a contested claim. The contract, statute and causal facts must be read together.
The operational design domain can be an important underwriting representation. A ship certified to navigate an open-water route under specified visibility may later be routed through a congested strait outside its assessed conditions. The change may alter both expected loss and the applicable policy analysis. The insurer should ask which restrictions are safety parameters, which are contract terms, what exactly was represented at placement, and whether any breach affected the loss under the governing law.
P&I and third-party liabilities
The International Group's P&I account describes mutual indemnity arrangements for third-party marine liabilities including bodily injury, cargo claims, collision, pollution and wreck removal. A P&I Club is a separate contractual and mutual insurance institution; admission of an autonomous vessel, any additional survey and any restrictions depend on the relevant Club rules and entry conditions. It would be unwarranted to claim that all such ships receive identical automatic P&I cover.
Consider an autonomous collision that damages a conventional tanker and causes injury. The claimant's primary rights derive from collision law and applicable statutes and treaties; the owner's P&I response depends on insured liability, procedural obligations, limits and exclusions. A contractual claim by the carrier's cargo interest may follow a different route and trigger different defenses. The same accident can involve first-party hull indemnity and third-party P&I recovery without the two policies becoming interchangeable.
A remote operating company may owe contractual duties to the shipowner or, under particular legal regimes, direct duties to third parties. The owner may remain responsible under applicable law even after delegating functions. Separate professional indemnity, product liability or cyber coverage for the provider can support subsequent recourse. Those covers should be coordinated with the vessel's own programs, including the extent to which one insurer may seek recovery from a vendor rather than from a human operator.
Cargo and the concept of seaworthiness
Cargo exposure extends the causation chain beyond navigation. Uncrewed vessels can lose temperature control, experience delayed intervention following stowage shifts, or fail to secure dangerous cargo. The legal carrier's duties can encompass adequate systems and processes for cargo preservation. In Carey's cargo-law analysis (2026), functional seaworthiness is examined as extending to remote centers, their reliability and the competence of remote personnel, while traditional nautical-fault defenses and software defects generate unresolved interpretive questions. The analysis is scholarly argument, and any judicial outcome must be grounded in the governing carriage regime and relevant decisions.
The carrier and cargo insurer are distinct. A cargo insurer may indemnify a cargo interest for covered loss and then seek recourse against the carrier, depending on the policy and applicable rights. The carrier's defense may rely on convention exceptions, contractual allocations or facts of due diligence. A vendor software defect adds another potential recourse target but does not itself prove carrier liability. The chronological chain is cargo loss, covered claim, payment, then recovery assessment.
Pollution and compulsory financial security
Marine environmental claims often involve specific liability conventions that are independent of general fault analysis. The Bunkers Convention rules provide a shipowner-centered framework for covered pollution from bunker fuel, including security requirements for ships above 1,000 gross tonnage and direct claims against insurers in the convention's scope. The CLC liability regime treats specified oil-carrying vessel pollution under its own strict-liability and compensation architecture. An autonomous vessel's mode of operation does not convert these instruments into general compulsory insurance for every damage category.
Environmental and salvage exposure is particularly acute when the ship cannot provide effective immediate onboard response. A grounding can cause bunker release, port closure, wreck marking or removal and third-party property damage. Each category needs a precise source of legal obligation, a convention or domestic scope check, an applicable insurance contract and a causal reconstruction of preventable harm. The existence of an automation defect may be relevant to recourse but is not always necessary to establish a convention-based owner's liability.
Wreck removal and limitations of liability
The Nairobi Wreck Removal Convention supplies defined circumstances for locating, marking and removing dangerous wrecks and provides financial-security duties for qualifying ships of 300 gross tonnage and above. It allows direct insurer action and permits participating states to extend certain provisions to their territory, making geographic applicability an essential preliminary question. A damaged autonomous craft offshore may fall within the Convention only after checking state participation, its location, threshold and the local implementing rules.
Under the LLMC limitation architecture, qualifying maritime claims may be subject to limitation, with a demanding exception for personal acts or omissions involving intent or recklessness with knowledge of probable loss. The 1996 Protocol and later effective amendments changed amounts. Different contracting states and reserved exclusions can produce different results. Neither a software product defect nor a remote operation center automatically removes or confirms limitation. The relevant claimant, legal person seeking limitation, claim class, fund, applicable treaty version, and jurisdiction must all be identified.
An insurance limit and a liability limitation are different objects. A legal fund may cap certain recoveries, whereas the policy limit caps the insurer's contractual obligation. Statutory direct action may further structure claimant rights. Treating these numbers as equal without identifying the instrument and policy creates material error in severity modeling.
Cyber coverage and concurrent causation
The distinction between a compromised ship and a defective ship is often central. The LMA5402 and LMA5403 models offer examples of marine cyber exclusion and endorsement wordings, but the issuing association explicitly states that its clauses are illustrative and may be changed or not used by insurers. Accordingly, the existence of LMA5403 is not evidence that cyber losses under a particular hull or P&I policy are insured.
Imagine a collision after a hostile update corrupted onboard route validation, accompanied by a remote operator's failure to intervene. The investigator must determine the effective proximate causes and available legal theories. A policy may respond to physical ensuing damage while excluding the malicious cyber event, or use broader language that excludes the ensuing damage; the outcome turns on actual clauses and governing law. Where multiple causes operate, a claimant or insurer cannot infer an allocation by simply labeling a single event cyber or marine.
The role of the producer and provider
The owner purchases operational capacity from builders, sensor providers, platform developers, communications suppliers and remote-center operators. These actors may each contract to a performance standard. Warranty, defect, negligence, statutory product liability and contract caps may affect ultimate recourse. The effect of incorporation, arbitration and governing law cannot be assumed. A provider's professional indemnity cover may contain territorial exclusions, technology-specific limits or cyber restrictions that diverge sharply from the vessel's hull and P&I placement.
An alternative legislative position is advanced by Lin, Wu, and Lu (2026), who argue for owner-centered strict liability, limitation and compulsory insurance as a coherent future approach to autonomous-vessel torts. This is a normative proposal, not an existing universal legal rule. It usefully exposes a policy design tension: prompt victim compensation favors a solvent, identifiable defendant, while fault-sensitive recourse encourages operational diligence among multiple vendors. Existing international conventions provide examples of owner-centered security in specific pollution and wreck cases, not a general MASS insurance statute.
Underwriting information and contract drafting
A marine risk submission for autonomous operation should identify the named assured and management entities, all remote centers and subcontractors, ship and route, expected mode schedule, manning, control-transfer rules, remote-operator workloads, permitted operational envelope, availability requirements, shore dependencies, class and safety certificates, software modification authority, patch history, cyber architecture and audited emergency evidence. The statement of insured activities should match documented operations. Underwriters can then decide specific hull, P&I, cargo, cyber and vendor cover rather than relying on ambiguous autonomy descriptors.
A consistent placement file should also identify material exclusions and their exceptions, aggregation language for common software incidents, deductibles by occurrence, jurisdiction clauses, notification windows, salvage and mitigation duties, claim cooperation, and preservation of electronic evidence. Reinsurance agreements should be checked for event definitions, hours clauses where relevant, cyber exclusions and recoverability of underlying liabilities. A contract that names its exclusions clearly may be more valuable to both sides than one that silently assumes conventional crew arrangements.
Research Evidence Quality and Sensitivity
Hierarchy and scope of sources
The source set distinguishes treaty and statutory texts, adopted non-mandatory IMO guidance, national administrative guidance, class requirements, insurer-market model clauses, professional mutual-insurance descriptions and legal scholarship. This hierarchy prevents a familiar methodological error: treating the word code as if it always means binding law. Resolution MSC.595(111) calls for government application and remains non-mandatory; binding duties continue to arise through applicable conventions, legislation, certificates, approvals and contracts. Conversely, a voluntarily used technical code can be consequential to judging whether conduct was reasonable even though it does not itself create automatic private liability.
The adopted resolution and annex text provides unusually direct evidence about functional approval and safety-management intentions. Its operative document is available through a full-text third-party reproduction matched to the official IMO reference, while the official hosted original link could not be fetched in this research environment. The official IMO FAQ independently confirms number, adoption, date, instrument status and regulatory scope. This documented access distinction is more reliable than presenting a mirrored PDF as a separately authenticated copy.
The regulatory facts are relatively strong; claims of measured casualty reduction are much weaker. A policy paper supporting strict liability proves the existence of an argument, not its judicial acceptance. An industry page describing P&I proves the market's broad risk classes, not the terms of an unproduced certificate. A technical class rule proves requirements for covered designs, not that every implementing ship met them. These distinctions govern all conclusions in this monograph.
Geographical and temporal variation
The relevant legal time can be vessel design approval, date of construction contract, contract placement, software change, casualty, filing of action, or treaty amendment entry into force. For IACS cyber requirements, the revised July 2024 applicability superseded earlier withdrawn versions and has vessel-category qualifications. For the MASS Code, July 1, 2026 marks invited use of a non-mandatory instrument. For national workboats, UK guidance defines a different operational scope. A technically accurate statement becomes legally misleading if applied to a different class of vessel or a different date.
Comparative analysis also requires identifying the flag state, vessel registry, charter structure, location and nature of casualty, port state, governing law and forum, applicable direct-action rights, and coverage terms. The UK Marine Insurance Act is a valuable analytical baseline for English-governed policies, but it is not universal statutory insurance law. UNCLOS Article 94 is an international obligation of states, not directly the owner-insurer coverage clause in every court. These jurisdiction filters should appear in both legal opinions and insurer claims systems.
Difficulties of statistical identification
Reliable autonomous-vessel frequencies require a countable exposure base and consistently classified loss events. Vessel-years are inappropriate when an autonomous mode operates for only a fraction of the year; route-segment hours, autonomous nautical miles or voyages by mode can be more informative. A collision rate per autonomous voyage should distinguish remotely controlled, supervised autonomy and other mode categories, and should identify whether the denominator includes close-quarter trials or long-distance commercial passages.
The lack of mature internationally trading autonomous SOLAS experience in the current IMO assessment prevents an empirically validated comparison of global autonomous and conventional collision loss ratios on equal exposure. A conventional casualty dataset can inform hazard definitions and baseline severities but cannot produce an unbiased causal effect of autonomy without design, route, vessel, weather and reporting adjustments. Trial successes are conditional on selected environments and professional supervision; extrapolating them to open-ended commercial operation requires explicit transfer assumptions.
The target estimand for a comparative study may be E[Y(autonomous)-Y(conventional)|X] at specified vessel, route, season, traffic and weather covariates X. Observational deployments are likely selected: owners choose simpler voyages for early trials, supervision is unusually intense and reporting thresholds differ. Without an identification strategy, the observed rate difference confounds technology with selection. Matching and hierarchical models can improve comparability, while unresolved unobserved selection must still be acknowledged. No such observed causal effect is claimed here.
Severity and dependencies
Loss severity combines hull replacement, cargo value, injury, environmental cleanup, wreck removal, defense cost, interruption, regulatory delay and recovery. The combined portfolio outcome is not obtained by simply adding maximum insured limits: coverage overlaps, legal exclusions, salvage, limitation and deductibles alter the actual indemnified amounts. Scenario analysis should specify which actor pays each component before reinsurance. Software-related incidents may affect a cluster of vessels and their shared recovery dependencies, producing a tail risk disproportionate to mean claim frequency.
A sensitivity exercise can hold marginal failure rates constant while changing correlation. For n vessels exposed to a shared update, independent single-vessel loss probability p implies variance np(1-p). With positive pairwise loss correlation ρ, a stylized homogeneous model produces variance np(1-p)[1+(n-1)ρ]. At n=100, p=0.01 and ρ=0.05, the variance is 0.99×5.95=5.8905 rather than 0.99 under independence. This is a constructed illustration of concentration, not a fitted fleet model; the homogeneous correlation must also satisfy mathematical consistency constraints.
Claims evidence and falsifiability
The most valuable forensic records are those that can discriminate among competing causes. A post-event screenshot claiming the vessel was in remote control is inadequate if earlier mode-transition logs contradict it. A vendor statement that software passed validation is inadequate if the deployed binary differs from the tested version. A cybersecurity alert may be relevant but does not establish whether a malicious action changed the physical navigation trajectory. Evidence should be secured with integrity checks, time synchronization, custody records and retention controls in a form suitable for independent expert reconstruction.
A causal hypothesis should specify what record would disprove it. If the hypothesis attributes a collision to link loss, logs showing continuous authenticated command connectivity and effective downstream control weaken that explanation. If a claim attributes the loss to a software revision, a comparison of certified deployed versions across affected and unaffected vessels can test the association. If the insurer invokes a cyber exclusion, the disputed clause and its causal trigger should be mapped to the actual evidence, not to generic technological descriptions.
Limits of this research
The study verifies official source texts, statutory provisions and original scholarly records accessible by November 17, 2026, and derives illustrative probability relations independently. It does not examine confidential marine policies, claims files, onboard binary artifacts, proprietary incident datasets or an adjudicated autonomy casualty fact pattern. It does not estimate an autonomous fleet loss ratio, determine a specific policy's legal coverage, or certify a vessel. The analysis treats one full-text IMO resolution as a mirrored reproduction and identifies this evidentiary qualification explicitly. External peer review was neither commissioned nor represented as completed.
Integrated Scientific Synthesis
An integrated causal and legal interpretation
The primary cross-source result is that operational responsibility and insurance responsibility are connected through evidence but remain governed by different legal instruments. The master may retain overarching safety responsibility under the non-mandatory 2026 MASS framework; the owner may face a statutory or tort claim; a carrier may owe a cargo-related duty; a remote operator or supplier may owe contractual duties; and the hull or P&I insurer responds only within its particular coverage. These positions can coexist following a single casualty, and a finding of technical causation does not mechanically establish every link.
The change introduced by autonomy is a shift in the information necessary to assign responsibility. Where a conventional investigation reviews bridge records, voyage data, crew testimony and machinery condition, an autonomous incident additionally demands signed mode history, system configuration, remote-center traffic, deployment provenance, sensor-quality evidence, network security records and proof of effective fallback. This represents an evidentiary transformation of an established maritime allocation system. The legal beneficiary of clearer records may be the injured claimant, the shipowner, an insurer, or a vendor, depending on what those records show.
Insurance insurability as specified operating capacity
An autonomous vessel is insurable to the extent that its hazards and insured interests can be defined, assessed, priced within uncertainty and contractually allocated. This assessment is strongest when based on bounded operating contexts, authorized functions, tested fail-safe behavior, auditable remote centers and an underwriting description consistent with real operations. The existence of a new technology does not by itself determine that a risk is beyond insurance. Neither does technological sophistication eliminate severe tail uncertainty or override exclusions.
Underwriters can use the MASS Code's concepts of operations, operational envelopes, fallback and logging as prompts for documentation, while recognizing the Code's voluntary status. They can test linkage among the safety documents, the assured's representations, the hull policy, P&I entry, cargo obligations, cyber wording and the vendor contracts. The most serious coverage gap is often a mismatch: a navigation function and service chain material to a claim were never described consistently across the policies, operating approval and contracts.
Ex post recourse and ex ante incentives
A victim-centered route to compensation and a supplier-centered allocation of innovation risk answer different policy objectives. The owner may be an accessible defendant under some regimes; the developer may be the cheapest party able to correct a defect. A direct-action convention can protect a claimant for a specified loss, and later subrogation or contractual indemnity can distribute the cost. The strict-liability proposal of 2026 identifies potential legislative reforms, but existing doctrine should be applied first before reform ideas are presented as current law.
Insurance can support prevention by requiring credible test records, maintenance change controls and accountable decision rights, and by differentiating premiums and deductibles when sufficient evidence exists. An insurer should resist rewarding superficial certificate counts when real world safety performance is unobserved. A validated operating-envelope test and a reliable failure-reporting process may be more decision-relevant than an abstract statement that an autonomy platform exceeds a human operator's performance.
Portfolio-level research priorities
Early actuarial evidence should be collected at the function and mode level. Each observation should state hours or miles at risk, geography, weather, traffic, communications, autonomy version and near-miss reporting criteria. Fleet records should identify shared software versions, remote centers, cloud dependencies and class of cyber architecture. This permits estimates of both single-vessel hazard and correlated multi-vessel vulnerability while avoiding ecological conclusions from mixed-purpose trials.
Comparative legal research should publish the exact version and territorial application of each instrument. Collision attribution, cargo seaworthiness, limitation rights, pollution and wreck security, cyber exclusions, class rules and professional-service indemnities each require a different comparison set. An integrated claims chronology should preserve dates of policy placement, software release, control transfer, casualty, notification, mitigation and settlement. Strong historical provenance makes later scientific revisions and financial dispute analysis possible.
Implications for related Science
The dossier belongs primarily to Marine and Aviation Insurance. It also has direct analytical relevance to Casualty and Liability Insurance through third-party tort and statutory recovery, Cyber Insurance through compromise and cyber-wording disputes, Risk Engineering and Loss Prevention through safety cases and fallback design, Claims Science through causal reconstruction, and Reinsurance through correlated fleet accumulation. These are topical connections; the single dossier retains one canonical research identity. Such cross-Science relationships do not convert the work into a separate new theory.
The monograph's independent contribution is a comparative synthesis across original legal obligations, functional safety provisions, underwriting language, maritime liability and evidence requirements, coupled with explicit constructed causal and accumulation examples. The novel work is the specific cross-document reasoning and clarity of operational attribution, not a newly declared scientific doctrine. Its scientific conclusions remain conditional on the source regimes and case facts stated throughout.
Conclusion
The adoption of the non-mandatory MASS Code in May 2026 establishes a detailed international regulatory vocabulary for safe, accountable autonomous maritime functions. The Code's July 1, 2026 application date and its preservation of human oversight must be read alongside UNCLOS, COLREG, national implementation and maritime contracts. These instruments supply the law and relevant standards against which actual remote and automated operations must be evaluated.
Responsibility for an autonomous-vessel casualty is determined through a documented chain of operational control, legally recognized duty, factual and legal causation, and a specific insured obligation. The owner, carrier, master, remote operator and technology supplier can each appear in that chain under different legal relationships. Hull insurance, P&I, cargo cover, pollution and wreck security, cyber clauses and professional recourse respond to different components. Their coverage should be established through their actual wordings and governing jurisdictions.
The strongest prospective basis for insurability is a transparent operational design, bounded routes and conditions, verified fallback and human intervention, robust cybersecurity, preserved incident evidence, and explicit insurer and service-provider allocations. Fleet-level underwriting additionally requires recognition of software and network dependence that can synchronize apparently independent casualties. These findings follow from a comparative analysis of present sources and elementary risk relations. Future empirical research must measure mode-specific exposures and verified loss outcomes before asserting that autonomy changes the global frequency or severity of marine insurance claims.
References
International Maritime Organization. (2026). Resolution MSC.595(111): International Code of Safety for Maritime Autonomous Surface Ships (MASS Code). MSC 111/22, Annex 16. Research source.
International Maritime Organization. (2026). FAQ: Autonomous shipping (updated for the MASS Code adopted May 2026). Research source.
International Maritime Organization. (2026, May 22). IMO adopts first global Code for autonomous ships. Research source.
International Maritime Organization. (2019). Interim Guidelines for MASS Trials. MSC.1/Circ.1604. Research source.
United Nations. (1982). United Nations Convention on the Law of the Sea, Part VII, Article 94: Duties of the flag State. Research source.
International Maritime Organization. (1972). Convention on the International Regulations for Preventing Collisions at Sea (COLREGs): IMO convention overview. Research source.
United Kingdom Maritime and Coastguard Agency. (2024). MGN 703: Training and competence of remote operators working with remotely operated unmanned vessels under the Workboat Code Edition 3. Research source.
United Kingdom Maritime and Coastguard Agency. (2023). The Workboat Code, Edition 3. Research source.
United Kingdom. (1906). Marine Insurance Act 1906, section 55: Included and excluded losses. Research source.
United Kingdom. (2015). Insurance Act 2015, section 11: Terms not relevant to the actual loss. Research source.
International Maritime Organization. (1976; 1996; amendments effective 2015). Convention on Limitation of Liability for Maritime Claims (LLMC) and 1996 Protocol: Convention overview. Research source.
International Maritime Organization. (2001). International Convention on Civil Liability for Bunker Oil Pollution Damage (Bunkers Convention): Convention overview. Research source.
International Maritime Organization. (2007). Nairobi International Convention on the Removal of Wrecks: Convention overview. Research source.
International Maritime Organization. (1969; 1992). International Convention on Civil Liability for Oil Pollution Damage and 1992 Protocol: Convention overview. Research source.
International Group of P&I Clubs. (2026). About the International Group of P&I Clubs. Research source.
Lloyd's Market Association. (2019, November 13). Property and Marine Cyber Clauses. Bulletin LMA19-031-PD. Research source.
International Association of Classification Societies. (2024). Addressing cyber resilience of ships: Revised Unified Requirements E26 and E27. Research source.
Carey, L. (2026). Carriage of Goods and Maritime Autonomous Surface Ships: The Liability of the Carrier Under International Sea Cargo Regimes. NUS Law Working Paper No. 2026/007; CML Working Paper No. 2026/2. Research source.
Lin, Z., Wu, X., & Lu, X. (2026). Strict liability for maritime autonomous surface ships: Rethinking tortious accountability in the age of autonomy. Humanities and Social Sciences Communications. Article in press. https://doi.org/10.1057/s41599-026-09633-3 Research source.