RESEARCH DOSSIER
Automated Driving Liability and Vehicle Insurance Allocation
Abstract. Automated driving changes the relationship between the person occupying a vehicle, the technological system performing the dynamic driving task, the entities supervising its operation, the manufacturer responsible for product safety, and the insurer advancing compensation after a crash. This monograph examines how legal responsibility for road harm is attributed and how automobile-insurance payments are allocated when the driving task is partly or fully automated. The investigation integrates technical definitions, operational design domains, takeover obligations, vehicle-holder liability, product-defect claims, compulsory third-party coverage, first-party physical damage and injury benefits, manufacturer recourse, systemwide loss accumulation, and the evidence required to distinguish them. Primary sources include United Kingdom automated vehicle legislation, German road traffic law, European Union motor-insurance and product-liability instruments, United States federal vehicle-safety reporting and California authorization materials, Japanese operational approval records, and international technical standards. Comparative legal interpretation is combined with actuarial reasoning and reproducible hypothetical claim calculations. The analysis finds that an automated driving classification does not itself designate a civil defendant or an insurer: those outcomes depend on the applicable statutory compensation rule, contemporaneous operating mode, insured interest, insurance contract, causation evidence, and any subsequently established rights of recovery. First-response compensation and ultimate economic incidence are therefore separately ascertainable. The paper identifies practical consequences for investigation, insurance reserving, correlated fleet risk, consumer protection, software evidence, and cross-border comparison. Empirical safety claims remain confined to the available evidence and its exposure, reporting, and selection limitations. The original contribution is a documented comparative synthesis and transparent worked analysis; no new legal doctrine, scientific theory, or actuarial prediction is asserted.
Keywords: Automated driving; vehicle insurance; autonomous vehicles; motor liability; automated driving systems; product liability; compulsory automobile insurance; crash attribution; operational design domain; vehicle-holder liability; insurer subrogation; claims evidence; insurance allocation; software defects; comparative regulation; cyber-physical risk.
Introduction
A collision involving an automated vehicle presents a deceptively simple question: who pays? The scientific problem is more exacting. A vehicle may be owned by one person, insured through a policy purchased by another, deployed by a fleet operator, supplied with an automated driving system by a manufacturer, updated by a software provider, and monitored remotely by an organization exercising duties defined by regulation. Each relationship can be relevant without providing an automatic answer to liability. A road victim may obtain payment under a statutory insurance scheme even when final fault or product defect remains disputed; the paying insurer may then pursue a legally supported recovery. A successful insurance claim can consequently precede the determination of ultimate loss incidence.
The World Health Organization's road-safety assessment situates the inquiry within the established international burden of traffic injury. Automated driving presents potential changes to crash patterns, but its eventual effect on serious harm must be demonstrated in a comparable exposure population. The rare-event demonstration analysis explains why directly proving fatal-crash improvements through accumulated driving miles may require extraordinarily large observational exposure. Neither prospective benefit nor current technical ambition substitutes for observed evidence.
This study draws on statutory text and official technical and supervisory records available for examination as of November 16, 2026. It distinguishes enacted legal rules, commencement and implementation, technical approval, insurance contracts, empirical observations, and explicit hypothetical illustrations. Jurisdictional differences are substantive: a German holder-liability claim, a United Kingdom automated-vehicle insurer claim, and a United States state-law negligence suit may arise from similar mechanics but rest on different legal predicates. The purpose is to determine which analytical questions are stable across these systems and which answers depend on the specific legal setting.
The foundational research question concerns the passage from factual operation to legal compensation and then to insurer accounting. Technical standards define the driving task; legislation establishes duties and rights; policies define the insured risk within mandatory limits; claims evidence connects a specific event to the applicable sources of obligation. The result of this ordered inquiry is an allocable claim, not merely a label attached to a vehicle.
Scientific Definitions and Primary Foundations
The object of allocation
Automated driving liability is the legally determined responsibility arising from injury, death, or damage connected to a vehicle whose driving functions are performed, to a specified extent, by an automated system. Vehicle insurance allocation is the assignment of payment and residual loss among injured claimants, insurers, insured parties, vehicle holders, fleet operators, manufacturers, technology suppliers, and other potentially responsible entities under applicable insurance and liability rules. These definitions identify objects of comparative study, not a claim that a single worldwide rule governs them.
The relevant event must be specified. At a minimum, an analysis should record the location and date of the collision, the vehicle configuration and approved operating domain, whether automation was actually engaged, the role required of a human driver, any requests to intervene, the action of the system and other road users, the kind of injury, the statutory source of compensation, and the effective contracts. A manufacturer recall, a regulatory defect investigation, a tort judgment, and an insurer's indemnity payment answer different institutional questions. The United States safety-defect notification law, for example, creates manufacturer-related safety notification obligations; it does not itself resolve every state-law civil claim for accident damages.
Six terms require consistent use. Harm means the physical or financial consequence of the crash. Causation concerns the factual and legally relevant contribution of conduct or product condition to that consequence. Liability concerns an enforceable legal obligation owed by an identified entity. Coverage concerns whether a particular insurer owes a defense, benefit, or indemnity under its policy and controlling law. Recovery concerns a subsequent claim for reimbursement, contribution, or subrogation against another responsible entity. Paid loss is the amount actually disbursed; incurred or reserved loss is an estimate or recognition that can change as evidence develops. The distinctions matter because reimbursement may occur years after the victim's payment and may affect net insurer results without reversing the original compensation.
Defining automation, operational domains, and insurance interests
The SAE J3016 classification (2021) supplies the familiar six levels, 0 through 5, and separates assistance to a human driver from systems that execute the complete dynamic driving task within a defined operational design domain. Level 2 assistance can simultaneously influence steering and speed while a human driver remains the driver. Level 3 permits a system to execute the task under specified conditions while a fallback-ready human responds to an intervention request; Levels 4 and 5 differ principally in domain boundaries and fallback capability. The meaning of a level depends on the specified feature and conditions, rather than on a marketing description of the whole car. Safety assurance, statutory authorization, and coverage require additional analysis.
The operational design domain (ODD) may be restricted by roadway class, speed, weather, daylight, map coverage, roadworks, infrastructure, or operating area. A claim at a geographic boundary can turn on whether the software was operating inside the authorized conditions and which actor had responsibility during the transition. A Level 4 service approved for limited routes cannot be treated as having Level 5 universal capability merely because passengers need not actively steer.
Insurance interests likewise differ. A private owner may insure physical damage to the vehicle while the legal system requires third-party liability coverage for its use. A fleet operator may hold commercial liability, motor, cyber, general liability, and excess policies; a supplier may have product liability insurance. The NAIC's automobile coverage explanation describes common U.S. distinctions between liability, collision, and comprehensive insurance, while noting state-law and contract differences. It would be incorrect to infer that a policy naming an automated vehicle covers every form of manufacturer product defect or every commercial passenger activity.
Three decisions after one crash
Suppose an automated fleet vehicle hits a pedestrian. The first decision is the victim's recoverable entitlement under local law, including whether liability is strict, fault-based, or governed by an injured-person benefit scheme. The second is the immediately responding payer: possibly the motor insurer, a statutory fund, a first-party injury scheme, or a defendant satisfying a judgment. The third concerns final cost allocation after investigation: a supplier may be liable for a defect; an operator may bear responsibility for forbidden deployment; or a third party may have caused the incident. These stages may occur in a different temporal sequence.
The EU motor-insurance amendments strengthen and coordinate rules protecting victims of road accidents, but do not abolish distinct national civil-liability systems. The United Kingdom's automated-vehicle insurer provisions create a particularly explicit route by which the insurer can face direct statutory responsibility in qualifying automated-driving accidents, with possible later recovery rights. Thus, identifying the initial payer is both a protection question and a distinct underwriting problem. The empirical object is the entire sequence of payment and recourse, with each decision tied to its own legal and evidential basis.
Automation Level and Control
Driving-task responsibility and handover
Operational responsibility is temporally specific. Engagement state, activation eligibility, user warnings, operator monitoring, takeover demand, response time, and fallback maneuver together determine which actions were available and required. A request to intervene is relevant only where the applicable feature legitimately presupposes a fallback-ready human and the request is timely and intelligible under the governing rule. A crash in the seconds surrounding a transition cannot be assigned solely by the label “automated.” It requires chronology.
German legislation illustrates the point. Under StVG section 1a, the statutory conditions for highly or fully automated functions and permissible use are expressly addressed; StVG section 1b addresses duties of the vehicle driver, including readiness to resume control in prescribed circumstances. Duties are not identical across all types of automated operation. For vehicles with autonomous functions in specified operating areas, StVG section 1d establishes a different statutory setting, while section 1f on operational participants addresses responsibilities of those participating in the operation, including the vehicle holder and technical supervision. A human seated in a vehicle, a remote technical supervisor, and the legally relevant driver may accordingly occupy different roles.
The UNECE Regulation No. 157 concerns approval of automated lane keeping systems within its specific regulatory scope. Its documented requirements supply technical variables for checking operation and transitions, yet approval of one system against one regulation does not determine tort responsibility in all countries. The documentary revision itself notes the separate legally authentic text. In technical evidence, approval compliance should be recorded as an evidentiary circumstance and assessed alongside the facts of failure, instructions, maintenance, and use.
The causal importance of a takeover request depends on the event sequence. For illustration, consider a feature that issues a valid prompt at time t=0. The record shows an escalating warning at t=2 seconds and an impact at t=4 seconds, with a plausible hazard already present before the prompt. A finding that the human was nominally expected to respond says little without testing whether avoidance was physically feasible at t=0, whether the prompt met regulatory requirements, and whether the system should have executed a safe fallback. This illustration establishes questions to test, not a universal safe reaction interval or a presumption of fault.
Recorded operational state and the evidentiary chain
Modern automotive investigation combines vehicle data, system event logs, cloud telemetry, driver-monitoring records, cameras, maintenance histories, software-version controls, roadway evidence, medical records, and independent witnesses. They differ in reliability, custody, temporal precision, and access. A reported active automation state may be contradicted by event data, a manual override, delayed network transmission, or an outdated vehicle-identification record. Investigators should preserve original timestamps, clocks and time-zone conventions, event hashes, recording-system descriptions, and every interpretation step.
Germany's StVG section 63a concerns storage and processing of defined events involving highly or fully automated driving functionality; StVG section 1g addresses data processing associated with vehicles with autonomous functions. These rules provide legal context for recording and access but must not be collapsed into a universal right for an insurer to obtain all telemetry. Whether a particular actor may inspect a particular record depends on its role, legal basis, necessity, and applicable privacy and procedural law.
Data evidence must be connected to causation. A software disengagement shortly before a collision might be the cause of a hazard, a response to it, or a routine consequence of a driver override. A defective object-recognition output may explain an unsafe maneuver, while the record of object recognition itself may be incomplete. Human intervention may be negligent, reasonable, or impossible. Expert reconstruction should document alternative causal accounts, visibility, braking capacity, vehicle dynamics, contradictory sensor channels, and reasons why one explanation is preferred. Claims staff should preserve both supporting and exculpatory artifacts; a single proprietary “fault score” cannot substitute for a legally relevant causal explanation.
Exposure, crash reporting, and claims about safety
The NHTSA Standing General Order requires specified organizations to report certain crashes involving automated driving systems and Level 2 advanced assistance, with different triggers for the two categories. Its 2025 amendment and later presentations identify problems of duplicate reporting, uncertain automation classification, varying access to telematics, and inconsistent observation. Reported incident totals are administrative counts, not automatically comparable safety rates. They lack uniform mileage or trips denominators, and the sets of vehicles and environments differ.
Suppose two operators each report ten crashes. One accumulated 50 million urban passenger miles, the other 500,000 motorway miles, and their reports may follow different ascertainment practices. Neither equality of counts nor any raw count ratio identifies which automated fleet is safer. Even a crude rate requires exposure, road mix, severity definitions, period, collision causation, and reporting completeness. The Kalra–Paddock sample-size study emphasizes how rare fatal events make direct statistical demonstrations slow; its result does not establish that any current vehicle is safe or unsafe.
For vehicle insurance, evidence should be maintained at the coverage-event level: crash frequency per comparable exposure unit, bodily-injury severity, repair expense, property-damage claims, fault and coverage disputes, development lag, and costs of defense or recovery. Safety data provide necessary inputs but cannot be repurposed into premium indications without adequate linkage to insured exposure, policy terms, limits, claim inflation, and the market in which the vehicles operate. The methodological consequence is to model automation features and operating conditions, rather than treating “autonomous” as a single homogeneous actuarial class.
Product and Operator Liability
The United Kingdom: statutory insurer response and future authorization
Part 1 of the Automated and Electric Vehicles Act 2018 established a distinctive framework for listed automated vehicles involved in accidents while driving themselves. Section 2 sets out specified circumstances in which the insurer is liable for damage caused to an insured person or other person; where the vehicle is uninsured, the owner may be liable under the section's conditions. This statutory mechanism begins with the qualifying event and the protected claimant. It must be read alongside the Act's definitions, exclusions, contributory-negligence provisions, limits and rights of recovery. Section 4 addresses certain effects of prohibited software alterations and failure to install safety-critical updates on insured-person claims, subject to its actual terms. Section 5 addresses recovery by the insurer or owner from other persons liable for the accident. The statutory text is more precise than any claim that “the manufacturer always pays.”
The Automated Vehicles Act 2024 develops authorization, responsibilities and in-use oversight for vehicles intended to drive themselves. It distinguishes the authorized self-driving feature and relevant operational roles, including circumstances involving a user-in-charge and operation without such a user. The Act's provisions require attention to commencement. The UK government's implementation programme describes the operational scheme as targeted for implementation in 2027, while individual statutory provisions and subordinate measures may have different commencement dates. For the November 2026 research cutoff, it is therefore essential to distinguish enacted legislative architecture from rules that have become operationally effective. The Law Commissions' joint recommendations explain the rationale for separately addressing safety authorization, in-use responsibility and legal accountability.
The relationship between the 2018 insurer-payment mechanism and the 2024 authorization structure is analytically central. A motor insurer's obligation to compensate a qualifying victim is not identical to the allocation of a safety-management duty under the authorization regime, and the latter does not automatically establish the whole causation analysis of an accident. A victim-facing insurance obligation can be clear even when investigators need months to identify a defective sensor or a breach of an operating duty. The reverse is also possible: proof that an entity violated an administrative requirement may not by itself resolve every civil-liability or coverage element. Both questions depend on statutory conditions.
Consider an insured vehicle recognized by the applicable statutory list that, while driving itself, collides with a cyclist. Initial analysis tests the s2 statutory event and covered damage before considering manufacturing fault. A subsequent engineering investigation reveals a demonstrable defective update. A legally available s5 recovery might shift part of the insurer's net loss to another liable entity; it does not retrospectively eliminate the insurer's initial victim-facing payment. A separate scenario, in which the feature was not in its relevant self-driving mode, requires examination of ordinary road liability and contract provisions rather than assuming the s2 route. These are legal mechanism illustrations, not determinations of actual claims.
Germany: holder liability, human duties, and defined operating areas
German road traffic law begins from a consequentially different baseline: StVG section 7 establishes vehicle-holder liability for certain harm arising in the operation of a motor vehicle, with statutory exceptions and limitations that must be read in context. A harmed person need not first reconstruct a software defect in order to identify every possible holder-liability theory. This does not mean that the holder, driver, technical supervisor, manufacturer and insurer are legally interchangeable. Different bases of responsibility may coexist and enable further contribution or recourse.
For highly or fully automated functions, §§1a and 1b address permissible use and human responsibilities, including recognized conditions for transferring control. For autonomous vehicles operating in defined areas, §§1d–1g and the German autonomous-vehicle operating regulation define approval, operating arrangements and participants' functions. The type of authorization, operational area and required technical supervision determine the relevant duty structure. A remote supervisor may be responsible for specified remote tasks without having physically operated the steering wheel. A holder can have statutory responsibilities even when the system carries out the driving task.
The allocation problem involves at least four evidence paths: the triggering facts for holder liability; any human driver's breach of a contemporaneous statutory duty; operational failings of regulated participants; and the existence and causation of a product or maintenance defect. A claimant may pursue a legally established pathway while inter-party recovery turns on different evidence. An insurer's ability to price the risk also depends on whether the insured fleet uses one common supplier, update platform and technical supervisor. Insurance risk is therefore attached both to vehicles and to the operational organization through which they are deployed.
European Union: motor compensation, approvals, and defective products
European Union law combines harmonized aspects of motor insurance, product liability and vehicle safety. They serve distinct functions. Directive 2009/103/EC, as amended by the 2021 Motor Insurance Directive amendments, concerns compulsory insurance against civil liability in respect of vehicle use, victim protections and relevant cross-border institutional arrangements. It does not make every automated-vehicle crash a product-liability event. National civil-liability and coverage implementation remains important. The analyst should verify whether a particular vehicle and use fall within national transposition of the applicable EU definition, especially when comparing ordinary road travel, confined routes and novel operating environments.
The EU General Safety Regulation 2019/2144 and Implementing Regulation 2022/1426 address vehicle type approval and technical specifications for automated driving systems within their scoped applications. Type approval can provide substantial evidence about the approved design and safety case. It neither provides automatic proof that an accident was unavoidable nor extinguishes a claim concerning a specific defect or negligent operation. A vehicle may satisfy an approval condition at entry into service and later experience a defective update, maintenance failure or domain mismatch. Accident investigation must identify the configuration actually used at the relevant time.
Directive (EU) 2024/2853, the recast Product Liability Directive, specifically addresses defective products in an updated technological environment, including software-related questions and economic operators. A May 2026 EU corrigendum revised its temporal applicability wording so that the new regime concerns products placed on the market or put into service after December 8, 2026. This research is dated November 16, 2026: the new directive's future application must not be retroactively treated as the substantive rule governing all previous placements or collisions. Applicable national transposition, transitional provisions, and the older legal regime for earlier products require examination in any actual claim. The distinction between date of product placement and date of crash is particularly important where vehicle software is updated after the original vehicle was sold.
An automated vehicle can raise an insurance claim and a potential product-defect claim simultaneously. They may share evidence but have different elements: the insurance question asks whether a statutory or contractual compensation duty arose; the defect question asks whether the product was defective under the applicable regime and whether the defect legally caused compensable damage. Manufacturer recovery may fail because defect proof is insufficient even where insurance compensation was properly due. Conversely, a proven defect may support a separate claim beyond a specific motor policy, subject to governing law and procedural limits.
The United States and Japan: administrative powers and separate liability law
The United States presents a division between federal motor-vehicle safety regulation and diverse state rules governing torts, insurance, product liability, vehicle operation and commercial services. The federal defect-notification statute empowers safety enforcement and recall-related processes. The NHTSA crash-reporting framework supplies a federal stream of specified incident data and supports safety investigations. Neither imposes a uniform federal strict-liability answer for all crashes involving automated vehicles. State laws, permit requirements, contract wording and judicial decisions remain central for compensation and insurance allocation.
California's autonomous vehicles programme distinguishes testing and deployment authorization, while its permit resources address evidence of required financial responsibility and operational reporting. A permit or insurance certificate is evidence of authorization or financial security within its scope; it does not itself prove a claimant's liability allegations, establish unlimited coverage, or eliminate state law distinctions among an owner, operator, driver, manufacturer and supplier. A sophisticated claims analysis must record whether the incident occurred during testing or permitted commercial operation, with or without a driver, and which applicable requirements governed that phase.
Japan's March 2023 Level 4 approval announcement documents the initial approval of a specific-driverless-operation setting rather than countrywide authorization for all vehicles and roads. Comparative analysis should separate that administrative milestone from the complete Japanese motor-liability and insurance framework. The existence of a Level 4 approval says which operational conditions were authorized; it does not by itself establish the legal result of an injury claim. National compulsory automobile liability, operator arrangements and any applicable product claims must still be identified before assigning payment responsibilities.
The common lesson of the U.S. and Japanese examples is evidentiary: approval status is relevant to lawful deployment, and system capability is relevant to technical causation, but neither can be substituted for the legal test that establishes liability, contract coverage or compulsory benefits. This approach avoids importing a British insurer-first statutory provision into a jurisdiction that has enacted no equivalent.
Defects, updates, and common-cause losses
Vehicle hardware, software models, positioning services, maps, communications, maintenance and operational policies can contribute to a single crash. Several cars may share the same component or update. A design defect can therefore create correlated claims across vehicles and insurers, even when drivers and routes are independent. The practical underwriting unit may be the deployed software version or supplier relationship as much as the vehicle-year. The insurer should inventory supply dependencies, rollback capability, distribution timestamps, cybersecurity exposure, recall status and affected vehicle identifiers.
The Schellekens (2015) liability analysis identifies how liability rules may influence automated-vehicle development incentives; that economic and legal argument is not an empirical attribution of fault in individual accidents. The more recent Roussou and colleagues (2026) review surveys legal-framework and liability-allocation research, indicating both scholarly attention and jurisdictional heterogeneity; its available review findings cannot replace examination of a statute applicable to a named claim.
Software correction timing also matters. A claim against a manufacturer might concern a defect in original design, an update that introduced a hazard, or failure to issue a required remedy. A claim against an operator might concern postponed installation of an available mandatory safety update, or deployment outside the certified domain. A third-party cyberattack introduces further questions about foreseeability, evidence, exclusions and recovery. Under no circumstance does an insurer's operational label such as “cyber loss” itself decide the rights of a pedestrian compensated under motor insurance law.
Coverage and Claims Responsibility
The insurance contract and compulsory compensation
Insurance analysis begins with the applicable statutory compensation baseline and the actual documents, not a presumed industry practice. A compulsory motor third-party liability policy principally protects against specified civil-liability exposure to others; its mandatory limits and protected claimant categories vary. First-party physical damage, personal injury protection, medical payments, uninsured motorist, collision, comprehensive, product liability, technology errors-and-omissions, commercial general liability and umbrella coverages have distinct definitions and exclusions. The NAIC coverage guidance offers a useful U.S. taxonomy but does not fix terms for every jurisdiction, form or fleet arrangement.
A policy may identify a “driver” even where lawful operation assigns part of the dynamic driving task to an automated system. Contracts must be interpreted alongside mandatory law, public policy, the declarations, insured-person definitions, covered use, vehicle schedule, territorial limits and applicable endorsements. Questions can arise over permissive use, commercial passenger service, remote supervision, modifications, maintenance, cyber events, recall activity and the identity of the named insured. None can be resolved by treating automation as a single exclusion across all policies. Under compulsory liability law, a contract limitation may be ineffective against a protected victim while remaining relevant to recourse or the relationship between insurer and insured, depending on the governing legal regime.
The IAIS Insurance Core Principles provide a supervisory framework for adequate governance, risk assessment and fair conduct; the framework is not a worldwide automobile insurance policy form. It is useful for evaluating insurer systems that gather evidence, communicate coverage reasoning, administer claims and monitor outsourced automated-fleet services. A statutory payment is still subject to the exact governing rule, not a generic expectation that technology providers will absorb every claim.
Allocation sequence, contribution, and recoveries
The operational claims sequence can be documented as follows in prose. Investigators establish the event, location and victims. They identify applicable law and coverage schedules. They record technology mode and relevant insured use. They determine the initial right to payment and calculate its legally eligible amount. The insurer or other responsible body pays according to the applicable obligation, while making appropriate reserves for outstanding loss and expenses. The claim file preserves potential contribution, subrogation and product claims and pursues them only when substantive and procedural conditions permit.
The sequence separates gross and net insurance cost. Let C be the total gross compensation and allocated expenses recognized by the insurer for one covered event, and R the actually recoverable amount from other liable parties. Net cost, in a purely arithmetic description, is C − R. A requested recovery is not necessarily a realized recovery. R must be bounded by legal entitlement, proof, counterparty ability to pay, settlement terms, allocation of deductibles, costs and any limitations periods. When recovery depends on pending litigation or uncertain defect proof, recording the entire claimed amount as certain cash would misrepresent the insurer's position.
Contribution among several insurers is a distinct problem from contribution among tortfeasors. Motor, commercial liability, manufacturer product liability and excess policies may insure different entities and hazards. Whether defense costs erode limits, whether an insurer has a duty to defend, how overlapping coverage is allocated, and when excess layers attach depend on contract wording and governing law. The mere existence of two policies does not create a 50:50 allocation. Multiple defendants likewise do not establish equal causal shares. The ultimate distribution must be tied to recognized rights and actual recoveries.
The United Kingdom's statutory recovery structure demonstrates why victim payment and ultimate recourse can lawfully occur at different moments. The German holder-liability provision illustrates how a different primary compensation predicate may apply, while manufacturer liability and insurance recovery are additional inquiries. Under the EU motor-insurance framework, cross-border motor compensation also requires attention to national implementation and the relevant guarantee or claims-handling institution. Comparative generalization is appropriate at the level of analytical steps, not a fictional uniform liability formula.
Worked first-response and recourse examples
The following illustrations are hypothetical, use no real claims data, and create no legal entitlement. Assume a collision produces $300,000 of eligible third-party bodily injury and property compensation, $20,000 of covered allocated loss-adjustment expenses, and a separate $80,000 of physical damage to the insured vehicle. Assume all stated coverage conditions and limits are satisfied. The initial third-party insurer records $320,000 gross cost; a first-party physical damage insurer may separately pay $80,000 according to that distinct policy. That does not imply the two amounts fall under one limit or one insured interest.
In hypothetical Case A, there is a valid first-instance insurer duty toward the injured third party and subsequent engineering evidence supports an enforceable manufacturer recovery of $120,000, of which only $90,000 is ultimately collected after settlement. The insurer's recognized economic cost associated with the $320,000 gross third-party claim is $230,000 if all those amounts concern that insurer and there are no additional expenses. Calculating $320,000 − $120,000 = $200,000 before collectability has been established would overstate the realized recovery by $30,000. The case illustrates reserve and recovery measurement, rather than determining when accounting rules permit recognition of a recovery asset.
In hypothetical Case B, the same gross compensation is paid but no product defect or operator fault can be established against another party. The gross $320,000 remains with the paying insurer, subject to its own reinsurance and contract arrangements. The absence of recourse does not imply that victim compensation was improper. From the claimant's standpoint, both Case A and Case B could produce the same timely payment; from the insurer's standpoint, they create different ultimate loss ratios. This is why public claims-paid data alone cannot measure manufacturer liability shares.
In hypothetical Case C, an automated fleet suffers 40 similar losses in one quarter after a defective common software release. Assume average eligible indemnity of $25,000 per claim and $5,000 average associated expense. The scenario produces 40 × ($25,000 + $5,000) = $1,200,000 gross cost before any recovery. The total is an arithmetic illustration, not an estimate of real defect incidence. Forty losses sharing one release also violate the independence assumption in a simple vehicle-by-vehicle Poisson calculation. Aggregation, systemic deductibles, event definitions and reinsurance treatment require examination of actual contracts and exposure data.
A fourth illustration concerns split responsibility. A court may find an operator's improper geographic deployment and a software defect both causally relevant, while a separate driver or road user's conduct also affected the collision. A statement such as “manufacturer 60%, operator 40%” has no authority unless produced by a governing legal determination or properly framed settlement. Even then, that allocation of underlying liability may differ from insurance recovery because policy coverage, limits, exclusions, insolvent defendants and contractual contribution rules intervene. For analysis, legal shares and insurance cash flows must remain separately recorded.
Loss development, pricing, and concentration
Pricing an automated-vehicle portfolio requires more than collision frequency. A basic expected gross loss cost per insured exposure unit can be represented as expected covered claim count times mean covered cost, with exposure and coverages consistently defined. More complete models distinguish claim frequency, injury severity, property severity, allocated expenses, defense cost, recoveries and tail development. Symbols do not imply the existence of a validated predictive model. They simply expose the components that must be separately measured.
Automated operation could reduce some human-error-related collisions while increasing certain repair costs, concentration of technology risk, defense complexity or manufacturer recovery opportunities. Every claim requires empirical evidence about the specific population. Differences in repair costs can arise from sensor calibration, replacement parts, specialized labor, battery and electronic systems, vehicle downtime, and requirements for recommissioning. Statistical inference must also account for usage intensity, dense urban operation, passenger services, weather, speed and changed claims reporting. A simple before-and-after fleet comparison can confound deployment expansion with technological change.
Pricing uncertainty extends to legal development. If the probability of a successful product recovery is estimated from selected litigated cases, the sample is conditioned on disputes that reached litigation; routine settled claims may differ materially. Evidence preservation practices influence recoverability and potentially create selection in observed manufacturer recovery rates. Insurers should distinguish recovery amounts claimed, awarded, settled, collected and outstanding. A large theoretical recovery portfolio may provide little reduction of paid claims where proof, liquidity or enforcement is weak.
The IFRS 17 insurance-contract framework governs relevant financial reporting by insurers subject to that standard, including measurement of insurance contracts under its specified rules; it is not a legal determination that a software supplier caused a collision. Technical actuarial reserves, financial-statement figures, capital requirements and legal settlement obligations must therefore be reported with their own definitions. A manufacturer recovery, if recognized, requires its applicable accounting treatment and evidence; it cannot be assumed to offset all gross liability before realization.
Insurance solvency, claims handling, and incentives
Automated fleets may be operated by firms with significant technology dependence and changing commercial viability. Liability insurance plays a public-protection role where a liable party lacks liquid resources. The international insurance supervisory principles support risk-management and consumer-treatment analysis, including the insurer's capacity to assess aggregate exposure and respond to claims. In an insolvency or operational shutdown, the identities of the motor insurer, service operator, liable manufacturer and any statutory fund become especially important.
Compensation speed and recovery incentives can align or diverge. The insured or insurer may wish to preserve evidence for a later defect claim, while the victim requires prompt payment for medical expenses and lost income. Strong claims procedures record relevant vehicle logs promptly without delaying uncontested eligible compensation solely to resolve disputes among commercial defendants. Recovery rights can incentivize supplier accountability; their cost and probability affect the ultimate economics of automated deployment. The liability-incentive literature helps articulate the question, while the actual extent of deterrence remains an empirical matter.
A balanced supervisory response focuses on enforceable policy terms, access to necessary evidence, accurate claims communication, exposure concentration and financial resilience. Technological safety assessment and insurance sufficiency are complementary, separately auditable concerns. A well-approved system can still generate covered claims, and a financially sound insurer can compensate losses arising from a product defect for which eventual recovery is uncertain.
Evidence, Alternative Models, and Methodological Limits
Source hierarchy and legal effective dates
This investigation grades sources according to the proposition asserted. Primary statutes establish enacted rules, subject to scope, commencement, amendment and local implementation. Official regulators provide evidence of permits, reporting duties, approved configurations and regulatory interpretations. Technical standards define engineered concepts and performance tests within their own scopes. Scholarly articles offer legal arguments, reviews and methodological results. Accident data and insurer records support claims about incidence only when their population and ascertainment are known. No category can silently replace another.
Four time points must be distinguished: when legislation was enacted, when a provision came into force, when a vehicle or software was placed on the market or entered service, and when a collision occurred. In the European product-liability transition, these dates can determine which defect regime applies. In the United Kingdom, the government's 2027 implementation programme prevents a 2026 analysis from pretending that every authorization provision already operates in its future completed form. In Germany, currently effective text and regulatory amendments must be dated. Legal statements in this monograph are therefore descriptive at the stated November 16, 2026 cutoff, not guarantees of unchanged law for future claims.
Where authoritative legislation is accessible mainly through official online sections or compiled translations, the available text and its legal context must be kept explicit. The German citations identify individual statutory sections rather than implying that an English unofficial paraphrase has the force of the German original. The UNECE ALKS documentary revision itself states that a separately identified source constitutes the legally authentic text. These details are scientifically important because a plausible interpretation drawn from a search excerpt is not equivalent to the governing instrument.
Causal identification and measurement
A crash count is meaningful only with a denominator and observation mechanism. Let Y denote reported crashes, E comparable exposure, and p the ascertainment fraction. The observed rate Y/E may undercount true incident frequency when p is below one; changing p across periods can create a false trend. NHTSA expressly documents differences in reporting capabilities and changes to how engaged-system classifications were displayed in the Standing General Order data notes. This limitation is relevant even if every publicly displayed number was copied without transcription error.
The identification challenge also includes selection into automated service. Early deployments may choose geofenced routes, good weather, low-speed operation and extensive remote monitoring. A privately driven fleet in a contrasting environment is not an automatically valid control group. A credible comparison requires common outcome definitions, exposure units, time and geography, roadway mix, operating restrictions, data completeness and appropriate adjustment or matching. An engineering claim that automation eliminates one error mechanism cannot alone quantify the net crash effect after introducing other failures.
Insurance severity has separate ascertainment challenges. A crash in which technology reduces physical injury may still require expensive vehicle repairs; a severe claim may take years to develop in medical evidence, litigation and payment. A rising bodily injury average can reflect inflation, claim mix, legal developments or altered retention practices rather than an increase in physical crash severity. The rare-event statistical analysis supports caution about safety confirmation but does not yield a universal actuarial premium rating factor.
Sensitivity testing and transparency of assumptions
Hypothetical scenarios can demonstrate dependence without presenting invented estimates as observations. Suppose a fleet insurer expects 100 gross claims per 10 million annual vehicle miles at an assumed $30,000 average cost, for $3 million expected gross loss. If improved prevention reduces frequency by 20% while repair and injury severity increases by 15%, the illustrative expected cost is 80 × $34,500 = $2.76 million, a reduction of 8%. If severity rises by 30% instead, cost becomes 80 × $39,000 = $3.12 million, an increase of 4%. No empirical values are asserted; the calculations show how competing effects change the net outcome.
Suppose additionally that recoveries of manufacturer-related costs are pursued on 25 of those 80 hypothetical claims, with an assumed average face-value claim of $10,000 and a 50% realized collection rate. Expected collected amounts under these stipulated conditions equal 25 × $10,000 × 0.5 = $125,000. Treating all $250,000 of sought recovery as certain would overstate the offset by $125,000. Changing the collection rate to 20% lowers realized recovery to $50,000. This simple sensitivity exercise explains why product-defect evidence, legal enforceability and counterparty solvency are material actuarial variables.
An alternative calculation changes the unit of exposure. A commercial robotaxi operating 60,000 miles annually and a private vehicle operating 8,000 miles may share a vehicle-year but differ greatly in trip frequency, passenger exposure and collision opportunity. Rates per vehicle-year, mile, trip or passenger-mile answer different questions. No conversion is valid without measured mileage and use patterns. When frequency data are stratified by ODD and software version, small-cell credibility and concentration must be assessed. These are established methodological considerations rather than a newly introduced model.
Data privacy, auditability, and missing evidence
Automated vehicles can generate granular location, camera, telemetry and occupant information. Relevant evidence retention must respect rights and lawful purposes. The General Data Protection Regulation imposes requirements including lawfulness, purpose limitation and data minimization in its scope. Accident investigation should specify the legal basis for access and processing, select data proportionate to the disputed event, and implement retention and disclosure controls. An insurer's interest in defense or subrogation does not amount to unrestricted access to personal records from every sensor or third-party device.
Provenance questions are technical and legal. What firmware and machine-learning model version was installed? Who signed the update? Were data overwritten? Does a remote timestamp align with the onboard clock? Could a detector misclassify a road object? Was an adverse inference legally available from missing data? Can the manufacturer, insurer and claimant independently evaluate relevant conclusions without revealing unrelated trade secrets or personal data? A reproducible claims record documents the original source, transformation, chain of custody, competing interpretations and degree of uncertainty.
Research on automated vehicle risk faces an incomplete longitudinal record. Public NHTSA data do not provide a representative global accident census. California program documents concern participants in particular authorization categories. Germany's regime uses different operating concepts from the United Kingdom's authorization design. Product-liability law is evolving. Insurance policy forms and private recoveries may be unavailable. The 2026 comparative literature review helps identify legal research trajectories, but this monograph does not claim an exhaustive systematic review or proprietary claims-database validation. Its generalizable conclusions concern demonstrably distinct obligations and methods of inquiry; numerical safety and premium effects remain contingent on future evidence.
Alternative legal interpretations and adverse cases
A complete analysis tests explanations under which automation is incidental to the crash. A human-driven vehicle may be struck by a fully automated vehicle because a third party unexpectedly crosses a lane; an automated vehicle may be stationary when another driver collides with it. A compliant system may encounter a residual unavoidable hazard. Conversely, a vehicle may move autonomously outside its permitted area despite warnings and maintenance requirements. Each case challenges the assumption that the presence of a system fixes either cause or defendant.
The investigation also tests contrary coverage outcomes. A policy may respond to the victim subject to compulsory protection while excluding a separate contractual benefit to an insured who altered safety-critical software. A product defendant may be insolvent. A cyberattack may produce coverage disputes across different insurance lines. A recovery action may be time-barred or legally unavailable. A manufacturer may have fulfilled a recall duty but still face a contested product claim. The legal results require their own provisions and evidence, not analogies from unrelated jurisdictions.
Comparative Scientific Synthesis and Research Conclusions
Comparative legal findings
The jurisdictions examined show five important structural differences. First, the United Kingdom enacted an explicit insured-automated-vehicle compensation mechanism in the 2018 legislation and subsequently enacted a wider authorization framework whose full operational implementation is planned for 2027. Second, Germany preserves a consequential vehicle-holder liability basis under StVG section 7 while assigning technology-specific obligations to drivers and operational actors under separate provisions. Third, EU law supplies harmonized compulsory motor-insurance objectives and separate product-safety and product-liability instruments, with member-state implementation and temporal transitions. Fourth, U.S. federal safety obligations coexist with state-based liability and insurance regimes. Fifth, the Japanese Level 4 authorization example illustrates the limits of inferring insurance duties from approval status alone. These are comparisons of legal architecture, not rankings of national safety outcomes.
The common analytical invariant is a sequence of traceable questions: what activity was authorized, what system actually operated, what harm occurred, what legal compensation duty arose, what insurer or fund was bound to pay, and which final recoveries are enforceable. The order is methodologically useful because errors at one stage propagate to the next. A classifier that confuses a Level 2 assistance feature with a Level 4 automated service may incorrectly identify the human driving role. A claim system that confuses initial compulsory payment with product recovery may deny victims timely benefits. An actuarial analysis that treats demanded recoveries as collected cash may understate reserves.
Cross-border comparability improves when these questions are entered in a common evidence inventory while the legal answers remain local. Meaningful comparative variables include automation feature and ODD, actual control state, registered vehicle and use, operating permit, human and remote duties, applicable civil-liability source, claimant status, coverage type and limit, paying entity, elapsed time to payment, gross damages, legal and operational expenses, recoveries sought and collected, and the applicable statute's effective date. This is a documentation checklist derived from the reviewed sources, not a proprietary scientific theory or new regulatory doctrine.
Implications for insurance contracts, claims, and supervision
Insurers underwriting automated fleets should map material operational dependencies that conventional driver-age or garage-territory variables cannot fully describe. Vehicle model alone may conceal software releases, outsourced mapping, remote technical supervision, geofenced operating areas, shared maintenance providers and update practices. Contract analysis needs clarity about commercial deployment, insured entities, bodily-injury protection and interfaces with product-related insurance, while maintaining statutory victim safeguards. Claims systems need prompt collection of technically relevant data with privacy controls and credible independent dispute resolution.
Product-safety governance and insurance pricing require distinct feedback loops. Incident investigation may identify a defect and trigger corrective action or recall under federal motor-safety obligations or applicable European approval requirements. Insurers may revise expected frequency, severity or accumulation assumptions based on observed evidence. A software patch that addresses one failure mode should be evaluated for possible changes elsewhere in the ODD. The existence of insurance can finance compensation but cannot validate engineering safety by itself.
The distribution of risk between insurer and manufacturer has implications for incentives. A strong and legally effective right of recourse may transfer ultimate cost to a supplier whose defective design caused injury, supporting investment in safer products. High evidentiary costs or insolvency may make the same right economically weak. Where manufacturer recovery cannot be substantiated, an insurer must evaluate its gross obligations without assuming a collectible offset. The 2015 liability incentive analysis provides a scholarly starting point, and the working examples show how recovery probabilities change expected net cost. The normative desirability of a particular allocation is a separate policy judgment informed by victim protection, safety incentives, insurability and administrative costs.
Testable research questions
The next empirical task is to build privacy-compliant, multi-fleet matched exposure cohorts with collision outcomes stratified by ODD, automation level, system version, incident severity and claims maturity. It should measure the extent to which reporting practices differ and estimate sensitivity of observed results to missing records. Independent validation would be needed before comparing a technology fleet with conventional vehicles or recommending a rating factor.
A complementary legal and claims study would track real victim-facing payment times and subsequent manufacturer or operator recoveries. It should distinguish a filed demand from a collected recovery, classify claim adjudication outcomes by governing law and policy type, and document court or regulator interpretations. Such a study could test whether front-end compensation mechanisms reduce delay while preserving effective product accountability, and whether they shift the administrative cost of establishing software causation to insurers.
Finally, insurance accumulation research should measure cross-vehicle common causes: a shared software version, supplier outage, mapping error, cyber incident or operational supervision failure. Observed joint-loss distributions, reinsurance wordings and extreme-event scenarios would be needed to assess solvency exposure. These questions follow from existing insurance economics, engineering safety and comparative liability law and are presented as research needs rather than claims that a new model has been discovered.
Conclusion
Automated driving makes the time-specific distribution of the driving task central to crash reconstruction, while preserving distinct legal questions about liability, compulsory compensation, contractual coverage and ultimate economic loss. A technical taxonomy does not establish the defendant; administrative approval does not establish causation; a paid insurance claim does not establish a defective product; and a successful product recovery does not erase the original victim-facing compensation duty.
The United Kingdom, Germany, the European Union, the United States and Japan illustrate how different legal institutions address these questions. They share an interest in verifiable operational evidence and protection of road users, but distribute obligations through specific statutes, policy terms, operating approvals and legal processes. The transition to software-intensive mobility adds correlated system exposure and novel evidentiary tasks. Insurers must quantify gross obligations, distinguish uncertain from realized recoveries, and maintain defensible claim records capable of supporting fair payment and later recourse.
The scientific result is an original comparative interpretation grounded in official documents, established technical definitions, selective scholarly analysis and reproducible hypothetical arithmetic. The boundaries of that result are explicit: national law must be verified for the relevant date and claim, observed crash counts cannot replace exposure-adjusted safety analysis, and no numerical premium or safety improvement is inferred from a hypothetical example. A rigorous insurance allocation follows the actual event, the applicable duty, the covered interest, the paying obligation and the enforceable subsequent recovery, with each conclusion supported by its own evidence.
References
SAE International (2021). J3016_202104, Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles. SAE International. https://saemobilus.sae.org/standards/j3016_202104-taxonomy-definitions-terms-related-driving-automation-systems-road-motor-vehicles
National Highway Traffic Safety Administration (NHTSA) (2021–2026). Standing General Order on Crash Reporting, as amended 2025; public data and methodological limitations. U.S. Department of Transportation. https://www.nhtsa.gov/laws-regulations/standing-general-order-crash-reporting
United Nations Economic Commission for Europe (UNECE) (2025). UN Regulation No. 157, Revision 1, Uniform Provisions Concerning the Approval of Vehicles with Regard to Automated Lane Keeping Systems; documentary text dated March 27, 2025, identifying ECE/TRANS/WP.29/2022/59/Rev.1 as authentic text. https://unece.org/sites/default/files/2025-06/R157r1e.pdf
United Kingdom Parliament (2018). Automated and Electric Vehicles Act 2018, Part 1, including sections 1–5. UK Public General Acts 2018 c. 18. https://www.legislation.gov.uk/ukpga/2018/18/part/1/enacted
United Kingdom Parliament (2024). Automated Vehicles Act 2024, enacted text. UK Public General Acts 2024 c. 10. https://www.legislation.gov.uk/ukpga/2024/10/body/enacted?view=plain
UK Centre for Connected and Autonomous Vehicles (2025). Automated Vehicles Act Implementation Programme. UK Department for Transport. https://www.gov.uk/guidance/automated-vehicles-act-implementation-programme
Law Commission of England and Wales and Scottish Law Commission (2022). Automated Vehicles: Joint Project and Recommendations. Law Commission. https://lawcom.gov.uk/project/automated-vehicles/
Federal Republic of Germany (current official consolidated text, accessed 2026). Straßenverkehrsgesetz (StVG), § 1a, Kraftfahrzeuge mit hoch- oder vollautomatisierter Fahrfunktion. https://www.gesetze-im-internet.de/stvg/__1a.html
Federal Republic of Germany (current official consolidated text, accessed 2026). Straßenverkehrsgesetz (StVG), § 1b, Rechte und Pflichten des Fahrzeugführers. https://www.gesetze-im-internet.de/stvg/__1b.html
Federal Republic of Germany (current official consolidated text, accessed 2026). Straßenverkehrsgesetz (StVG), § 1d, Kraftfahrzeuge mit autonomer Fahrfunktion in festgelegten Betriebsbereichen. https://www.gesetze-im-internet.de/stvg/__1d.html
Federal Republic of Germany (current official consolidated text, accessed 2026). Straßenverkehrsgesetz (StVG), § 1f, Pflichten der am Betrieb Beteiligten. https://www.gesetze-im-internet.de/stvg/__1f.html
Federal Republic of Germany (current official consolidated text, accessed 2026). Straßenverkehrsgesetz (StVG), § 1g, Datenverarbeitung. https://www.gesetze-im-internet.de/stvg/__1g.html
Federal Republic of Germany (current official consolidated text, accessed 2026). Straßenverkehrsgesetz (StVG), § 7, Haftung des Halters, Schwarzfahrt. https://www.gesetze-im-internet.de/stvg/__7.html
Federal Republic of Germany (current official consolidated text, accessed 2026). Straßenverkehrsgesetz (StVG), § 63a, Datenverarbeitung bei Kraftfahrzeugen mit hoch- oder vollautomatisierter Fahrfunktion. https://www.gesetze-im-internet.de/stvg/__63a.html
Federal Republic of Germany (2022, as amended). Autonome-Fahrzeuge-Genehmigungs-und-Betriebs-Verordnung (AFGBV): approval and operation of autonomous vehicles in designated operating areas. https://www.gesetze-im-internet.de/afgbv/BJNR098610022.html
European Parliament and Council (2021). Directive (EU) 2021/2118 amending Directive 2009/103/EC relating to insurance against civil liability in respect of the use of motor vehicles and enforcement of the obligation to insure against such liability. Official Journal of the European Union. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32021L2118
European Parliament and Council (2024). Directive (EU) 2024/2853 on liability for defective products and repealing Council Directive 85/374/EEC. Official Journal of the European Union. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024L2853
European Union (2026). Corrigendum to Directive (EU) 2024/2853, published May 7, 2026, including applicability wording for products placed on the market or put into service after December 8, 2026. Official Journal of the European Union. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024L2853R(01)
European Parliament and Council (2019). Regulation (EU) 2019/2144 on type-approval requirements for the general safety of motor vehicles and systems. Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2019/2144/oj/eng
European Commission (2022). Commission Implementing Regulation (EU) 2022/1426 laying down rules for application of Regulation (EU) 2019/2144 as regards uniform procedures and technical specifications for type approval of automated driving systems of fully automated vehicles. Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg_impl/2022/1426/oj/eng
United States Congress. 49 U.S.C. § 30118, notification of defects and noncompliance. United States Code, current codification. https://www.law.cornell.edu/uscode/text/49/30118
National Association of Insurance Commissioners (2026). What You Should Know About Auto Insurance Coverage. NAIC Consumer Information. https://content.naic.org/article/what-you-should-know-about-auto-insurance-coverage
California Department of Motor Vehicles (2026 access). Autonomous Vehicles Program; testing, deployment and regulatory information. State of California. https://www.dmv.ca.gov/portal/vehicle-industry-services/autonomous-vehicles/
California Department of Motor Vehicles (2026 access). Autonomous Vehicles Program Permit Resources. State of California. https://www.dmv.ca.gov/portal/vehicle-industry-services/autonomous-vehicles/autonomous-vehicles-program-permit-resources/
Ministry of Land, Infrastructure, Transport and Tourism, Japan (2023). Approval of first specified automated operation under Level 4 framework (March 30–31 announcement). Japanese Government. https://www.mlit.go.jp/report/press/jidosha07_hh_000442.html
Kalra, N., and Paddock, S. M. (2016). Driving to safety: How many miles of driving would it take to demonstrate autonomous vehicle reliability? Transportation Research Part A: Policy and Practice, 94, 182–193. https://doi.org/10.1016/j.tra.2016.09.010. https://doi.org/10.1016/j.tra.2016.09.010
Schellekens, M. (2015). Self-driving cars and the chilling effect of liability law. Computer Law & Security Review, 31(4), 506–517. https://doi.org/10.1016/j.clsr.2015.05.012. https://doi.org/10.1016/j.clsr.2015.05.012
Roussou, J., Ziakopoulos, A., Kontaxi, A., and Yannis, G. (2026). Literature review of legal frameworks and liability allocation in connected and automated vehicles. Research in Transportation Business & Management. https://doi.org/10.1016/j.rtbm.2026.101804. https://doi.org/10.1016/j.rtbm.2026.101804
European Parliament and Council (2016). Regulation (EU) 2016/679 (General Data Protection Regulation), especially Articles 5–6. Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng
World Health Organization (2023). Global Status Report on Road Safety 2023. WHO, Geneva. https://www.who.int/teams/social-determinants-of-health/safety-and-mobility/global-status-report-on-road-safety-2023
International Association of Insurance Supervisors (2026 access). Insurance Core Principles and ComFrame: online standards tool, especially ICP 19 (conduct of business) and ICP 16 (enterprise risk management). IAIS. https://www.iais.org/icp-online-tool/
IFRS Foundation (2023–2026). IFRS 17 Insurance Contracts, official standard overview and effective date. IFRS Foundation. https://www.ifrs.org/issued-standards/list-of-standards/ifrs-17-insurance-contracts/