RESEARCH DOSSIER
Connected Vehicle Data Ownership and Motor Insurance Privacy
Abstract. Connected vehicles produce continuously renewable records of movement, machine state, driver inputs, communications, and safety events. These records support maintenance, road safety, crash reconstruction, and insurance underwriting, yet their evidentiary and economic value depends on who generates them, which person or organization may obtain them, how they are transformed, and the legal purpose for which an insurer uses them. The term data ownership often compresses several independent relationships: a vehicle user's statutory data-access entitlements, the personal-data rights of drivers and passengers, contractual control over a platform, intellectual-property or trade-secret claims, and a recipient's permission to process information for insurance purposes. This monograph examines these relationships through a comparative analysis of European data-access and privacy law, United States regulatory enforcement and consumer reporting, selected international privacy regimes, vehicle cybersecurity governance, and the actuarial literature on telematics. It distinguishes raw and preprocessed vehicle data from derived indicators and insurance risk scores; maps the pathway from a recorded driving event to a possible premium, underwriting, or claim decision; and analyzes consent, alternative lawful bases, purpose limitation, reidentification, security, evidentiary reliability, and selection. Two transparent hypothetical calculations illustrate why partial participation and event misclassification can distort apparently precise insurance scores. The central finding is that reliable governance requires separately justified access, processing, transfer, inference, and decision rights at each stage. The analysis offers a source-grounded comparative synthesis and operational research questions without proposing a new theory, doctrine, or universal rule of data ownership.
Keywords: Connected vehicles; automobile insurance; motor insurance; telematics; data access; data ownership; EU Data Act; GDPR; consent; lawful basis; secondary use; driving behavior; risk scoring; insurance fairness; data provenance; cybersecurity; consumer reporting; privacy governance.
Introduction
A connected automobile is simultaneously a physical risk object, a mobile sensing platform, a communications endpoint, and an insured asset. Its electronics may record velocity, acceleration, braking, steering, airbag activation, diagnostics, location, timestamps, service events, and interactions with applications or cloud systems. The relevant data can support different decisions about a single vehicle: preventive maintenance, roadside assistance, crash investigation, fraud detection, premium calculation, liability assessment, or warranty management. The NAIC telematics overview identifies mileage, driving time, GPS location, acceleration, braking, cornering, and airbag events among the variables used in usage-based motor insurance. The same observation becomes legally and statistically different when used for immediate roadside assistance, an individualized premium adjustment, or transmission to a consumer reporting agency.
The scientific object of this Research Dossier is the governance of connected-vehicle data as they move toward motor insurance decisions. The object has three irreducible empirical dimensions. Provenance determines what was measured, by which instrument, under whose physical control, and with which transformations. Processing authority determines which actors may collect, access, store, infer from, and share the data for defined purposes. Consequential use determines whether the data influence a quote, renewal, premium, coverage restriction, claim, fraud referral, or dispute. The 2025 European Commission vehicle-data guidance confirms the importance of vehicle-specific access questions under the Data Act. The EDPB connected-vehicle guidelines situate the same industry within data-protection obligations.
The unit of analysis must be specified for every claim. It may be the vehicle, the trip, the insured policy, the actual human driver, a household, a commercial fleet, an insurer's book, or a jurisdiction. Vehicle identification does not necessarily identify a particular driver; identifying a policyholder does not establish who drove at a given moment; and a record relevant to a collision does not automatically establish the insured's legal responsibility. Throughout this work, insurer includes organizations making or receiving motor insurance decisions, while distinct obligations of manufacturers, platforms, brokers, analytics suppliers, and consumer reporting agencies are separately identified.
The method is doctrinal and comparative analysis of original legal instruments, regulator documents, and selected scientific studies, coupled with elementary, fully disclosed analytical calculations. Sources are evaluated at the level they support: legislative text establishes legal rules; an agency order establishes the order's terms; an allegation identifies what was alleged; a study describes its methods and observed associations; and an illustration demonstrates mathematical possibility. The review is internationally oriented, with concentrated legal treatment of the European Union and United States and a bounded comparison with Brazil, Japan, and the United Kingdom. No proprietary telematics feed, insurer dataset, individual driver history, or original accident file was analyzed.
Scientific Definitions and Primary Foundations
Connected-vehicle data and the insurance information chain
A connected-vehicle record is information generated, observed, stored, processed, or transmitted through a vehicle's electronic systems or connected services. The underlying measurement may originate in a vehicle sensor, smartphone, telematics dongle, roadside system, repair device, navigation service, or human entry. Some devices report continuously; others transmit selected events. An insurer may receive a manufacturer-generated extract rather than the physical sensor measurement itself. The Data Act's legal definitions of connected products and product data govern a defined European access regime; they do not exhaust all potentially relevant vehicle information in insurance practice.
The information chain may be described in ordinary sequence: physical event; sensor sampling; local software processing; storage; transmission; attribution to a vehicle or person; feature engineering; actuarial or business scoring; administrative decision; and eventual retention, sharing, correction, or deletion. Every stage changes the available evidence. A hard-braking flag represents an interpretation of sampled motion under a threshold, not a direct observation of negligent driving. A risk score summarizes a model's output, not an independently established future loss. A claims photograph and an event-data trace may concern the same crash but have different chains of custody and different uncertainty.
What the word ownership does and does not resolve
In ordinary speech, vehicle data may be said to belong to the driver, owner, manufacturer, or service provider. As a comparative legal question, a single comprehensive title to every item of information is a poor description of existing regimes. A user may possess statutory rights to access and transmit product data while another person remains the data subject of personal information contained in those data. A manufacturer may operate the collection infrastructure while an insurer acts as an independent controller of its subsequent decision process. A platform contract may allocate database access while mandatory privacy law limits personal-data processing. These relationships can overlap without collapsing into property ownership.
The EU Data Act, Regulation 2023/2854 gives qualifying users access to certain readily available data from connected products and related services and establishes conditions for third-party sharing. In particular, the user of a connected product is a legally defined role, which can include a person or legal entity entitled through ownership, rent, or lease; it is not necessarily the individual driving when a record is created. The GDPR's data-subject protections protect identifiable natural persons under their own definitions and rules. Consequently, a fleet operator's product-data access request and an employee driver's personal-data rights require different analysis. A vehicle user's instruction to share data with an insurer does not by itself supply every GDPR legal basis needed for personal data relating to other people.
This distinction also applies to inference. Data subject access, portability, and transparency rights have specific statutory limits. The right to obtain certain observed product data under the Data Act does not automatically require disclosure of every proprietary actuarial model, complex derived score, trade secret, or confidential investigation. Equally, a contractual assertion that a proprietary model owns its outputs does not remove all duties concerning the personal information used to produce them. In each case the relevant legal provisions and the exact category of information must be identified.
The insurance decision is a separate legal event
Motor insurance obligations arise from policies, compulsory liability rules, and local supervisory requirements. A telemetry value may inform the risk classification but does not itself vary coverage, establish fault, or determine indemnity. The NAIC account of usage-based insurance describes behavior-linked rating practice; whether a particular variable can enter a regulated rate depends on the applicable jurisdiction and approved rating rules. For example, California authorities expressly distinguish their state's rules against using driving data to set private-passenger auto insurance premiums from the privacy enforcement issues arising when data are sold to brokers, as explained in the California GM settlement record.
For analysis, distinguish a sensor event, a scoring inference, an insurer's decision, and a legal obligation. The first can be technically accurate while the second is statistically uncalibrated; the second can be predictive while the third exceeds an authorized purpose; and the third can be lawful while an injured person's compensation rights are determined under independent motor liability statutes. The scientific subject therefore includes the interfaces between these different stages.
Historical and institutional development
Usage-based insurance expanded as vehicle sensors, GPS, cellular communications, smartphones, and cloud analytics made mileage and driving behavior economically observable. Earlier motor rating relied primarily on variables such as claims history, location, vehicle characteristics, and permitted classification factors. Telematics creates a second channel: repeated observations of exposure and conduct during coverage. The Gao, Meng, and Wüthrich (2022) survey of actuarial telematics research discusses data cleaning, trip attribution, privacy, and claims-frequency prediction. Its treatment shows that the analytical challenges of feature construction and identifying drivers are central to the insurance use case, rather than peripheral matters of technology implementation.
Connected cars also distribute decisions among institutions. Original equipment manufacturers and cloud contractors may control ingestion. Insurers select rating rules and products. Analytics firms provide scores or data extracts. Consumer reporting agencies may aggregate, furnish, and dispute information. Supervisors and privacy authorities regulate different activities. The CFPB's description of Telematics OnDemand confirms the existence of a U.S. consumer-reporting route for driving behavior used in auto insurance pricing. A single insurance score can therefore depend on contractual and regulatory relationships extending far beyond the insurer-policyholder pair.
Vehicle Sensor Data Provenance
Measurement, sampling, and the limits of rawness
Vehicle data are generated through instruments with finite resolution, temporal precision, sampling intervals, firmware, calibration, and failure modes. A road-speed value recorded at one-second intervals cannot reconstruct every brief maneuver; GPS coordinates may wander in urban canyons or under poor satellite reception; a phone may register movement while its owner is a passenger. Vehicle CAN-bus signals, accelerometers, and smartphone-derived estimates differ in provenance. A label such as raw data is therefore relative to a processing stage: a vehicle controller may already have filtered the underlying sensor signal before exporting it.
Reconstructing provenance requires at least the source device and firmware version, measurement unit, sampling and missing-data rules, clock handling, transformations, event thresholds, driver or vehicle assignment, transmission receipt, retention history, and accessible audit metadata. These are criteria for interpreting evidence, not a proposed new legal standard. A claim that a driver exceeded a speed threshold must be tested against the specific sensor, map, speed-limit reference, and timestamp reconciliation. An apparent aggressive-braking episode requires conditions such as a pedestrian crossing, road grade, weather, and traffic density before it can be treated as a behavioral interpretation.
Ma, Zhu, Hu, and Chiu's context-sensitive insurance telematics study illustrates the research use of detailed driving context in rate-making analysis. Its statistical findings do not create a universal causal mapping between each recorded maneuver and moral fault, individual blame, or entitlement to insurance. This boundary is essential where commercial pricing categories are later presented to policyholders as objective measurements of driving quality.
Driver attribution and multiple occupants
A car frequently has more than one driver over its lifetime, while a smartphone may accompany a passenger, ride-share occupant, or employee using a fleet vehicle. A vehicle-level observation is a driver-level observation only after a justified attribution step. Driver recognition from behavior itself can be technically possible, which introduces an additional privacy consequence: ostensibly pseudonymous time-series traces can act as behavioral signatures. The Gao, Meng, and Wüthrich (2022) survey describes the use of neural methods to distinguish driver patterns from driving data. This observation matters when an organization claims that stripping a name from trip records makes them nonpersonal.
Pseudonymization reduces direct identification but leaves the possibility of reidentification with reasonably available additional information. In a shared car, a vehicle account holder, policyholder, human driver, child passenger, and occasional guest have different interests and sometimes different rights. The EDPB's mobility guidance analyzes vehicle ecosystems where several actors and devices may participate. A consent interface addressed only to the registered owner therefore cannot be assumed to resolve all occupants' personal-data rights.
Access under the European Data Act
The Data Act's connected-product rules are especially relevant to aftermarket insurance services. Subject to statutory scope and exceptions, users must be enabled to access certain readily available product data and related service data, and may request transmission to third parties. Article 4 governs user access; Article 5 addresses sharing with a user-designated third party; Article 6 constrains the third party's subsequent uses. The Commission's automotive guidance of 2025 explains Chapter II applications to vehicle data and is an official interpretive guide, while the Data Act itself is the legislative source.
The distinction between data made readily available and more substantially derived information is material. A manufacturer may hold inferences or proprietary engineering assessments generated by sophisticated processing; these are not interchangeable with the product's accessible sensor outputs for purposes of statutory access. Likewise, trade-secret safeguards and technical constraints must be evaluated under the law's actual provisions. A hypothetical insurer requesting every factory model prediction and all raw internal test data would be asking a different question from an insurer receiving mileage and event records on a user's lawful instruction.
The Data Act generally applies from September 12, 2025, with staggered application for specified provisions and product-design obligations. It should be read with transitional provisions rather than represented as an unconditional universal duty imposed on every pre-existing vehicle in exactly the same form. The EU legislative summary provides the general application date, while precise exceptions belong to the Regulation.
Provenance in claims and litigation
Telemetry can assist crash timing, vehicle state, impact reconstruction, and suspected fraud review. Its relevance depends on whether the insured peril, injury, liability, repair cost, or policy condition is in dispute. A reconstruction that uses a stored event record must record its provenance, the access method, integrity check, event definitions, calibration assumptions, clock alignment, chain of custody, and possible data gaps. These methods follow general evidentiary reasoning; admissibility, production, privilege, and insurer disclosure obligations are jurisdiction-specific.
Consider a vehicle whose systems report a sudden deceleration while a crash file reports a rear-end impact. The observation supports that the vehicle decelerated under the recorder's detection logic. Determining whether braking was too late, whether collision avoidance was possible, and whether legal negligence occurred requires additional scene evidence, vehicle dynamics, traffic law, and potentially expert reconstruction. A loss adjuster should preserve both the machine record and its limitations so that apparently precise digital evidence does not displace the broader factual inquiry.
Integrity across platform changes
Connected-vehicle systems are modified by firmware updates, sensor replacements, new supplier pipelines, and changes in data schemas. A variable with the same display name may acquire a different calculation after an update. Longitudinal actuarial analysis can then mistake a processing change for a behavioral shift. The UK VCA account of UN Regulations 155 and 156 addresses vehicle cybersecurity management and software-update management as regulated processes. These requirements help locate a source of technical governance; they do not, by themselves, establish a lawful basis for an insurer to process driver location data.
Data lineage therefore matters for claims about change over time. Where a model is trained on pre-update signals and applied after a firmware change, the analyst must test measurement equivalence, missingness, and recalibration. Comparability is an empirical result, not a presumption established by unchanged field names.
Consent and Processing Purposes
Separating product-data access from personal-data processing
A vehicle user may authorize an insurance company to receive selected connected-product data. If those data contain personal information, the manufacturer and the insurer must separately satisfy applicable data-protection requirements. Under the GDPR's Article 6 framework, processing requires an identified lawful basis; consent is one available basis among others, and contractual necessity, legal obligations, and legitimate interests have their own conditions and limits. The Data Act's rules for third parties include purpose constraints. Neither instrument should be collapsed into a claim that signing an insurance telematics form transfers general ownership of all future vehicle information.
A defensible analysis names the actor, data category, specific purpose, applicable legal basis, retention period, recipients, and individual rights. The purpose of measuring annual mileage for a disclosed pay-per-mile option differs from collecting second-by-second coordinates for broad insurance marketing. The former may be achievable with a single cumulative figure; the latter requires an independent necessity and proportionality assessment. Where processing is based on consent, its legal validity and withdrawal consequences must be examined under the controlling law rather than inferred from the existence of a touchscreen acknowledgment.
Consent as an operational event
Meaningful consent depends on what was represented at the moment of the decision, including specificity, freely given choice, and intelligible information where the relevant law requires those features. In practice, one enrollment flow may cover connectivity services, navigation, maintenance, personalized advertising, accident assistance, risk scoring, and data-broker transfers. Combining these choices can obscure the distinction between services necessary to operate the vehicle and optional commercial uses. The EDPB guidelines emphasize context-sensitive privacy-by-design approaches, including locally processed data where feasible.
Consent must also be understood as time-dependent. A person who agrees to usage-based rating for the present policy may later change insurers, sell the car, withdraw consent, or discover that a shared vehicle's previous owner remained connected to the app. Data collected for prior contractual performance may be retained where another lawful obligation justifies that retention; continued optional marketing or unrelated profiling requires its own authority. The scientific record should separately track cessation of new collection, continued historical retention, and suppression of future onward disclosure.
Device access and ePrivacy
European connected vehicles can fall within the legal analysis of terminal equipment for certain ePrivacy Directive purposes. The ePrivacy Directive 2002/58/EC establishes rules for access to or storage of information on terminal equipment, subject to the Directive's conditions and exceptions. The EDPB connected-vehicle guidelines discuss this issue as distinct from the lawful basis for subsequent personal-data processing under the GDPR. A valid basis at one stage does not silently dispense with obligations at the other.
For an insurer, this distinction influences the technical route of collection. Reading vehicle-local storage with a dedicated app, obtaining a cloud export that has already been lawfully collected, and asking the user to download a history may implicate different acts and entities. A compliant implementation requires the precise facts of access, technology, and applicable national transposition, not merely a generic assertion that data are already stored somewhere.
Data minimization and purpose compatibility
The GDPR's principles of purpose limitation and minimization connect lawful processing to specified uses and necessary data. An insurer evaluating annual exposure might need verified mileage and coverage dates; it does not necessarily need exact night-by-night destinations. A claims investigation might require narrowly bounded location or crash information for one incident but have limited justification to inspect months of unrelated trips. The legal inquiry becomes more demanding when a company proposes using a retained dataset for a materially new purpose, such as training an unrelated commercial classifier or transferring detailed journeys to a third-party broker.
Privacy-preserving design can change the amount of information disclosed without eliminating insurance usefulness. An on-device computation may produce an aggregate distance figure, bounded risk-related counts, or a verifiable claims-event extract. The recipient receives a result adequate for the insurance purpose while reducing unnecessary raw trace dissemination. This is a design possibility, not evidence that every aggregate remains anonymous: repeated summaries can reveal patterns, and a score may itself be personal data if associated with an identifiable person.
Automated scoring and individual rights
Motor insurers use diverse decision systems, from simple mileage discounts and manual underwriting to algorithmically generated offers. Where personal data are involved, the GDPR's Article 22 conditions become relevant specifically to decisions based solely on automated processing that produce legal or similarly significant effects, subject to the provision's exceptions and safeguards. An algorithm's involvement alone does not prove that Article 22 applies to each insurance quote. Equally, nominal human oversight does not establish meaningful involvement without examining the actual decision process.
Transparency and challenge procedures have scientific as well as legal value. A driver may dispute an attributed trip, a speed-limit database error, a device carried as a passenger, or a score affected by a missing sensor period. A correction mechanism must be able to identify the underlying error and its impact on the decision, not merely replace a complaint with a generic customer-service script. Specific access, rectification, objection, portability, and erasure entitlements follow their respective provisions and exceptions rather than a single unlimited right of deletion.
Multiperson vehicles and organizational users
Household, rental, fleet, and ride-share contexts reveal why consent by one account administrator cannot simply bind everyone who generates data. A company may be the Data Act user and contract holder; a driver may be an employee; another traveler may be a passenger whose precise location is exposed. Such situations engage distinct labor, privacy, contract, and insurance considerations. The UK ICO's 2025 technology horizons report treats connected transport as an environment in which vehicle-derived data and organizational systems create privacy risks.
A fleet contract that promises broad telematics access is relevant to commercial rights, yet its presence cannot alone determine every worker's privacy protections. Conversely, limiting driver-level collection can affect loss-prevention feedback, accident investigation, and insurance pricing, so institutions must specify precisely what functions they retain and what data they omit. The proper comparison is between alternatives that accomplish a defined purpose with different data granularity.
Secondary Use and Cyber Exposure
Onward disclosure and the U.S. connected-car enforcement record
A central risk arises when information collected during vehicle service is passed into an insurance reporting and pricing ecosystem not apparent to the individual at the original interface. In January 2026, the FTC finalized an order addressing allegations about General Motors and OnStar's collection and sharing of location and driving data. The FTC's final order announcement describes a five-year ban on certain disclosures to consumer reporting agencies and longer-term requirements involving affirmative consent and consumer control, with specified exceptions. The official case docket identifies the January 14, 2026 decision and order. An FTC settlement resolves allegations through the terms of an order; it should not be treated as a judicial finding that every allegation was independently proven at trial.
California's separate action illustrates overlapping but distinct regulatory interests. The May 2026 California GM announcement reported a $12.75 million settlement addressing alleged sales of California drivers' location and behavioral data and included restrictions on further sales. The announcement described the agreement as subject to court approval. The state's privacy enforcement register records the terms and their relation to purpose limitation and minimization. The original reports distinguish the alleged privacy harm from the assertion that the data actually caused premium increases for California motorists, because California's private-passenger rating law has specific restrictions on driving-behavior variables. The same dataset can therefore implicate privacy harms in one market even where a particular insurance rating use is restricted.
Consumer reporting and disputes
A vehicle-data intermediary may develop consumer reports or indicators used by insurance companies for underwriting. The U.S. CFPB's consumer-reporting directory explicitly describes LexisNexis C.L.U.E. and Telematics OnDemand as sources of insurance claims and driving-behavior information. When a covered consumer report is involved, the Fair Credit Reporting Act can require accuracy, permissible purposes, access, dispute processes, and adverse-action notices according to their statutory conditions. The FTC's FCRA statutory resource provides the federal legal framework. Classification depends on the provider, report, recipient, and purpose; it should not be asserted for every direct insurer-device data transmission.
The scientific significance of this intermediary layer is that an erroneous trip assignment or misclassified driving event can propagate into more than one underwriting decision. Correction at the first collector may be insufficient when downstream entities retain cached versions or separately generated scores. A robust dispute inquiry needs to identify where the information entered the chain, which transformation occurred, what the receiving decision-maker relied upon, and whether the correction propagated to subsequent recipients.
California privacy design and asymmetric choice
In March 2025, the California Privacy Protection Agency announced a decision concerning American Honda Motor Co. The Honda enforcement announcement states that the agency required changes to privacy practices and a $632,500 payment, following allegations concerning excessive verification burdens, asymmetrical consent-choice interfaces, and difficulties using authorized agents. These are concrete issues of consumer control. A nominally available opt-out can be less usable than an opt-in if its path requires more steps, more personal information, or repeated contact.
For connected-vehicle insurance, interface design is part of the observed consent mechanism. Researchers should record which settings are presented, the sequence of enrollment, whether a second purpose is hidden in a broad term, the friction of withdrawal, and the treatment of household users. A policyholder's recorded click without the surrounding interface context is a weak basis for claims about informed choice.
Cybersecurity and software-update governance
A connected-vehicle data pipeline expands attack surfaces to in-vehicle networks, mobile applications, cloud APIs, repair interfaces, scoring contractors, and insurer systems. Threats include unauthorized extraction of location histories, identity-linked trip data leaks, credential misuse, compromised vehicle applications, falsified event submissions, and ransomware affecting an insurer's evidence availability. UN Regulations 155 and 156 address vehicle cybersecurity-management and software-update-management obligations within their relevant approval frameworks, as described by the Vehicle Certification Agency. These engineering regimes are complementary to privacy duties and insurer security governance; certification of a vehicle-management process does not certify every downstream insurer's privacy compliance.
Security and data minimization interact. Retaining precise journeys indefinitely increases the exposure if a database is compromised and makes reidentification easier. Deleting all event history immediately, however, may undermine a legally justified claim investigation or a required audit trail. A sound retention decision identifies a purpose-specific minimum, documented lifecycle, access controls, logging, disposal, and the legitimate requirements that justify exceptions. The strongest privacy posture cannot be assessed from a single encryption claim; security depends on key management, endpoint authorization, data volume, vendor access, change control, incident detection, and response.
Inference and reidentification through driving patterns
Sensitive information can be recovered from records that lack obvious names. A sequence of origins, destinations, and travel times may reveal home, work, school routines, medical visits, social interactions, or unique commuting patterns. Stable driving-style vectors may also distinguish individuals. The Gao, Meng, and Wüthrich (2022) survey discusses a method for assigning trips to drivers from driving patterns, illustrating how a technique useful for improving data quality can also weaken claims of anonymity. This supports a general methodological caution: removing direct identifiers is a transformation whose protection must be evaluated against likely linkages and adversarial knowledge.
Privacy tests should therefore distinguish genuinely anonymous aggregate statistics, pseudonymous personal records, reversible identifiers, and event records containing precise geography. Labels such as deidentified or aggregated are insufficient without knowledge of group size, released dimensions, time window, external auxiliary data, and the incentives of possible recipients. Because telematics traces are longitudinal and highly structured, the privacy effect of releasing a series can differ substantially from that of publishing one isolated observation.
Cross-border data governance
International automotive manufacturing and reinsurance operate across legal boundaries. A vehicle may be produced in one country, registered in another, connected to a cloud provider elsewhere, and insured by a domestic carrier employing a multinational service vendor. The data's physical route does not determine all legal obligations. Under EU law, the GDPR's territorial scope and international-transfer rules require their own analysis; the Data Act establishes different data access and sharing arrangements. Under Brazilian law, Lei 13.709/2018 (LGPD) establishes personal-data rights and processing principles in a separate legal system. Japan's Personal Information Protection Commission legal materials provide an institutional entry point into the APPI framework, with the current Japanese law and amendment record controlling precise local interpretation.
The United Kingdom retains its own data-protection and connected-transport governance after leaving the European Union. The ICO's connected-transport discussion identifies practical privacy issues but is a research and supervisory perspective, not a substitute for the UK GDPR or applicable statutory instruments. An insurer cannot infer that a lawful EU transfer automatically confers unrestricted rights to repurpose data in Brazil, the United States, or the United Kingdom. Each receiving activity needs a jurisdiction- and role-specific assessment.
Evidence, Alternative Models, and Methodological Limits
Actuarial significance and evidence quality
The central actuarial question is whether additional measured behavior provides out-of-sample predictive value for a defined insured loss, conditional on exposure, territory, vehicle, policy terms, and existing rating variables. There are related but distinct questions: whether a factor is predictive, whether it causally reflects a safer behavior, whether the data were lawfully obtained, and whether the factor is admissible for rating in the applicable jurisdiction. No amount of model performance can substitute for answers to the last two questions.
The Boylan, Meyer, and Chen (2024) review included 45 studies meeting its eligibility criteria, of which 27 received a good-quality assessment. These counts describe the reviewed literature and its evaluative method; they do not establish a universal percentage improvement in motor loss ratios or a single validated intervention effect. The Gao, Meng, and Wüthrich (2022) survey surveys trip-based and heatmap-based claims-frequency approaches while discussing data cleaning and transparency. The studies support research feasibility and methodological concerns, not the assumption that a particular insurer's new score is sufficiently calibrated for every population.
Insurer portfolios differ by age distribution, vehicle type, urban density, coverage, deductible, claims-reporting behavior, and driving exposure. Model performance observed in one participating user group must be tested again in a different group and period. An insurer might improve its predictive model while also excluding consumers with older vehicles, incompatible telematics hardware, or limited digital access. These are measurable selection effects rather than abstract concerns.
Hypothetical participation-selection arithmetic
Consider a deliberately hypothetical group of 1,000 insured drivers with identical coverages and a common exposure period. Assume 600 elect to share telematics data, and 400 do not. Within the volunteers, suppose 30 expected claims are observed; within the nonvolunteers, 32 expected claims are observed. These assumed counts give observed frequencies of 30/600 = 0.050 and 32/400 = 0.080 claims per driver-year, with a combined frequency of 62/1,000 = 0.062. If an analyst trains and evaluates only on volunteers and announces a market-wide expected frequency of 0.050, the implied 50 claims per 1,000 underestimate the constructed portfolio count by 12, a 19.35% shortfall relative to 62. The arithmetic is exact for the stated example; the assumed frequencies are not empirical estimates of real-world volunteer or nonvolunteer behavior.
The example demonstrates that selected participation and population prediction are separate objects. A real analysis would need to assess selection by risk, device availability, incentives, age, access, and driving preferences. Inverse-probability methods or stratification can improve transportability only when selection probabilities and relevant confounders are sufficiently measured and positivity conditions hold. In the example, the opt-in rate of 60% is an exposure characteristic, not proof that the nonparticipants' losses can be inferred from volunteer telemetry.
An insurer cannot automatically translate the example's difference into evidence that consumers who refuse tracking are riskier. That would require empirical controls and consideration of competing explanations: more privacy-sensitive drivers could be safer, older vehicles could be harder to connect, or pricing incentives could change participation. The causal direction is unresolved by the example. Disclosure and fairness analysis should treat selection as a testable measurement problem.
Hypothetical event-classification arithmetic
Consider another explicitly hypothetical dataset containing 10,000 reported trip events. Suppose the algorithm correctly classifies 9,650 and misclassifies 350, producing 96.5% overall accuracy. This percentage alone does not tell a motor insurer how many actual risky events the model detects, because sensitivity and specificity require the positive and negative class counts. If 100 of the 10,000 events are true positive instances and all 100 are missed, the classifier can still be correct on 9,650 other events, subject to its remaining false-positive assignments. A high headline accuracy can coexist with poor performance on a rare category.
More generally, let the prevalence of an actual event be p, sensitivity be Se, and specificity be Sp. The fraction of observations labeled positive is p·Se + (1 − p)·(1 − Sp). The fraction of positive labels that correspond to actual events, when the denominator is nonzero, is p·Se / [p·Se + (1 − p)·(1 − Sp)]. For an illustrative p = 0.02, Se = 0.90, and Sp = 0.95, only 0.018 / (0.018 + 0.049) ≈ 26.87% of positive labels are true positives. These quantities are hypothetical sensitivity inputs, not validated performance claims about any device or insurer. They show why the base rate and false-positive rate matter when a high-stakes adverse inference is made from a behavioral event.
Event misclassification can be differential by phone mounting, road geometry, vehicle class, and sensor hardware. A model that applies identical thresholds to unequal physical conditions can generate systematic score differences even before the rating algorithm is considered. Research should record confusion matrices across important subgroups, measurement drift, human review procedures, and the error distribution for consequential decisions.
Privacy, predictive power, and a concrete alternative
Granular position histories may help distinguish urban from rural mileage, identify exposures near particular road networks, or corroborate a crash. But the additional predictive value of exact coordinates must be tested against less intrusive alternatives, including verified distance, road class, time-of-day aggregates, local event counters, and declared use. A controlled comparison can hold train/test divisions and coverage definitions fixed, then evaluate predictive loss, calibration, stability, bias, and privacy exposure under progressively coarser feature sets.
An insurer should resist declaring that privacy necessarily eliminates useful risk measurement. Conversely, a marginal increase in an internal predictive metric does not automatically justify retaining every precise trip. The economic evaluation needs the incremental underwriting value, consumer welfare effects, legal permissions, operating costs, cyber consequences, and possible selection bias. These are separate empirical dimensions; reducing them to one profit metric obscures the governance question.
Competing interpretations of conduct and fairness
Motor insurance behavior scoring often uses measures such as rapid acceleration, hard braking, nighttime driving, mileage, and harsh cornering. These variables may correlate with expected loss at portfolio level. Their interpretation depends on route conditions, traffic, schedule constraints, emergency maneuvers, and geographic exposure. Nighttime travel may reflect shift work rather than a voluntary taste for risk. Hard braking may reflect protective response to another road user's unsafe act. A fairness inquiry should examine both actual predictive contribution and whether the rating policy places disproportionate burdens on groups for reasons that are unlawful or unjustified under the applicable regime.
Fairness has multiple measurable meanings: errors in driver attribution, within-group calibration, distribution of discounts and surcharges, access to opt-out alternatives, effects on protected or otherwise vulnerable groups, complaint outcomes, and recourse. Measures can conflict, particularly when underlying event frequencies or opportunity to be measured differ. A transparent empirical report states which metric is evaluated and how the insurer's legal environment constrains permissible data and rating variables. The NAIC Model Bulletin on AI systems is a supervisory model directed at insurer governance of AI-supported decisions; its status as a model bulletin must be distinguished from enacted state law in any specific jurisdiction.
Case-study limits and negative evidence
The GM/OnStar enforcement record establishes the terms of a real federal order and California action, the relevant enforcement allegations, and specific regulatory concerns. It does not support a quantified estimate of every affected driver's premium increase, a finding that all automobile manufacturers use comparable data-broker arrangements, or a universal conclusion that all insurers rely on these feeds. The FTC case materials and California's release should be interpreted according to their actual procedural status and stated facts.
Similarly, regulatory guidance is evidence of an authority's interpretation, not a randomized trial of privacy interventions. Peer-reviewed telematics studies may investigate prediction or driver behavior without auditing source legality. A statute can establish a right but cannot show that every vehicle manufacturer provides a technically usable interface. A consumer-reporting directory can establish a reporting channel but not the extent to which all insurers purchase or use it. These differences delimit what this documentary study can establish.
A reproducible research design
For future empirical work, an appropriate unit might be a vehicle-policy-year linked to trip-level exposure and legally authorized risk features, with explicit controls for jurisdiction, coverage, claims development, device, firmware, and enrollment. Researchers should preregister a question about prediction or privacy outcomes, define the sampling frame, identify legal permissions for obtaining personal data, and specify the consequences of opt-out and missing data. The minimal test set should separate training periods from later policy periods to detect drift and leakage; the audit trail should distinguish a score's generation from a premium decision. A second research design could compare consumer comprehension and actual withdrawal success across alternative consent interfaces, while recording which privacy duties apply.
Such studies require live data access and ethical, contractual, and regulatory approvals appropriate to the population. The present dossier supplies legal and methodological foundations but reports no original estimates of premium elasticities, crash reductions, fairness disparities, or industry-wide privacy compliance rates. Its examples are deliberately constructed and exactly reproducible from the printed assumptions.
Comparative Scientific Synthesis and Research Conclusions
A rights-and-purpose comparison across institutions
The European Union's Data Act addresses the accessibility and sharing of qualifying connected-product data; the GDPR protects personal data through independent processing principles and individual entitlements; and ePrivacy may separately regulate certain device-access acts. These instruments are designed for different legal questions and cannot be treated as interchangeable permission slips. The EU vehicle-data implementation guidance and EDPB sectoral analysis are complementary evidence for this distinction.
In the United States, sectoral insurance supervision, federal consumer reporting law, unfair-or-deceptive-practices enforcement, and state privacy laws establish a different pattern. The FTC's 2026 GM order is company-specific and enforceable through its terms; it is not a comprehensive federal vehicle data ownership statute. California's Honda decision and GM settlement announcement illustrate distinct enforcement contexts involving data-subject choice, proportionality, and onward use. The CFPB's reporting company directory identifies a further consumer-reporting layer relevant when the statutory conditions apply.
Brazil's LGPD recognizes personal-data processing rights under its own law; it is not an automotive product-data access analogue to the EU Data Act. Japan's APPI likewise requires country-specific legal interpretation. In the United Kingdom, connected-transport supervisory materials must be read alongside binding domestic law. The comparison supports a general analytical result: exact access rights, privacy rights, and insurance permissions arise from different instruments with different trigger conditions. No uniform cross-border concept of exclusive ownership can replace that statutory mapping.
The source-to-decision chain in motor insurance
The most defensible way to review a contested motor insurance decision is to follow the data backward. Begin with the actual rating or claims outcome and identify the variable relied upon, then the score or transformation, the observed events, the originating devices, the transfer path, the asserted legal basis, and the relevant policy or regulatory rule. This is a method of evidence reconstruction assembled from the scientific and legal sources, not an independent normative doctrine. It permits a reviewer to distinguish a correct observation used for an impermissible purpose from a lawful purpose compromised by a defective measurement.
A corresponding forward test begins with collection and asks whether each downstream use remains within the stated purpose, whether changes in recipients or methods alter the lawful basis, and how the insured could correct or challenge a damaging inference. These directions are complementary. The first reconstructs a historical decision; the second controls foreseeable secondary use.
Institutional implications for insurers and data suppliers
For insurers, the most consequential governance questions are practical and answerable: Which fields are genuinely necessary for each named insurance purpose? Is the lawful basis documented? Does the data subject know the recipient and retention period? Can the vehicle user get a relevant extract without supplying an unnecessary lifetime trajectory? Can the rating model identify driver attribution errors and demonstrate calibration across covered exposures? Can the insurer trace a disputed decision to the exact supplier record and version? Are data-broker or consumer-reporting obligations triggered? Can downstream corrections be propagated and evidenced?
For manufacturers and data holders, accurate field documentation, predictable access procedures, purpose-separated sharing, security controls, and software change logs improve both competition and evidentiary reliability. For public authorities, vehicle cybersecurity supervision, insurance rating oversight, data-protection enforcement, and consumer reporting remedies govern overlapping but distinct failures. These implications follow from the mapped institutional functions; a single regulator or contract cannot plausibly stand in for all of them.
Research conclusions and testable questions
The documentary evidence supports five conclusions. First, connected-vehicle data cannot be governed adequately by attributing every record to a single owner: user access, personal-data rights, service control, and insurance permissions are different relations. Second, precise data provenance determines what a motor insurer can legitimately infer from a signal; ownership rhetoric offers no substitute for sensor validity or driver attribution. Third, consent and other legal bases must be evaluated for the specific stage and purpose, with continuing attention to withdrawal, deletion, retention, and shared vehicles. Fourth, secondary transfers can produce independent harms and obligations even when a particular insurance rating use is unavailable in a given jurisdiction. Fifth, selecting a participation-based telematics dataset can distort claims estimates and fairness comparisons, while high event-classification accuracy may conceal poor precision for rare events.
These conclusions generate directly testable questions. Does driver-level attribution error increase for multi-occupant vehicles and particular hardware configurations? Does a coarse, on-device mileage extract retain most of the predictive value of full location traces? How do contractual enrollment designs influence comprehension and withdrawal rates? Which consumer-reporting correction pathways actually propagate revised scores to insurer decisions? Do recorded sensor and firmware changes alter the distribution of apparent behavioral risk? What is the out-of-sample insurance effect of telematics among participants and nonparticipants after accounting for selection? Each question identifies a potential empirical inquiry, not a claimed discovery already established by this manuscript.
Conclusion
Connected-vehicle telemetry brings risk exposure, driver behavior, automobile engineering, and motor insurance administration into a shared informational environment. The physical vehicle produces observations; platforms transform and distribute them; insurers may incorporate selected information into decisions; and several legal systems allocate different rights and duties to the actors along the way. Evidence from European legislation, privacy guidance, U.S. enforcement, consumer reporting, and actuarial scholarship demonstrates why this chain requires continuous source attribution and purpose-specific interpretation.
The comparative result is precise. A vehicle user's right to obtain certain product data is distinct from a natural person's rights concerning identifiable information. A company's possession of telemetry differs from its authority to reuse that information for insurance. A statistically useful behavior indicator is an inference conditional on measurement and population, while a lawful premium or claim decision remains subject to insurance rules, privacy obligations, and contestable evidence. Rigorous motor insurance analysis must preserve all these distinctions as it moves from a sensor reading to a consequential outcome.
The dossier's original contribution is an integrated, explicitly bounded analysis of the links among data provenance, legal access, processing purpose, statistical validity, secondary disclosure, and decision accountability. It presents no new theoretical system, claims no original observational dataset, and makes no unsupported claim of universal legal harmonization. Its findings identify the information and legal conditions under which future empirical comparisons can be meaningful and reproducible.
References
Boylan, J., Meyer, D., & Chen, W. S. (2024). A systematic review of the use of in-vehicle telematics in monitoring driving behaviours. Accident Analysis & Prevention, 199, 107519. DOI: 10.1016/j.aap.2024.107519. https://pubmed.ncbi.nlm.nih.gov/38458008/
Brazilian Ministry of Development, Industry, Commerce and Services. (2024). Legislação sobre proteção de dados pessoais (including Lei nº 13.709/2018, LGPD). https://www.gov.br/mdic/pt-br/canais_atendimento/ouvidoria/legislacao/lgpd
California Department of Justice, Office of the Attorney General. (2026, May 8). When it comes to data privacy, consumers must be in the driver's seat: $12.75 million General Motors privacy settlement. https://oag.ca.gov/node/622991
California Department of Justice, Office of the Attorney General. (2026 access). Privacy enforcement actions: General Motors. https://oag.ca.gov/privacy/privacy-enforcement-actions
California Privacy Protection Agency. (2025, March 12). Honda settles with CPPA over privacy violations. https://cppa.ca.gov/announcements/2025/20250312.html
Consumer Financial Protection Bureau. (2025 access). LexisNexis C.L.U.E. & Telematics OnDemand. https://www.consumerfinance.gov/consumer-tools/credit-reports-and-scores/consumer-reporting-companies/companies-list/comprehensive-loss-underwriting-exchange/
European Commission. (2025, September 15). Guidance on vehicle data, accompanying Regulation (EU) 2023/2854 (Data Act), C/2025/5026. Official Journal of the European Union. https://eur-lex.europa.eu/legal-content/EN/ALL/?uri=CELEX:52025XC05026
European Data Protection Board. (2021, March 9). Guidelines 01/2020 on processing personal data in the context of connected vehicles and mobility related applications (version 2.0). https://www.edpb.europa.eu/documents/guideline/guidelines-012020-on-processing-personal-data-in-the-context-of-connected_en
European Parliament and Council. (2002). Directive 2002/58/EC concerning the processing of personal data and the protection of privacy in the electronic communications sector. https://eur-lex.europa.eu/eli/dir/2002/58/oj
European Parliament and Council. (2016). Regulation (EU) 2016/679, General Data Protection Regulation (GDPR). https://eur-lex.europa.eu/eli/reg/2016/679/
European Parliament and Council. (2023). Regulation (EU) 2023/2854 on harmonised rules on fair access to and use of data (Data Act). https://eur-lex.europa.eu/eli/reg/2023/2854/oj/eng/
European Union Publications Office. (2025 access). Rules on fair access to and use of data (Data Act): Legislative summary. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=LEGISSUM:4723374
Federal Trade Commission. (2026, January 14). FTC finalizes order settling allegations that GM and OnStar collected and sold geolocation data without consumers' informed consent. https://www.ftc.gov/news-events/news/press-releases/2026/01/ftc-finalizes-order-settling-allegations-gm-onstar-collected-sold-geolocation-data-without-consumers
Federal Trade Commission. (2026). General Motors LLC, et al., In the Matter of, Docket C-4828; January 14 decision and order. https://www.ftc.gov/legal-library/browse/cases-proceedings/2423052-general-motors-llc-et-al-matter
Federal Trade Commission. (2026 access). Fair Credit Reporting Act: Statutory resource. https://www.ftc.gov/legal-library/browse/statutes/fair-credit-reporting-act
Information Commissioner's Office (United Kingdom). (2025). Tech horizons report 2025: Connected transport and privacy technology analysis. https://ico.org.uk/about-the-ico/research-reports-impact-and-evaluation/research-and-reports/technology-and-innovation/tech-horizons-and-ico-tech-futures/tech-horizons-report-2025/
Ma, Y.-L., Zhu, X., Hu, X., & Chiu, Y.-C. (2018). The use of context-sensitive insurance telematics data in auto insurance rate making. Transportation Research Part A: Policy and Practice, 113, 243–258. DOI: 10.1016/j.tra.2018.04.013. https://scholarsmine.mst.edu/civarc_enveng_facwork/1312/
National Association of Insurance Commissioners. (2023, December 4). Model bulletin: Use of artificial intelligence systems by insurers. https://content.naic.org/sites/default/files/inline-files/2023-12-4%20Model%20Bulletin_Adopted_0.pdf
National Association of Insurance Commissioners. (2026, September 1). Telematics: Insurance topics. https://content.naic.org/insurance-topics/telematics
Personal Information Protection Commission (Japan). (2026 access). Laws and policies: Personal information protection. https://www.ppc.go.jp/en/legal/
Gao, G., Meng, S., & Wüthrich, M. V. (2022). What can we learn from telematics car driving data: A survey. Insurance: Mathematics and Economics, 104, 185–199. DOI: 10.1016/j.insmatheco.2022.02.004. https://www.sciencedirect.com/science/article/pii/S0167668722000233
United Kingdom Vehicle Certification Agency. (2026, May 22). Cyber security and software updating: UN Regulations 155 and 156. https://www.vehicle-certification-agency.gov.uk/connected-and-automated-vehicles/cyber-security-and-software-updating/