RESEARCH DOSSIER
Smart Building IoT Vulnerabilities and Cyber Physical Claims
Abstract. Smart buildings concentrate physical processes, digital control, service dependencies, and contractual risk within the same economic asset. This monograph investigates how vulnerabilities in connected sensors, control networks, supervisory systems, remote maintenance access, and external platforms become threats to insured property, continuity of operations, third-party interests, and insurance portfolios. It integrates operational technology security, building physics, incident evidence, comparative insurance contract analysis, frequency-severity mathematics, and accumulation assessment. Primary materials include official device and operational technology guidance, manufacturer vulnerability notices, model property endorsements, judicial opinions, and supervisory texts. Peer-reviewed building automation research is examined within the available abstract and repository evidence. Documented incidents and testbed results are distinguished from hypothetical physical scenarios, market-level insurance statistics, covered losses, and insurer payments. The analysis establishes a causal accounting sequence linking a digital event to a changed operating condition, exposure to physical hazards, failed or successful protective barriers, economic loss, and contractual response. It demonstrates why an outage can be economically severe without tangible damage, why physical damage does not by itself establish property coverage, and why a shared digital dependency may concentrate losses across geographically separated risks. Reproducible illustrative calculations identify the sensitivity of expected indemnity and portfolio variance to damage conversion, coverage terms, and correlated events. The resulting scientific synthesis places operational evidence, actual policy wording, and common-dependency measurement at the center of cyber-physical underwriting and claims analysis. No population estimate of building-specific insured loss frequency or claim of a new named theory is made.
Keywords: Smart buildings; Internet of Things; operational technology; building automation systems; cyber-physical risk; property damage; business interruption; cyber insurance; engineering loss prevention; sensor integrity; control systems; coverage exclusions; non-affirmative cyber exposure; forensic reconstruction; frequency-severity analysis; accumulation; insurance claims.
Introduction
The scientific subject
A building is an engineered system in which thermal conditions, air quality, electricity, water, access, fire protection, and mechanical assets are managed to support a specific use. A smart building adds digitally mediated observation, decision, and control. Its insured exposure therefore depends on the interactions among the structure, replaceable equipment, firmware, software, communication networks, operating procedures, external services, users, and the legal allocation of interests. A single malfunction can cross several of these domains in succession.
The insured property may belong to an owner, equipment to a lessee, goods to an occupant, and a management platform to a contractor. Responsibility for repair may sit with a facilities manager whose own insurance protects only certain liabilities. Business interruption losses may be sustained by a tenant while the owner's immediate financial interest is rent. The practical scientific problem is to relate an event to injured interests without assuming that a building and its occupants constitute one insured entity.
Cyber insurance supplies the risk-transfer perspective; risk engineering and loss prevention supply the study of physical causation, safeguards, and recovery. This dossier specializes in their intersection. It covers commercial, residential, institutional, and mixed-use buildings as exposure settings, with particular attention to connected building controls. It does not treat building control faults unrelated to digital compromise as proven cyber events. Equally, it includes accidental digital failure when the technical facts and policy wording make it relevant to cyber-related exposure.
Three questions organize the investigation. First, what device, network, and organizational conditions make digital disturbance capable of changing building operation? Second, through what physical and contractual mechanisms does that disturbance produce an insured loss? Third, what observations and denominators allow the severity, frequency, and dependence of such claims to be assessed with defensible uncertainty?
Research method and evidentiary boundaries
The method is integrative documentary research. Technical standards and official guidance establish concepts and prevention objectives; manufacturer notices establish specified product issues; scientific publications establish the bounds of reported experiments or reviews; actual legal and model policy texts establish particular contractual mechanisms; supervisory documents establish regulatory concerns. The causal and mathematical analyses are original applications of established engineering, probability, and insurance reasoning to the defined subject.
The source corpus is purposive rather than a counted systematic review. Searches targeted operational technology and building automation, IoT capability baselines, actual building-linked event narratives, model cyber exclusions, physical loss jurisprudence, regulatory exposures, and aggregate claims-market measurement. Source identity, dating, accessible passages, and limitations are recorded in the companion evidence ledger. Repository abstracts are used only for their expressed findings. Model clauses are examples of wording, not evidence of uptake or an individual insured contract. Cases are jurisdiction-specific and selected to expose differences in contractual interpretation. There is no collected proprietary building claims dataset, verified portfolio of actual bound endorsements, or live penetration test. The numerical cases throughout are constructed illustrations, not actuarial estimates.
The main analytical distinction is between hazard, event, material condition, economic consequence, and entitlement to indemnity. A credible connection across these categories requires evidence at each material step. This distinction also permits the investigation to integrate malicious activity with accidental systems failure without conflating their legal characterization.
Conceptual and Historical Foundations
From building management to connected operational technology
Building automation historically developed around local measurement and control: temperature sensing, zone regulation, plant scheduling, alarms, and electrical or mechanical interlocks. The transition toward networked management and remotely operated services extends the control architecture beyond the physical premises. The economic rationale is straightforward. Central management allows one operator to observe several sites; automated scheduling reduces repetitive interventions; interconnected systems permit building operators to coordinate energy consumption and occupancy.
Li et al. (2023) describe this development in their building automation security review, which organizes technical exposure at management, automation, and field levels and discusses vulnerabilities across widely used protocols. Connectivity changes the distribution of authority: a person, application, or service outside a building can participate in determining its physical behavior. The same feature expands the perimeter that must be considered when estimating risk.
Stouffer et al. (2023) explicitly include building automation within the scope of operational technology security guidance. The safety, availability, and performance requirements of OT give cybersecurity an engineering dimension. In November 2026, the NIST publication record also identified Revision 4 as an initial public draft, with an open comment period through November 30. This study uses the finalized Revision 3 designation and does not attribute final requirements to its proposed successor.
An information system may be economically impaired because confidential records are disclosed. A control system can additionally be impaired because the signals used to represent a physical environment are wrong, late, or absent. Both are information failures, yet the cost architecture differs. A compromised identity database can produce privacy response expenses; a failed cold-room control can expose goods to temperature conditions that cause deterioration. The digital origin alone does not specify the loss category.
Functional architecture and control
For a stylized building, let x(t) denote the actual physical state at time t, y(t) the measurement delivered to a controller, u(t) the action decided by the controller, and d(t) external disturbances such as weather, occupancy, and utility availability. A simplified dynamic relationship is
x(t + 1) = f(x(t), u(t), d(t)).
The measurement relationship is y(t) = h(x(t)) + e(t), where e(t) is measurement error. Control actions may depend on measured state, business rules, schedules, alarms, and physical constraints. This ordinary state-space representation serves as a reasoning aid, not as a calibrated model of a particular building. The investment in engineering resilience concerns what happens when observations, commanded actions, or the infrastructure connecting them become unreliable.
A sensor may provide a wrong value because of calibration drift, disconnection, mechanical obstruction, or manipulation. A controller may faithfully implement an incorrect command. A valid schedule may become inappropriate after occupancy or weather changes. A correctly functioning valve cannot regulate water flow if the supply is interrupted. Accordingly, a digital anomaly must be understood in relation to both the physical plant and the environment.
At field level, sensors and actuators interact with equipment. At the automation level, controllers coordinate local processes. At the management level, servers, gateways, and interfaces aggregate information and orchestrate control. External support, identity, and cloud services may connect to one or more levels. This decomposition is useful for describing the scope of an incident, the authority of a compromised device, and the independence of protective systems. The location of the vulnerability in the architecture matters more to the eventual physical damage distribution than its mere existence.
| Layer or dependency | Typical function | Insurance-relevant exposure question |
|---|---|---|
| Field equipment | Measurement and physical actuation | Can failure permit unsafe physical conditions? |
| Local controller | Feedback, schedules and equipment coordination | Can local constraints survive supervisory loss? |
| Supervisory management | Monitoring, configuration, alarms | How many zones or buildings share authority? |
| Remote maintenance | Vendor diagnostics and service | Who owns credentials, logs and recovery duties? |
| External services | Identity, connectivity, hosted analytics | What fails when access or service is unavailable? |
| Independent protection | Mechanical or electrical safeguards, local alarming | Is the barrier independent of the initiating dependency? |
Risk classification and economic setting
The same building technology can carry different exposures depending on occupancy. A temporary temperature excursion in an empty office may be inexpensive. A similar event in a refrigerated warehouse may threaten inventory; in a hospital it may disrupt an essential service or require relocation. A residential building with dependent occupants may have high emergency assistance needs even if the cost to restore network service is modest. Exposure should be expressed in terms of affected functions, duration, occupancy, repairability, and the capacity to continue operating safely.
A vulnerability is a condition that can contribute to a harmful digital event. A threat is a possible adverse circumstance or action. An incident is an observed event within a defined scope. A physical hazard is a condition capable of causing injury or damage. The resulting loss measures should be individually specified: affected building time, damaged quantities, repair cost, business contribution lost, additional mitigation expenditure, and liabilities. A claim is a request for benefits or indemnity under a particular contract. Reported claim, accepted claim, outstanding case reserve, ultimate loss, and paid amount are distinct measures.
These definitions prevent category drift. The number of connected devices is not the number of policies, and neither is the number of claims. A vulnerability announcement is not an incident tally. A verified physical incident is not an insurance payment record. Claims science begins where technical records, physical evidence, contract language, and financial measurement can be joined.
IoT Vulnerabilities and Building System Exposure
Device capability and the installed system
Fagan et al. (2020) establish an acquisition-oriented IoT device capability baseline that describes technical capabilities useful to common cybersecurity controls. The NIST technical capability catalog elaborates those device-level abilities. These materials give acquisition and assurance teams a way to examine what a component can support. The existence of a capability does not establish that an installer enabled it, that a service organization maintains it, or that neighboring equipment can safely interact with it.
A building inventory must therefore contain more than a device count. Useful fields include role, operational criticality, installed version, responsible organization, support dates, authorized communication pathways, change history, and availability of a tested configuration backup. A network device with little physical authority may be pivotal if it is the sole operational connection between local controllers and the alarm service. Another device with a high technical severity score may have limited consequence in a particular installation if it is isolated and does not control critical functions.
The distinction between vulnerability severity and insured loss severity is central. A technical rating assesses a defined technical problem under assumed conditions. The economic loss distribution depends on configuration, access, propagation, physical tolerances, recovery resources, and contractual response. Multiplying device vulnerability scores by insured values is not a defensible loss model. It omits the conversion mechanisms that determine both probability and magnitude.
Networked trust and common dependencies
Li et al. (2023) examine communication weaknesses in BACnet, KNX, LonWorks, and Modbus and categorize signal corruption, delay and blocking as relevant consequences. Their review provides a technical taxonomy, not an estimate of the frequency of insured building losses. The architecture gives rise to multiple possible exposure pathways: a loss of visibility at management level, an inability to transmit commands, or incorrect control information at a local controller. The consequences vary with design and safeguards.
Remote management changes the location of exposure. The authority to coordinate several facilities may be concentrated in a provider or administrator. A shared platform may also distribute configuration errors simultaneously. Underwriting therefore needs to identify shared software, management services, maintenance vendors, and identity arrangements as portfolio dependencies. The eligibility of a supplier-dependent loss under insurance must be evaluated separately.
A manufacturer advisory provides a narrow example of what can be known. The Johnson Controls Metasys security advisory catalog lists an information disclosure issue affecting older building automation versions and recommends an upgrade. The catalog documents the product issue and identified affected versions. It does not establish that a specific building used such a version, that a malicious actor accessed the system, or that a controller changed a physical process. For underwriting, a meaningful response verifies installed equipment and compensatory controls. For claims investigation, a matching product issue is a lead that must be joined to event-specific evidence.
Organizational access and the supply chain
The 2014 U.S. Senate hearing record includes a majority staff analysis of the 2013 Target breach. Its third-party vendor discussion identifies a facilities-related HVAC contractor as relevant to the incident and draws on published reporting and expert analyses. It is an instructive example of vendor trust crossing organizational boundaries. It does not demonstrate that the attackers manipulated heating equipment, that temperature-related damage occurred, or that a particular insurance payment was made.
In scientific terms, a vendor relationship and a control connection are different observations. A person may possess credentials for a procurement portal without having access to operational controllers. A maintenance account may be authorized for one building but be mismanaged across a fleet. The evidence should identify the exact path of authority rather than infer it from the contractor's industry classification.
A complete procurement specification therefore connects security requirements to service responsibilities. Who approves remote sessions? Who retains change records? Who is responsible for vulnerability disclosure and patch testing? Who can restore configuration if the main platform fails? Who bears the cost of equipment that no longer receives vendor support? These questions matter to the owner, the provider, and their insurers because the ability to mitigate and reconstruct an incident is distributed among them.
Lifecycle, remediation, and legal context
Connected devices can remain in service while software versions, certificates, interfaces, and support arrangements change. A one-time audit gives only a dated observation. Updating a control system may itself impose operational risk if it changes compatibility or requires downtime. The engineering objective is a controlled transition supported by authorized testing, preserved configurations, and a safe continuity arrangement.
The European Commission CRA summary identifies Regulation (EU) 2024/2847 as a product-oriented cybersecurity regime for hardware and software with digital elements made available on the Union market. It reports different applicability dates: entry into force December 10, 2024, certain conformity assessment provisions from June 11, 2026, reporting obligations under Article 14 from September 11, 2026, and general application December 11, 2027. As of this research cutoff, these stages must not be collapsed into a claim that all requirements are already operative. The summary is an official explanatory publication, not a replacement for reading the regulation and its exceptions before an individual legal decision.
Product regulation can improve the availability of cybersecurity information and influence manufacturer behavior. It does not by itself determine the insurance coverage of a building owner, assign all liability for a failed installation to a supplier, or prove that a violation caused particular damage. The insurer's contract, applicable law, and forensic causation remain separate questions.
Loss prevention evidence
Risk engineering should focus on demonstrated operation of safeguards rather than the mere presence of security products. A useful control record identifies the asset and function, control objective, operating owner, test date, test conditions, result, deviations, and corrective action. Tests proposed here are authorized assurance activities, not instructions for accessing systems. They can establish whether a building retains safe local operation during planned loss of a nonessential supervisory service, whether recovery copies are usable, whether alerts reach an accountable operator, and whether vendor access is governed.
The required depth is proportional to consequences. Equipment protecting life or critical services demands safety-aware change management and competent engineering review. An isolated test environment and a planned rollback protect occupants and equipment during assessment. A cybersecurity audit should not induce the physical loss it is intended to prevent.
Physical Damage, Operational Interruption, and Causation
The conversion from digital event to physical loss
Cyber-physical causation is a sequence of changes in operating conditions. A digital event can change a command, interrupt observation, compromise a schedule, or reduce service availability. Physical loss may follow if the affected process exceeds its tolerable limits and safeguards fail to interrupt the progression. Exposure to unsafe environmental conditions, duration of the excursion, physical susceptibility, and timing of mitigation determine damage.
For example, in an illustrative cold-storage facility a control failure can interrupt refrigeration. The immediate effect is an operating state, not the value of destroyed stock. The temperature trajectory depends on insulation, loading, initial conditions, outside temperature, and whether cooling is partly maintained. Product deterioration depends on product-specific limits, actual duration, and condition assessment. Some goods may remain usable; others may require disposal according to safety and quality criteria. A claim for inventory loss requires proof of actual disposition and value, not simply the presence of an alarm.
A loss of building heating can be treated similarly. In cold conditions, sufficiently prolonged loss of heat may make water-bearing equipment susceptible to freezing. A very different result occurs if local heating remains available or the outage is brief. The U.S. Department of Energy energy systems cybersecurity guidance reports a Finnish building heating disruption associated with denial of service and repeated system restarts. This is a public official illustrative account of interrupted operation, not a verified record of frozen pipes, bodily injury, or paid insurance claims. Its relevance is the connection between digital service failure and building function.
A physical pathway should be reconstructed at the granularity required by the incident. If a pump fails mechanically, a network alarm may occur after rather than before the damage. If a control failure precedes a leak, its timing alone does not establish causation. If an unauthorized change occurred while the equipment was already deteriorating, both may contribute to loss. The examiner has to compare rival explanations with actual event and physical evidence.
Protective barriers and failure independence
A protective barrier interrupts a harmful sequence. It may prevent unauthorized changes, constrain the physical operating range, detect deterioration, isolate equipment, or provide alternate service. Some barriers are implemented through digital control; others use separate mechanical or electrical protection or operator intervention. The effectiveness of a barrier is conditional on the conditions under which it is needed.
Two alarms using the same failed sensor are not independent observations. Two supervisory platforms relying on one identity service may fail together. A local protection system whose authority can be removed through the same centralized maintenance process deserves separate analysis. Conversely, an independent physical safeguard may prevent equipment damage even when central visibility is unavailable. A diagram showing two protective devices is insufficient evidence of independence; their power, sensing, control, maintenance, and common failure modes matter.
A conventional event tree is useful for organizing these pathways. Let A be an incident initiating an unsafe process, B be failure of the relevant protective barrier, and D be material damage. In a simplified sequence where damage requires A and B,
P(D) = P(A) × P(B | A) × P(D | A, B).
Other mechanisms may exist, and multiple paths require consistent event definitions and overlap control. This identity is not an assumption of independence. The conditional terms matter precisely because digital failure can affect several barriers. A barrier failure observed in one incident is also not an estimate of its annual probability.
Incident reconstruction and alternative explanations
Li et al. (2024) report hardware-in-the-loop detection experiments with one baseline, two physical fault conditions, and two cyberattack conditions. Their study combines real control and network hardware with modeled building and HVAC behavior and reports average classification accuracy of 90.2% in the described evaluation. The accessible research record supports the existence of those experiments and their reported result. It does not establish the performance of the method in every building, with every physical fault, or in a claims adjustment population.
The significance for claims science is the similarity of observed manifestations. A control outage, unexpectedly warm space, high energy use, or equipment stoppage could result from a mechanical fault, a network disturbance, a software error, or intentional interference. Investigation must combine the physical process history with the digital history. A detector output is evidence of a classification, not a substitute for proving causation, coverage, or intent.
A loss investigation benefits from synchronizing controller logs, facility schedules, network and identity records, operator statements, independent measurements, service calls, parts inspections, utility records, and weather data. The record should preserve time zone, clock corrections, logging gaps, custody, and restoration actions. Where urgent action is needed to protect people or property, preservation must accommodate safety; the investigator should document which evidence could not be retained and why.
A counterfactual comparison asks what physical and financial outcome would probably have occurred had the disputed digital event been absent, while retaining the relevant weather, occupancy, equipment condition, and other external circumstances. A pre-event failure history, adjacent unaffected zone, appropriate engineering analysis, and independently recorded measurements may constrain this comparison. There is no requirement that an alternative explanation be impossible before any finding is made; the relevant evidentiary standard is determined by the purpose of the investigation and applicable law. Confidence in the engineering conclusion should be communicated distinctly from the applicable legal standard.
Physical and financial loss accounts
The loss account separates seven components: tangible asset repair or replacement; spoiled or contaminated stock; digital restoration; incremental mitigation and emergency expenses; business income loss; third-party liability; and separately measured deductions or recoveries. Additional categories may be needed for the actual insured interest. Costs such as redesign or security upgrades may combine remediation, future prevention, and betterment. Their allocation should be transparent instead of labeling the entire expenditure as restoration.
For business interruption, the calculation requires a baseline economic activity that would have occurred absent the covered event, a supported actual recovery path, and the policy basis of indemnity. Gross revenue alone is often inappropriate because some expenses are avoided during closure. Lost contribution, continuing expenses, mitigation costs, extra expense, waiting periods, and restoration periods must be distinguished. A landlord's rent interruption, a tenant's lost production, and an operator's service penalties are different exposures.
Consider an illustrative tenant whose net contribution (after avoided variable costs) is USD 25,000 per full day. Four days of complete cessation produces USD 100,000 gross contribution shortfall, assuming no delayed recovery of sales. If a hypothetical time deductible excludes the first 24 hours, the remaining three days produce USD 75,000 before monetary limits and other terms. A separately eligible USD 10,000 emergency expense would change the total to USD 85,000 before limits, if the expense is not otherwise counted. If one day of sales is recovered at no incremental expense, the factual baseline changes. This example uses an expressly assumed contract, not an interpretation of any cited form.
Life safety and liability
A building may be temporarily unfit for its intended use even though the structure and mechanical equipment remain physically intact. Evacuation, temporary relocation, manual security, and emergency support can be costly. Bodily injury, property damage to tenants, and breach of service obligations can create separate claims against owners, operators, and contractors. These legal liabilities depend on such matters as duty, breach, causation, contract terms, defenses, and jurisdiction.
Life-safety functions require particular caution in risk engineering. Decisions about testing, restoration, and manual operation belong with competent personnel responsible for the safety of occupants and physical systems. From an insurance perspective, a realistic scenario should account for emergency measures and the resulting claims pathways without assuming that any hypothetical hazardous sequence actually occurred.
Coverage Interfaces, Claims, and Portfolio Dependence
Contract structure and affirmative cyber exposure
An economic loss becomes an insurer liability only insofar as a valid contract responds under applicable law. The analysis begins with the identified insured, relevant asset or interest, effective wording, insured event definition, applicable exclusions, causal requirements, conditions, and financial limitations. The policy schedule and endorsements may substantially modify the base form. The effect of a deductible or exclusion cannot be evaluated without knowing which coverage section and loss component are involved.
Standalone cyber insurance may respond to defined security failures, system failures, response expenditure, interruption, and specified liabilities. Property, equipment breakdown, crime, general liability, technology professional liability, and specialized endorsements may also be relevant. The labels of these product families are descriptive. They do not establish the terms, the presence of a cyber exclusion, or a particular payment.
The LMA model clause announcement identifies LMA5400 and LMA5401 as illustrative forms and states that parties may adopt modified terms. This is important evidence of contractual diversity, not proof that either form governs all property insurance.
Model property endorsements and resulting damage
LMA5400, the Property Cyber and Data Endorsement, first excludes specified cyber and data-related loss, then creates limited exceptions. Paragraph 2 addresses insured property loss from an ensuing fire or explosion directly resulting from a defined Cyber Incident, with an additional restriction where a Cyber Act is involved. Paragraph 3 addresses insured physical damage to data-processing media and certain copying costs, rather than insuring the economic value of data itself.
The distinction between Cyber Act and Cyber Incident in this wording is material. A malicious operation and a nonmalicious digital failure can be treated differently even when both produce similar equipment damage. Fire or explosion is also a specific condition of the paragraph-2 exception. A heating failure leading to freeze damage, for example, does not become an ensuing-fire event merely because the initiating disturbance is digital.
LMA5401, the Property Cyber and Data Exclusion, contains broad exclusions for defined Cyber Loss and specified data-related consequences and expressly addresses concurrent or sequential contributing causes. It has no equivalent ensuing-fire provision in the reproduced form. Under both wordings, the defined digital system can include third-party systems. These observations concern the reproduced text; legal enforceability, amendments, interactions with other forms, and the applicable burden of proof must be assessed for an individual contract.
| Hypothetical consequence | Basic coverage issue |
|---|---|
| Control interruption without damaged property | Defined cyber interruption trigger, waiting period and dependent service coverage |
| Accidental digital failure followed by fire | Whether property wording contains an effective resulting-damage exception and required causal relationship |
| Malicious interference followed by fire | Whether a Cyber Act restriction, other exclusion, or affirmative cyber-physical endorsement controls |
| Water escape after prolonged control failure | Whether the property peril survives cyber exclusions; no automatic extension from a fire-specific exception |
| Hardware intact but software unusable | Electronic property definitions, restoration coverage and jurisdiction-specific physical loss interpretation |
| Injured tenant or damaged tenant goods | Liability of relevant parties and distinct liability/first-party coverage |
The table is a research comparison of questions; it is not a coverage decision. An affirmative physical cyber endorsement could deliberately buy back coverage otherwise excluded, subject to its own terms. A policy with silent wording creates a different interpretive and pricing problem. A well-designed program records both positive grants and exclusions across all layers; the two need to be compared using actual scenarios.
Physical loss jurisprudence
Two U.S. opinions illustrate the importance of contract language and jurisdiction. In National Ink & Stitch (2020), the U.S. District Court for the District of Maryland considered a businessowners computer coverage endorsement expressly encompassing electronic records and software. Following ransomware and impaired computer performance, the court granted the insured's summary judgment motion and recognized the loss at issue as within coverage under the particular wording and Maryland law. The case concerns computer equipment and data, not a judicial finding of building equipment destruction.
In EMOI Services (2022), the Supreme Court of Ohio interpreted a different electronic-equipment endorsement. It held that physical loss or damage to the covered media was required; software encryption without physical damage to the relevant hardware did not establish coverage. It reversed the appellate judgment and reinstated summary judgment for the insurer. The two decisions involve different policy texts and legal jurisdictions. They do not determine how courts elsewhere would classify building control impairment or actual fire, flood, or mechanical damage.
An adjuster should therefore distinguish factual physical alteration, inability to access software, diminished performance, inability to use a location, and the legal trigger governing each coverage. The investigator's job is to establish the material condition and cause with appropriate confidence. The coverage analyst applies the particular contract and law to those facts. The court decisions support this separation rather than a generalized rule that all software impairment either is or is not physical damage in insurance law.
Overlap, gaps, and recovery
A hypothetical loss may produce a USD 300,000 equipment repair cost, USD 80,000 digital restoration cost, and USD 180,000 business income shortfall. The economic total of these mutually exclusive components is USD 560,000. Suppose solely for illustration that the property section accepts the equipment repair, has a USD 50,000 deductible and a USD 500,000 remaining limit. Its payment is USD 250,000. Suppose a cyber section separately accepts restoration and the income shortfall, but has a combined USD 25,000 deductible and USD 200,000 remaining limit. Its payment is the lesser of USD 235,000 and USD 200,000, namely USD 200,000. Total payments are USD 450,000 and unrecovered loss USD 110,000, ignoring recoveries, expenses outside indemnity, tax, waiting periods, and other clauses.
If the property section instead excludes the equipment loss and no other coverage responds, the hypothetical insurance recovery is only USD 200,000. The difference is entirely contractual in this example; the physical event and overall economic loss have not changed. Alternatively, if both sections respond to some of the same dollars, other-insurance, contribution, allocation, and applicable indemnity rules are relevant. The two gross coverage amounts cannot simply be added without checking overlap.
The claims record should link each item to a loss component, claimant, payee, funding source, applicable coverage, claim status, amount, and recovery. Subrogation against a vendor may depend on contract limitations, liability evidence, and applicable rights. A later recovery must be recorded in relation to the relevant gross and net measures. This is particularly important when a single occurrence yields claims by several tenants, multiple insurers, or overlapping coverage sections.
Supervision and non-affirmative cyber
The Prudential Regulation Authority's cyber underwriting statement defines prudential cyber underwriting exposure to encompass malicious and accidental digital causes and tangible as well as intangible assets. Its UK supervisory scope must be distinguished from a worldwide legal definition of covered loss.
The New York Department of Financial Services (2021) Cyber Insurance Risk Framework identifies silent exposures in non-cyber products and calls for a formal risk strategy and clearer coverage. The European Insurance and Occupational Pensions Authority's (2022) non-affirmative cyber supervisory statement addresses clarity of terms, scenario assessment, concentration, accumulation, reinsurance, and solvency considerations. These materials signal a common prudential concern: an insurer should understand its exposure even where the policy is not marketed as cyber insurance. They do not prescribe an identical policy trigger across jurisdictions.
Lloyd's and Cambridge Centre for Risk Studies (2022) discuss physical cyber risk scenarios in relation to geopolitical uncertainty. Scenario analysis can clarify groups of connected exposures and coverage pathways. It should preserve its status as forward-looking analysis rather than be presented as an observed distribution of insured losses.
Portfolio accumulation and reinsurance
A property portfolio typically records location, construction, occupancy, asset values, and physical protection. Cyber-physical assessment adds the dependencies through which the same digital event could affect several properties. An otherwise diverse geographic portfolio may rely on one remote service, equipment family, platform, vendor, or common recovery contractor. Such concentrations can affect both event severity and recovery duration.
For insured payment X_i at site i and aggregate S across sites,
E(S) = sum of E(X_i),
Var(S) = sum of Var(X_i) + 2 times the sum of Cov(X_i, X_j) over distinct pairs i < j.
Dependence does not change the mean when all individual marginal distributions remain fixed; it can substantially change the variance and tail distribution. Consequently, geographically independent property perils and shared cyber dependencies need separate consideration. Physical natural hazards and cyber dependencies may also interact, for example when a utility interruption and reduced remote visibility coincide.
Take 1,000 identical insured buildings, a constant USD 40,000 payment for each affected site, and an illustrative annual common event probability of 1%. If that event always affects 100 specified or uniformly sampled buildings, the conditional portfolio loss is USD 4 million and annual expected payment is USD 40,000. For a fair independence comparison, suppose the 100 affected buildings are uniformly selected from the 1,000 when the common event occurs; each has marginal probability 0.001 per year. If each site instead independently has the same annual marginal probability 0.001, expected aggregate payments remain USD 40,000. The common-event construction gives variance of affected count 100 squared multiplied by 0.01 × 0.99 = 99; the independent construction gives 1,000 × 0.001 × 0.999 = 0.999. The portfolio payment variance ratio is about 99.1. These deliberately simple assumptions illustrate dependence, not the real probability or distribution of an attack.
The insurance and reinsurance responses may differ even with identical underlying event losses. Occurrence definitions, related event clauses, hours provisions where present, aggregate deductibles, exclusions, and sublimits determine the layer recovery. A common dependency does not automatically establish one occurrence. Conversely, geographically separated losses may be contractually related if the wording and facts support that result. A stress test should report gross losses, direct insurer covered loss, recoveries under specified treaty terms, and retained net loss separately.
Evidence Validation and Sensitivity Analysis
What the available evidence measures
A sound study distinguishes direct observations from documents that identify only possible mechanisms. The NIST publications establish guidance concepts and scope. The manufacturer notice establishes an identified product issue. The Senate record and DOE account supply public event narratives with distinct evidentiary boundaries. The building automation literature compares threats, faults, and detection in the stated technical setting. Judicial decisions and model endorsements establish particular contractual interpretations or text. Prudential material addresses portfolios and governance. Each category is scientifically useful within its own observational scope.
The National Association of Insurance Commissioners (2026) cyber insurance market report provides an aggregate U.S. market measure. Its 2025 data show approximately USD 9.46 billion in direct written premium including alien surplus lines, compared with USD 9.14 billion in 2024; the domestic series is approximately USD 7.46 billion for 2025. These published premium figures give market context. They do not isolate connected buildings, count realized cyber-physical incidents, establish their share of insured property payments, or supply the exposure-years needed to estimate relevant building loss frequency.
A building-specific claims dataset would need a defined geographic and temporal population, observed policies, buildings or building-years, applicable policies and limits, incident identifiers, damage measures, claim components, dates, and payment/reserve development. Where building count is impractical, another exposure basis might be described, such as insured site-years, controlled equipment-years, or value-years. Those bases answer different questions. A change in reporting rules or market participation should not be interpreted automatically as a change in hazard.
| Observation | Valid interpretation | Invalid inference without further data |
|---|---|---|
| Disclosed vulnerability | A specified technical issue exists in stated products | Number of damaged or insured buildings |
| Documented BAS interruption | Operational effect in the reported setting | Amount of property damages or insurer payments |
| Testbed classification result | Performance in specified experimental conditions | Field-wide claims attribution accuracy |
| Model policy exclusion | Contractual treatment in that illustrative text | Frequency with which it appears in bound policies |
| Aggregate cyber market premium | Market volume under the source reporting scope | Building-specific claim frequency or loss ratio |
Causal identification and missing data
A dataset of reported claims is conditioned on insured status, reporting behavior, the ability to recognize a loss, policy triggers, deductible thresholds, and insurer practices. Incidents below a deductible may not appear. Events recorded under property insurance may not carry a cyber classification. Events found by sophisticated monitoring may be overrepresented among firms implementing controls. A retrospective association between controls and claim frequency can therefore be biased in either direction.
Suppose monitored buildings report more events than unmonitored ones. The difference could reflect detection rather than greater underlying hazard. Suppose older buildings show fewer cyber claims; these may have less connectivity, weaker monitoring, or less affirmative coverage. A comparison should control for occupancy, asset value, operating and environmental conditions, policy terms, relevant exposure duration, and selection into insurance. The remaining uncertainty must be explicit.
Censoring and development also matter. A property damage estimate can change after equipment disassembly; business interruption losses develop over the restoration period; liability claims can emerge later. Date fields should distinguish event, discovery, notice, report, payment, reserve evaluation, and closure. A loss ratio calculated from incurred claims and earned premium differs from one based on cash paid and written premium. A study must report the respective periods, gross or net basis, and treatment of reopened claims.
Illustrative detection sensitivity
The experimental work by Li et al. (2024) reports a joint classification method for specific fault and attack conditions. Why might even a high-quality technical detector be insufficient as the sole claims attribution evidence? The base rate of relevant incidents in the population matters.
For an independent hypothetical binary classification example, assume that 1% of evaluated anomalies actually arise from malicious manipulation, sensitivity is 90%, and the false-positive rate among nonmalicious anomalies is 5%. Bayes' identity gives
P(malicious | positive) = 0.90 × 0.01 / (0.90 × 0.01 + 0.05 × 0.99) = 0.153846.
Only about 15.4% of positive detections would represent malicious events under these assumed conditions. The example is separate from the cited multiclass HIL experiment and does not estimate the performance of its detector. If the base rate in a specialized investigation is higher, the posterior changes. The intended lesson is that a technical classification must be interpreted with evidence about its sampling population and prior probability, including selection into the claims investigation.
Frequency-severity and conversion uncertainty
Let N be the number of distinct technical incidents in a specified exposure period. For an illustrative model assume E(N) = nλ for n homogeneous site-years and a constant incident rate λ. Assume incident-level expected payments conditional on physical damage status are stable; an expectation relation is
E(S) = nλ [(1 - q)m_0 + q m_1],
where q is the conditional probability of tangible damage given the defined incident, and m_0 and m_1 are expected insurer payments including zero-payment outcomes, respectively without and with tangible damage. No independence assumption across buildings is needed for this mean formula if the marginal expectations and incidence conditions are as stated; dependence remains relevant to aggregate risk. For real data, heterogeneous exposure, incident deduplication, correlated severity, and time-varying protection require more detailed treatment.
Assume 2,000 building-years, λ = 0.03 incidents per building-year, and q = 0.10. Sixty expected incidents include six expected damage events. Let a non-damage event have an assumed 60% probability of eligible payment averaging USD 20,000 among positive payments, so m_0 = USD 12,000. Let a damage event have an assumed 75% probability of payment averaging USD 180,000 among positive payments, so m_1 = USD 135,000. Expected annual insurer payment is 60 × (0.9 × 12,000 + 0.1 × 135,000) = USD 1,458,000, or USD 729 per building-year. All inputs are hypothetical and the result omits expenses, capital cost, reinsurance, and profit.
| Conditional physical damage probability | Expected payment per incident | Expected aggregate payment |
|---|---|---|
| 5% | USD 18,150 | USD 1,089,000 |
| 10% | USD 24,300 | USD 1,458,000 |
| 20% | USD 36,600 | USD 2,196,000 |
The sensitivity isolates the conversion probability while holding other inputs fixed. In practice, changing a building's safeguards might alter the incident frequency, probability of damage, repair duration, and severity simultaneously. The scenario therefore has an explanatory function rather than demonstrating a causal effect of an intervention. In the illustrative base case, an absolute one percentage-point change in q changes expected annual payment by 60 × 0.01 × (135,000 − 12,000) = USD 73,800. The magnitude shows why an insurer needs empirical evidence about conversion from technical disruption to physical loss rather than relying exclusively on threat frequency.
A contract scenario can change the conditional expected payment without changing the physical probability. For example, a new exclusion might move part of m_1 toward zero, while a newly purchased physical damage layer might increase it. A genuine pricing exercise needs contract-version information, inflation, policy limits, expenses, conditional recoveries, and a credible loss distribution. The manuscript's calculations intentionally do not produce commercial rates.
Validation steps and unresolved empirical quantities
The study's material claims were assessed against the accessible checked passages, with source-specific limitations. Screenshots of the LMA clauses and EIOPA paragraph page support the specific wording discussion. Official Ohio opinion text was corroborated by a reproduced full opinion; the requested official PDF screenshot could not be retrieved. The NAIC table extract was available in indexed text; its screenshot was unavailable. The empirical building studies were accessible at repository abstract level, so no methods or dataset results beyond the reported content are attributed to them.
No source in this research corpus establishes an exposure-adjusted estimate of physical cyber claims for all smart buildings, verifies the size of coverage gaps across the world, proves actual claims payment for the cited Finnish event, or permits causal estimation of the loss reduction associated with a specific installed safeguard. These are explicit research boundaries. More complete original incident records, locally executed engineering studies, contracts actually in force, insurer claim files, and exposure denominators would be needed to answer those empirical questions.
Integrated Scientific Synthesis and Research Agenda
Joining engineering and insurance evidence
The most useful insured unit for this research is a specified site-function-interest-contract combination, linked to incident events and external dependencies. A building is not one risk in a simple sense: an owner, tenant, facility operator, contractor, and visitor can have different interests exposed through the same equipment. A policy is not one physical asset: it can insure several sites and different classes of loss. The linked unit permits the researcher to compare incident pathways with actual contractual responses.
This is an analytical application of established concepts, not a proposed new scientific doctrine. Its value is that it reconciles evidence from fields using different units. Engineering observes functions, protective barriers, and states. Cybersecurity observes assets, privileges, faults, and events. Insurance observes contracts, claims, payments, and recoveries. A relational empirical record can preserve their distinct identities and connect them without forcing them into a single count.
The analytical contribution of this study is to identify where the connection is commonly lost. A technical advisory without installed-version information cannot establish a site exposure. A recorded attack without physical measurements cannot establish the mechanism of property damage. A documented damage event without policy wording cannot establish indemnity. A paid claim without exposure denominators and event identifiers cannot establish a representative frequency or independence assumption. Each missing link produces a different scientific uncertainty and requires a different kind of further evidence.
Underwriting and mitigation evidence
At underwriting, a minimum defensible dataset should describe the scope of the building systems, physical damage susceptibilities, safe operating alternatives, asset and interruption values, major external service dependencies, recovery capabilities, and actual contemplated coverage. An inquiry about the existence of multifactor authentication, for example, must specify which remote access or identity functions are covered, exceptions, and how the control is maintained. A mechanical protection inquiry must establish its condition and independence.
The NIST Cybersecurity Framework 2.0 (Pascoe et al., 2024) gives organizations a general risk-management structure. For buildings, governance and asset identification can be joined to protection, detection, response, and recovery activities compatible with physical safety. The DOE operational technology guidance illustrates the practical importance of legacy equipment, access control, patch management, and inventories. A compliance label alone is less informative than a dated, scoped record of the concrete control and its performance.
The insurance contract may support prevention by setting conditions, encouraging loss control, or rewarding defensible reductions in risk. The incentive can fail if underwriting prices controls that are easy to declare but difficult to verify, or if a single compliance questionnaire does not distinguish high-consequence systems from ordinary endpoints. The appropriate analytical goal is to connect a verified control to the causal paths it can change and to acknowledge remaining uncertainty about actual effectiveness.
Claims coordination and forensic independence
At notification, the building operator must protect occupants and mitigate further damage. An insurer can expedite evidence collection and coordinate coverage analysis without treating the earliest incident label as final. The technical examiner should document competing explanations and relevant physical evidence. The loss accounting team should identify the ownership, economic interest, and cost components. The coverage team should identify governing wording, causal standards, exclusions and any reservation of rights. The insurer may later correct the claim classification as evidence develops.
An audit-ready loss record would connect each material conclusion to an original artifact, acquisition time, provenance, and scope. The absence of a log should be described as an evidence gap, rather than transformed into proof of concealment. Where a vendor supplies logs or remediation advice, independent corroboration may be especially useful if the vendor's own liability is disputed. The goal is a traceable account of what is known and how it was determined.
There is a separate ethical dimension. Occupancy, health, security, and access records can reveal sensitive information about individuals. Evidence collection should be proportionate to the purpose, restricted to authorized parties, and consistent with applicable data protection and legal retention duties. Insurance investigation must not expand a building control incident into an unnecessary disclosure of occupants' private lives.
Coverage design and accumulation management
The comparison of LMA5400 and LMA5401 demonstrates that the financial significance of an identical physical consequence can vary with a small but material contractual distinction. A portfolio modeled using policy labels instead of actual endorsement terms will misstate at least some payment distributions. The models should therefore retain versions, applicable loss triggers, tangible damage restrictions, exclusions, limit structures, and revenue or expense bases.
The prudential sources cited above encourage explicit attention to non-affirmative cyber and concentration. A practical implementation distinguishes count of sites using a dependency, aggregate insured value, aggregate interruption exposure, physical susceptibilities, shared protection and recovery arrangements, and contractual covered fractions under specified scenarios. It should evaluate the insurer's retained loss after reinsurance, including possible disputes where the treaty and original contracts differ in their event definitions.
A stress test should record what is assumed rather than embed a false sense of empirical precision. Useful scenarios include a regional operational disruption with limited local physical effects, a widely shared vendor service interruption, an accidental software change affecting multiple sites, and a malicious event with potential physical consequences. For each, specify the sites affected, ability to operate locally, duration assumptions, damage conditional on circumstances, recovery bottlenecks, relevant coverage, and aggregation provisions. The scenario results can then be compared with risk appetite and capital resources without treating stress assumptions as measured probabilities.
Research agenda and international comparability
Future empirical research should create a longitudinal site-level dataset with linked but distinct records for installed system versions, protective controls, operational incidents, physical outcomes, policy terms, and claims. A common event identifier would allow accumulation analysis without conflating independent physical causes. Confidential records can be de-identified and aggregated where necessary, while preserving exposure denominators and version changes.
A second task is evaluation of mitigation effectiveness. Prospective observations before and after a documented control change, matched sites, and transparent adjustment for confounders could provide stronger causal evidence than retrospective questionnaires alone. Event studies should document both disrupted and unaffected functions and compare physical faults with cybersecurity incidents. Claims analysts should additionally measure changes to reporting propensity and policy terms.
A third task is comparative contract research. A sample should be constructed by jurisdiction, insurer, product, period, and actual bound form rather than convenience selection of publicly accessible model wordings. Researchers should separately code positive grants, cyber exclusions, resulting-damage exceptions, tangible and intangible property definitions, dependency coverage, waiting periods, and aggregation language. Interpretation should reflect local law and procedural context. The legal examples here reveal relevant distinctions, but they do not represent a global coverage census.
Finally, cross-industry accumulation research should measure concentrations in service providers, control software, equipment families, and recovery resources. It must allow one physical site to participate in several dependency groups. A shared digital dependency can link sites that never share a geographic catastrophe, while geographic conditions can determine which sites experience physical damage from a common disruption. Research that observes both dimensions can improve stress testing and allocation of prevention and insurance capacity.
Conclusion
Smart building cyber-physical insurance risk arises through the interaction of digital authority, physical operating conditions, engineering safeguards, economic interests, and the contract terms that govern indemnity. Its investigation requires a continuous evidentiary connection from an identified event to a changed physical or operational state, a demonstrable loss, and an applicable contractual response.
The documentary research establishes that building automation belongs to operational technology, that connected device security involves both product capabilities and installed-system conditions, and that attacks and physical faults can present similar observable symptoms. Public incident narratives give evidence of third-party trust and heating disruption without establishing physical damage or actual insurance payment. The legal and model wording comparison shows that property damage and software impairment can receive different treatment according to the contract, cause, and jurisdiction. The prudential record identifies silent cyber and concentration as substantive underwriting concerns.
The original analytical synthesis concerns the relation between evidence and the insured loss distribution. In the frequency-severity illustration, conditional damage conversion and contract acceptance determine expected insurer payment; under equal marginal expectations, shared dependencies can greatly increase variance. These calculations demonstrate the importance of conversion and dependence and do not estimate the actual global smart-building loss distribution.
From this analysis follow three conclusions for scientific practice. Technical risk must be measured in relation to physical function and independent protection. Claims causation must join digital, physical, contractual, and financial records with explicit treatment of alternative explanations. Portfolio risk assessment must retain common digital dependencies and exact coverage identities alongside physical location and insured values.
The current public evidence supports these conclusions as a documentary and conceptual synthesis. It does not establish building-specific observed insurance frequencies, validated real-world conversion rates, or global coverage prevalence. The conditions for stronger inference are identified in the research agenda. The long-term scientific objective is an empirically grounded account of how the safety and continuity of built environments, digitally mediated operation, and the financing of residual loss interact.
References
Stouffer, K., Pease, M., Tang, C., Zimmerman, T., Pillitteri, V., Lightman, S., Hahn, A., Saravia, S., Sherule, A., & Thompson, M. (2023). Guide to Operational Technology (OT) Security. NIST SP 800-82 Rev. 3. National Institute of Standards and Technology. doi:10.6028/NIST.SP.800-82r3. Checked source. Accessed November 18, 2026.
Fagan, M., Megas, K., Scarfone, K., & Smith, M. (2020). IoT Device Cybersecurity Capability Core Baseline. NISTIR 8259A. National Institute of Standards and Technology. doi:10.6028/NIST.IR.8259A. Checked source. Accessed November 18, 2026.
National Institute of Standards and Technology. (n.d.). IoT Device Cybersecurity Requirement Catalogs: Technical Capabilities. NIST Cybersecurity for IoT Program. Checked source. Accessed November 18, 2026.
Li, G., Ren, L., Fu, Y., Yang, Z., Adetola, V., Wen, J., Zhu, Q., Wu, T., Candan, K. S., & O'Neill, Z. (2023). A critical review of cyber-physical security for building automation systems. Annual Reviews in Control, 55, 237–254. doi:10.1016/j.arcontrol.2023.02.004. Checked source. Accessed November 18, 2026.
Li, G., Ren, L., Pradhan, O., O'Neill, Z., Wen, J., Yang, Z., Fu, Y., Chu, M., Huang, J., Wu, T., Candan, K. S., Adetola, V., & Zhu, Q. (2024). Emulation and detection of physical faults and cyber-attacks on building energy systems through real-time hardware-in-the-loop experiments. Energy and Buildings, 320, 114596. doi:10.1016/j.enbuild.2024.114596. Checked source. Accessed November 18, 2026.
Johnson Controls. (2015; updated 2018). Metasys Building Automation System Information Disclosure Vulnerability (CVE-2018-10624). Security Advisories catalog, entry with March 17, 2015 and August 27, 2018 dates. Checked source. Accessed November 18, 2026.
U.S. Department of Energy, Federal Energy Management Program. (n.d.). Operational Technology Cybersecurity for Energy Systems. U.S. Department of Energy. Checked source. Accessed November 18, 2026.
U.S. Senate Committee on Commerce, Science, and Transportation. (2014). Protecting Personal Consumer Information from Cyber Attacks and Data Breaches. Hearing, March 26, 2014, 113th Congress; included Majority Staff Report, A Kill Chain Analysis of the 2013 Target Data Breach. U.S. Government Publishing Office. Checked source. Accessed November 18, 2026.
Lloyd's Market Association. (2019). Property and Marine Cyber Clauses. LMA market announcement. Checked source. Accessed November 18, 2026.
Lloyd's Market Association. (2019). Property Cyber and Data Endorsement. LMA5400, property direct and facultative model wording. Reproduced clause text, Insurance Endorsements. Checked source. Accessed November 18, 2026.
Lloyd's Market Association. (2019, November 11). Property Cyber and Data Exclusion. LMA5401, property direct and facultative model wording. Reproduced clause text, Insurance Endorsements. Checked source. Accessed November 18, 2026.
National Ink & Stitch, LLC v. State Auto Property & Casualty Insurance Co., 435 F. Supp. 3d 679 (D. Md. 2020), No. SAG-18-2138, memorandum opinion, Jan. 23, 2020, ECF No. 39. Court opinion reproduced by Justia. Checked source. Accessed November 18, 2026.
EMOI Services, L.L.C. v. Owners Insurance Co., 170 Ohio St. 3d 78, 2022-Ohio-4649 (Supreme Court of Ohio, Dec. 27, 2022). Checked source. Accessed November 18, 2026.
Prudential Regulation Authority. (2017). Cyber insurance underwriting risk. Supervisory Statement SS4/17. Bank of England. Checked source. Accessed November 18, 2026.
European Insurance and Occupational Pensions Authority. (2022). Supervisory statement on management of non-affirmative cyber exposures. EIOPA-BoS-22-414, EIOPA(2022)0035012. Checked source. Accessed November 18, 2026.
New York State Department of Financial Services. (2021). Cyber Insurance Risk Framework. Insurance Circular Letter No. 2 (2021). Checked source. Accessed November 18, 2026.
European Commission. (n.d.). The Cyber Resilience Act: Summary of the legislative text. Directorate-General for Communications Networks, Content and Technology. Summary of Regulation (EU) 2024/2847. Checked source. Accessed November 18, 2026.
National Association of Insurance Commissioners. (2026). Report on the Cybersecurity Insurance Market. NAIC, reporting year 2025. Checked source. Accessed November 18, 2026.
Lloyd's & Cambridge Centre for Risk Studies. (2022, June 30). Shifting Powers: Physical Cyber Risk in a Changing Geopolitical Landscape. Lloyd's Futureset. Checked source. Accessed November 18, 2026.
Pascoe, C., Quinn, S., & Scarfone, K. (2024). The NIST Cybersecurity Framework (CSF) 2.0. NIST CSWP 29. National Institute of Standards and Technology. doi:10.6028/NIST.CSWP.29. Checked source. Accessed November 18, 2026.