Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Cyber Insurance Public Incident Reporting and Underwriting Information

Abstract. Public cyber incident reporting has become an important component of the information environment within which insurers select risks, establish contract terms, estimate losses, and monitor accumulating exposures. Its actuarial relevance depends on the route from an underlying security event to detection, legally defined notification, availability to an underwriter, and the eventual measurement of insured loss. This monograph investigates that route through comparative documentary research on securities disclosure, privacy breach notification, operational resilience reporting, payment-triggered reporting, insurance applications, supervisory statistics, and selected issuer disclosures. The analysis distinguishes information available to investors from that supplied to competent authorities, affected individuals, insurers, and restricted industry exchanges. It establishes how event selection, reporting delays, evolving severity estimates, enterprise boundaries, contract definitions, and correlated dependencies alter the interpretation of observable incident histories. Original calculations demonstrate the sensitivity of observed incidence, conditional loss severity, expected indemnity, and aggregate risk to reporting probabilities and detection quality. These are hypothetical analytical demonstrations, not empirical estimates of market parameters. Evidence from three documented corporate cases illustrates the divergence between an initial announcement, subsequent operational or financial information, and eventual insurance recovery. The conclusion is that public reports are most valuable as dated, source-specific corroborating evidence linked to private underwriting information, exposure denominators, coverage terms, and verified loss development. Public disclosure alone cannot identify the probability or severity of insured claims, and better reporting can raise observed incident counts while underlying security improves. The study contributes a comparative interpretation and reproducible methodology for analyzing this information problem without asserting a new named theory or an empirically established causal effect of disclosure on premiums.

Keywords: Cyber insurance; underwriting; incident disclosure; breach notification; information asymmetry; reporting delay; exposure measurement; risk classification; loss intelligence; systemic cyber risk; actuarial data; operational resilience.

Introduction

Cyber insurance transforms uncertain financial consequences of digital operations into contractual obligations. Underwriting determines which exposures an insurer is willing to assume, the terms under which they can be accepted, and the compensation required for their expected costs and uncertainty. In this process, the public record of cyber incidents presents an attractive but structurally incomplete source of information. It identifies events that might otherwise remain invisible outside an organization and can reveal dependencies, affected operations, regulatory exposure, or delayed recovery. Its interpretation, however, requires knowledge of the purpose for which each item of information was originally produced.

The scientific object of this work is the relation between regulatory incident disclosure, the time at which information becomes available, and underwriting information. The central questions concern what public records actually establish, how reporting selects events from a larger risk population, and how insurers can translate observable history into decisions about future uncertain liabilities. The analysis proceeds through four distinct units: the security event, the affected organization, the disclosure or notification, and the insured claim. Each unit has its own time dimension and jurisdictional meaning. A fifth unit, the shared technology dependency, connects organizations exposed to a common event.

This work uses comparative legal and documentary analysis, critical evaluation of original published research, and mathematical illustrations. The research cutoff is November 18, 2026. The jurisdictional examples concern United States securities and health privacy obligations, European Union directives and regulations, Canadian federal private-sector privacy law, and Australian payment reporting. The relevant legal instruments are examined within their described scope; country-specific implementing rules, every U.S. state requirement, and every sectoral exception are beyond the documentary coverage. CIRCIA is examined as a statutory design with contemporaneous rulemaking status. The record of corporate cases is purposively selected to illustrate information sequences rather than to estimate frequency, average loss, compliance rates, or average insurance recovery.

The research examines a primary carrier application, historical academic analysis of actual policy filings, official statistical reporting, and public filings. No confidential insurer application or claim microdata, private incident-response files, bound policies for the named companies, or individual quoted premiums are available. Accordingly, findings about insurance practice are confined to the documented mechanisms and the explicit assumptions of the illustrations. The analysis of legal rules is educational and comparative; application to a particular insured contract requires the actual instrument, governing law, parties, and operative facts. Each external source is identified in the text and in the bibliography, with its examined scope recorded separately in the evidence record.

Conceptual and Historical Foundations

Risk, observation, and contractual consequence

An incident report is a representation of an event, formed within a specified reporting process. The event can involve unauthorized access, disruption of an information system, loss of confidentiality, manipulation of records, extortion, or a service dependency failure. Whether it generates a legally reportable case depends on the applicable threshold; whether it creates a recoverable insurance claim depends on the policy and facts. An event can have severe technical consequences yet impose little immediate financial cost, while a less technically dramatic interruption can create substantial lost revenue and additional expenses. A payment demand, an actual payment, and an indemnity reimbursing that payment are separate economic observations.

For underwriting, the appropriate objective is the conditional distribution of future covered losses over a defined policy horizon and exposure set. This includes probabilities of claim-generating events, financial severity before the insurance contract, contractual response, and the dependence of outcomes across insured organizations. A public incident report is one possible input into that estimation process. Its presence can strengthen the evidentiary basis for questioning an applicant or reviewing portfolio correlations. Its absence does not, without an observation model, establish an event-free history.

This distinction is especially important in cyber insurance because the reporting unit is often different from the underwriting unit. The listed parent corporation, the attacked operating subsidiary, the data controller, a service provider, and the insured entity named in the schedule may not be identical. The same intrusion can appear in the separate announcements of several customers. A corporate press statement may describe effects on an entire business group even if only a smaller legal entity holds the relevant insurance. Conversely, one insurance contract may insure several subsidiaries and cover consequences of a breach not separately reported under each subsidiary's name.

Historical incident classifications are shaped by the institutions that collect them. The study of reported events by Romanosky (2016) illustrates the problem: for the period 2005–2014, cost data appeared in only 887 of 11,705 observations. This specific historical evidence warns against equating an event database with a complete severity database. It does not establish that the same fraction characterizes current disclosures.

From confidentiality to operational resilience

Reporting developed in multiple institutional settings. Investor disclosure concerns information relevant to securities decisions; privacy notification serves affected persons and oversight bodies; operational reporting seeks timely awareness of disruptions and coordinated response; insurance notification initiates a contractual process. The same incident may traverse all these settings. A notice intended to warn a patient about compromised records can establish a privacy exposure without providing a meaningful business interruption estimate. A securities filing can quantify a corporate financial effect without identifying the insured contract terms or the number of persons affected.

Security governance standards contribute a different perspective. The NIST Cybersecurity Framework 2.0 (Pascoe et al., 2024) organizes the management of cybersecurity risk at an organizational level. A framework can give an insurer and an applicant a common vocabulary for discussing control ownership, monitoring, incident response, and recovery. The framework's publication supplies guidance; it does not supply a universally validated multiplier for premiums. A maturity assessment must establish the actual controls, their tested effectiveness, their scope, and the date of the assessment.

The expansion of regulatory reporting therefore changes the composition of accessible evidence. Public authorities can see activities that private organizations previously shared only selectively, while confidential reporting channels can create valuable aggregate intelligence without disclosing every underlying case. The institutional distinction has direct scientific consequences: an increase in available observations may reflect an increase in attacks, improved detection, stronger reporting obligations, or changes in information distribution. None of these effects can be isolated by counting published headlines alone.

Information asymmetry and the insurance process

Insurance transactions occur under asymmetric information. The applicant typically possesses detailed knowledge of its own systems and recent operational history, while an insurer combines that knowledge with its experience of numerous risks and external data. Both parties face uncertainty concerning vulnerabilities, threat evolution, and future financial consequences. An applicant may have investigated a prior event but remain unable to establish the entire scope. The insurer may possess useful aggregate claims experience but have limited knowledge of a particular operational environment.

Independent public information can reduce certain types of asymmetry by furnishing a dated point of comparison. If an application states that no material interruptions occurred within the relevant lookback period and a public filing documents a disruption, an apparent discrepancy warrants clarification. The wording of the question, identity of the affected entity, date range, and applicant's knowledge must all be reconciled. A difference between records can result from a narrower contractual definition, an acquisition, a late discovery, or an honest distinction between a technical event and an insured claim. It is an evidentiary question before it becomes an underwriting or coverage conclusion.

An insurer also selects and classifies risks. This selection changes the relationship between publicly observed risk and subsequent insured experience. Risks refused at underwriting contribute no claims to that insurer's usual policy exposure database, even if those organizations later experience publicly reported incidents. Accepted applicants with higher retentions can generate fewer payable claims than similar organizations buying broad first-dollar protection. Renewals and new business can have different histories. Prior-insurance status and coverage limits are therefore integral to interpreting claim frequencies.

Underwriting information as a decision-time object

Underwriting information should be defined by what was available and permissible at the decision date. A historian may have the final investigation and subsequent financial disclosures; an underwriter quoting a year earlier did not. A retrospective assessment that imports these later records into its original feature set measures a different forecasting problem. It can greatly overstate predictive performance.

A useful incident history records both when something occurred and when it could reasonably be known through the relevant channel. A date printed on a filing establishes its documentary time, not necessarily the onset of intrusion. The assessment of severity can change after that date. A record must preserve superseded descriptions so that a decision can be reconstructed using the version available at the time.

There is a further distinction between a disclosed historical incident and an ongoing exposure. A remediated compromise may have limited relevance to the risk during a future policy term, whereas an unresolved vendor dependency or repeated failure to contain intrusions may remain material. Historical reporting acquires forward-looking significance through mechanisms that persist: residual vulnerabilities, untested recovery, concentration, business model, governance deficiencies, or uncertainty about unresolved liabilities.

The primary scientific conclusion of this chapter is that public incident reporting offers conditional evidence about exposed organizations. Its value depends on the correspondence between the source's original object and the underwriting target. Transparency strengthens this evidence when it documents changing facts and enables corroboration. It also creates selection effects that must be handled explicitly in prospective risk estimation.

Primary Documents and the Scope of Incident Reporting

The United States securities disclosure regime

The SEC's 2023 cybersecurity disclosure rule establishes the ordinary timing for domestic registrant disclosure under Form 8-K Item 1.05 by reference to determination of materiality, with four business days for filing. The determination is to be made without unreasonable delay after discovery. The required subject matter concerns material aspects of the event and its impact or reasonably likely impact. The instructions contemplate amendments where required information was initially unavailable and do not demand technical detail whose disclosure would impede the response. The rule has specified exceptions and scope conditions; it does not turn every cyber incident into a public filing.

Two underwriting consequences follow. First, the event selected by Item 1.05 belongs to the class of incidents that the covered registrant determines to meet the securities-law standard. The existence of a report neither identifies a minimum covered loss nor specifies a common financial threshold across issuers. Materiality depends on the circumstances of the issuer and the information relevant to investors. Second, the filing deadline starts after a determination, not necessarily after occurrence or initial detection. Event-to-filing lag cannot be inferred from the four-business-day requirement alone.

The public character of securities filings does not extinguish the value of private communications. Gerding's selective disclosure statement (2024) explains that Item 1.05 itself does not prohibit additional discussions with counterparties. The statement is a director's explanation rather than a new statutory rule. The legal and contractual constraints on sharing particular nonpublic information must still be evaluated. Consequently, a carrier that has lawfully obtained additional evidence through the application process may possess a materially richer history than that obtained from securities filings alone.

For research purposes, a collection of Form 8-K documents needs accession identifiers, item numbers, amendments, filing timestamps, issuer identifiers, and the actual entities affected. A rule-based extraction should distinguish a disclosure under the specific item from an announcement under another filing item. An amendment should ordinarily be associated with the event it updates. The count of filings is not the count of distinct incidents, and the count of incident disclosures by listed parent groups is not the count of insured policy years.

European Union reporting

The NIS 2 Directive (European Parliament & Council, 2022) establishes significant-incident reporting for covered essential and important entities through national authorities or CSIRTs. Article 23 provides a warning within 24 hours of awareness and ordinarily an incident notification within 72 hours, with subsequent reports and a final report normally within one month after the notification. Trust service providers have a special shorter notification rule. National transposition and the actual entity classification matter to legal applicability. Notification to an authority is a distinct action from disclosure to the general public.

The timetable reflects the institutional purpose of staged reporting. Authorities benefit from early identification before the complete technical and financial impact is known. A subsequent report can refine the initial classification, mitigate false assumptions, and provide more detailed information. For an insurance analyst, this means that early and late versions may describe different evidentiary stages of one event. A revised estimate may change aggregate loss analysis without increasing the number of distinct events.

Articles 33–34 of the General Data Protection Regulation (European Parliament & Council, 2016) relate notification obligations to personal data breaches and risks to individuals. Article 33 provides for notice to the supervisory authority without undue delay and where feasible within 72 hours of awareness, subject to the risk threshold; Article 34 addresses communication to data subjects. This is a separate information population from securities materiality. A breach of sensitive records can warrant privacy notification even where its immediate financial consequences for a corporation are small.

The Digital Operational Resilience Act (European Parliament & Council, 2022) concerns the operational resilience of financial entities within its scope, including relevant insurers as financial entities. Article 19 establishes major ICT-related incident reporting to competent authorities, and Article 22 provides for annual anonymized and aggregated reporting by the European supervisory authorities. The DORA reporting technical standards (European Commission, 2025) specify, in the ordinary case, initial reporting within four hours after major classification and no later than 24 hours after awareness; an intermediate report follows within 72 hours of initial submission. The regulatory text has qualifications for particular circumstances. Those time limits cannot be converted into blanket deadlines for every firm or every policyholder event.

Underwriting research must preserve a boundary between an insurer as the reporting financial entity and the insurer as the risk bearer for its customers. A cyber event affecting the insurer's own claims platform may generate regulatory and operational consequences. A cyber event affecting an insured enterprise may generate claims under a cyber policy while the insurer's own systems function normally. Both matter to insurer financial resilience, but they enter different exposure populations and loss accounts.

Health privacy and Canadian reporting

The U.S. Department of Health and Human Services' Breach Notification Rule guidance describes individual notice, notice to the Secretary, and media notice for covered entities following qualifying breaches of unsecured protected health information. Individual notice is normally due without unreasonable delay and within 60 days of discovery; media notice concerns breaches affecting more than 500 residents of a state or jurisdiction. The threshold for prompt Secretary notice is 500 or more affected individuals. Smaller cases may be reported to the Secretary annually within the guidance's separate timetable.

These thresholds make the observable distribution discontinuous. A dataset restricted to larger publicly accessible cases cannot establish a universal relationship between record counts and loss. Nor should the analyst collapse the media threshold and the Secretary threshold into a single numerical rule: geography and audience differ. An organization providing several services may report one breach on behalf of multiple covered entities where the rules permit delegated action, further complicating identity resolution.

Canadian federal law follows a risk-based privacy approach. Sections 10.1–10.3 of the Personal Information Protection and Electronic Documents Act (Parliament of Canada, 2000, as amended) address reports to the Commissioner, notice to affected persons and maintenance of breach records. The reporting and notice trigger involves a real risk of significant harm, and covered reports or notices are made as soon as feasible after determining that the breach occurred. Record keeping encompasses a broader population of breaches than those meeting the reporting threshold under section 10.1.

This arrangement provides a useful contrast: the internal universe of recorded breaches has a distinct legal purpose and may be wider than the subset externally reported. Insurers examining an applicant's experience could, where lawfully appropriate, request summaries of its internal records and compare the material cases to external notifications. The results would need to be interpreted under the federal law's actual scope and relevant provincial arrangements. A record produced to protect an individual should not be presumed to have been created for commercial reuse.

Payment-triggered reporting and forthcoming rules

Part 3 of Australia's Cyber Security Act 2024 establishes a reporting obligation for qualifying ransomware payments, within 72 hours of making the payment or learning that another entity has made it, depending on the facts. Sections 29–30 constrain governmental use, disclosure, and secondary use of information. The Department of Home Affairs' Cyber Security Act implementation page identifies the implementing ransomware payment reporting rules. This analysis concerns the statute as made and the identified implementation framework, without purporting to certify every later legislative amendment.

A payment-report dataset is selected on payment behavior as well as exposure to extortion. Organizations refusing to pay in an otherwise similar event can be absent from that reporting population. Payment decisions depend on operational dependence, recovery options, demands, legal constraints, negotiations, and insurance arrangements. The selection is therefore economically endogenous. A database of payments cannot identify ransomware incidence without additional evidence about attacks and nonpayments.

In the United States, the reporting architecture under 6 U.S.C. § 681b (United States Congress, 2022) includes covered incident and ransom payment notifications and subsequent updates, with the effective dates of the relevant duties prescribed by the final rule. As of this research cutoff, the OIRA rulemaking review listing (November 18, 2026) showed RIN 1670-AA04 at pending review. This contemporaneous official status does not establish promulgation or an effective operational reporting obligation. The statutory design is discussed as such; any later implementation requires a new legal status check.

Comparison of reporting populations

Reporting setting Reporting object Time reference Principal information recipient
SEC Item 1.05 Material cyber incident of a covered registrant Materiality determination; ordinary four business days Public securities filing
NIS 2 Article 23 Significant incident in covered scope Awareness; staged 24/72-hour duties with exceptions CSIRT or competent authority
GDPR Articles 33–34 Personal data breach and applicable risk threshold Awareness or applicable communication duty Authority and, where required, affected persons
DORA and RTS Major ICT-related incident of financial entity Awareness and classification; staged reports Competent authority
HIPAA breach rule Qualifying unsecured PHI breach Discovery; recipient- and size-dependent duties Individuals, HHS, media where required
Canadian PIPEDA Breach with real risk of significant harm Determination; as soon as feasible Commissioner and affected persons
Australian Cyber Security Act Qualifying ransomware payment Payment or awareness of another's payment Designated Commonwealth body

The comparison summarizes the specific provisions and guidance discussed above, rather than asserting that national scopes or legal exceptions are equivalent. Four characteristics determine the evidentiary usefulness of each record: who is obliged to report, what threshold is applied, when the reporting clock begins, and which recipient can access which information. These characteristics are observable aspects of an institutional reporting process. They are also conditioning variables for any statistical analysis; ignoring them transforms different legal populations into an artificial common sample.

Information Timeliness and Underwriting Mechanisms

Observation as a staged process

Information timeliness has at least three relevant components: the interval between occurrence and detection, the interval between detection and the production of a report, and the interval between the report and its availability to an insurer. The latter may be positive when a report is restricted or published in another language, when an intermediary digitizes an official document, or when the insurer cannot reliably associate a company name with the correct applicant. A legitimately confidential report may never become public.

Let t0 represent event onset, td detection, tm determination of a reporting threshold, tr the first report to a specified recipient, tp its public availability when applicable, and tu the underwriting decision time. For a public reporting channel, a basic availability test is tp ≤ tu. Where the report is private, its actual lawful availability must be established directly. This notation describes existing temporal distinctions and does not assert a novel model.

The chronological fields should record their evidentiary quality: exact date from a filing, date stated retrospectively by the organization, interval of possible onset, uncertain date, or unavailable. Setting an unknown date to the filing date creates a spurious zero reporting lag. Imputing all unknown dates to the earliest possible date can bias the analysis in the opposite direction. If only event months are known, a study should acknowledge interval censoring rather than infer exact day-level performance.

Reporting speed and report completeness can move in different directions. An early notice is useful for escalation and portfolio monitoring but initially may contain incomplete information about the outage, affected records, or duration. A later report can be more specific but may arrive after the insurer's quotation or renewal. Versioned records permit an underwriter to use early signals in a conditional assessment without retrospectively claiming that later details were already established.

Conditioning on reporting

Let I indicate that an organization experiences a defined incident in an exposure year and R indicate that it becomes observable to a chosen data collector. For a group with incident probability p and probability q of observation conditional on an incident, the expected observed rate is p × q, assuming no false positive reports and at most one event per exposure year in this simplified example. Identifying p from the observed rate requires information about q. The observation probability can depend on severity, entity size, sector, legal scope, detection capability, reporting culture, and calendar time.

Consider two hypothetical portfolios, each with 10,000 organization-years. Portfolio A experiences incidents at a probability of 5% and observes 80% of them; its expected public count is 400. Portfolio B experiences incidents at 8% and observes 25%; its expected public count is 200. Comparing public counts would suggest that A is worse, although the assumed physical or operational event incidence is lower. The expected actual event counts are 500 and 800. This is a direct consequence of unequal reporting probabilities, not an empirical assertion about any industry.

Reporting changes can also reverse conclusions over time. Suppose a portfolio's actual incidence falls from 6% to 4%, while its observation probability rises from 30% to 70%. Its public rate increases from 1.8% to 2.8%. The rise of about 55.6% in observed reports accompanies a one-third decline in the underlying incident probability. An insurer using only the public rate would infer a worsening of security if it failed to account for the change in observation.

In practice, reporting probabilities are rarely known. When plausible bounds exist, sensitivity intervals can replace an unjustified single corrected estimate. If an observed rate is 3% and q is believed to lie between 0.30 and 0.75, the implied simple incident rate ranges from 4% to 10%. The bounds are conditional on the observation assumptions; they are not sampling confidence intervals. A sampling interval would additionally require the exposure count, model of dependence, and sampling process. When duplicates, uncertain classification, or false positive reports exist, even the simple relationship must be adjusted.

Information from an insurance application

A carrier can request information more specific to its intended contracts. The English-language AIG cyber application form (n.d.), in its prior-claims section, asks about significant incidents during a five-year period and, when applicable, the cause, corrective measures, available forensic documentation, and estimated losses. This primary application demonstrates the relevance of private event histories and post-event remediation. It is a single insurer document, not a census, common mandatory wording, or an executed policy.

The historical policy content analysis by Romanosky et al. (2019) examines filed insurance policy materials, underwriting questionnaires, and rate schedules from three U.S. states. It establishes a documented research route for distinguishing actual underwriting questions from general assumptions about insurers. Its sample and vintage constrain the generalizations that can be drawn. An insurer's current ability to obtain incident information and its rating response require current documents and decision data.

Public information can be used to formulate a focused follow-up rather than to infer an adverse answer automatically. When an applicant's description differs from a public source, the insurer may request reconciliation of corporate identity, lookback period, event type, reported severity, remediation, insurance history, and pending circumstances. A public report can raise a question even if the application did not ask about that event category. A contract consequence depends on the actual duty of disclosure and applicable law.

Questionnaire answers and technical controls must also be time-aligned. A control implemented after an incident may be relevant to prospective risk but could not have prevented that prior event. Evidence that the control presently exists is stronger if its scope and operation have been tested. An updated cybersecurity policy is different from proof of effective deployment across all covered operations. Documentation should preserve these distinctions rather than reducing each to a binary indicator of compliance.

Severity, timing, and loss development

Information often arrives in an order that differs from the order in which economic costs develop. Service interruption may be recognized quickly, restoration expenditures accumulate over weeks, and litigation can emerge months or years later. An early report can overestimate or underestimate ultimate losses. Insurer claims experience also develops as new costs are reported, reserves are established, claim eligibility is clarified, and payments are made.

Assume an incident has an early economic loss estimate of $400,000. A later business interruption analysis adds $300,000, and a liability settlement adds $500,000. Suppose these figures refer to distinct components and no offsets, producing $1.2 million of gross economic loss in the example. If a data system stores them as three separate incident totals, it can overcount frequency and miscalculate severity; if it retains only the first estimate, it misses $800,000 of later development. A proper record would preserve each component, the valuation date, and the evolving cumulative amount.

Event public reporting lag is analytically separate from claim reporting lag. To estimate ultimate insured losses, an insurer may use development methods appropriate for its exposure and reporting process. A public filing can provide a further external signal, but its lag distribution differs from the lag between occurrence and insurer notice. A reserving study must use the actual claims development process and applicable valuation convention.

System and entity linkage

A shared platform changes both exposure and timing. The provider may report first, while affected customers learn about the problem at different times. A customer may not itself be compromised by an attacker but may suffer a covered or uncovered interruption because its service provider is offline. Public dependency information can consequently matter even when an individual applicant has no reported direct security incident.

A data system should distinguish a root event from its organizational consequences. One root incident can create many entity-exposure records, separate first-party and third-party financial outcomes, and multiple legal disclosures. Entity matching should be based on stable identifiers where available, corroborated relationships, and the relevant corporate ownership period. Name similarity alone is weak evidence.

For prospective underwriting, verification of the insured's current use of a service provider is particularly important. A company that formerly used an affected supplier may have migrated before the relevant policy term. An apparent association with a vendor may instead relate to a subsidiary that is no longer within the insurance program. Portfolio analysis must capture actual services, reliance, alternatives, contractual responsibility, and concentration rather than merely flagging every organization mentioned near the supplier's name.

Comparative Measurement of Pricing Information

Data populations and denominators

Insurance pricing information is meaningful when its numerator and denominator describe a coherent population. An incident count from public registrants divided by an insurer's number of policies produces a ratio without a common exposure basis. A record count of affected persons is not the same as a count of breaches or organizations. Direct written premium is a measure of written insurance business, not an estimate of the cost of underlying cyber incidents. Claims paid, losses incurred, claims reported, and economic damage each represent different measurements.

Exposure denominators may include organization-years, insured policy-years, payroll, revenue, number of records, technology dependencies or transactions, depending on the peril and outcome. For a firm-level attack frequency study, organization-years and consistent event definitions are natural starting points. For expected claim costs, policy terms and coverage layers must be associated with the time during which the insurer bears risk. Revenue and company size may help classify exposures but are not substitutes for the policy-year denominator in a frequency model.

The 2026 NAIC market report provides a contemporary example of the distinction between insurance data and broader incident intelligence. Its reporting basis includes the statutory Cyber Supplement and alien surplus lines data from the International Insurers Department; it distinguishes primary business, excess business, and endorsements. Figure 1 reports total U.S. direct written premiums of approximately $9.14 billion for 2024 and $9.46 billion for 2025, including alien surplus lines. These market amounts do not measure the severity of public incidents or a change in individual prices.

The GAO cyber insurance study (2021) identifies limitations of historical loss information and inconsistent terminology as challenges for insurance product development. That earlier institutional evidence supports the importance of comparable information but does not demonstrate that collecting public reports alone solves the actuarial identification problem. Different market segments can also have different reporting coverage and access to private incident intelligence.

Sampling severity

An incident corpus can systematically overrepresent large or conspicuous losses. Consider a hypothetical exposure population with 100 incidents: 80 cost $10,000 each and 20 cost $500,000 each. Gross loss is $10.8 million, with average severity $108,000.

Suppose a public data collector observes 10% of small events and 90% of large events. Its expected observed sample has 8 small and 18 large events, or 26 total. Expected captured loss is $9.08 million. The ratio of those expected quantities is approximately $349,231. This is about 3.23 times the underlying mean severity. Large losses account for most of the measured value even though many small incidents are missing. A public database can therefore simultaneously undercount incidents and overstate the mean cost of all incidents.

The direction of selection is not universally upward. A reporting dataset based on data protection notices can omit costly interruption cases that do not involve personal data. Another public source can capture material financial disruption but omit a large number of small insured response expenses. In each case, the selection reflects the purpose of the reporting regime.

The operational risk study by Eling and Wirfs (2019) describes 1,579 cyber incidents in an operational risk dataset and an emphasis on observed event costs. Its accessible abstract supports the distinction between cost-based and record-count-based analysis; its full empirical methods were not independently audited here. The historical data examined by Romanosky demonstrate a separate problem of incomplete cost fields. These documented studies provide motivation for distinguishing event availability from financial measurement, while the numerical example above remains hypothetical.

A researcher should record explicitly whether a cost field represents an amount paid, an estimate, an accounting measure, a claim reserve, a court award, a demand, or a financial impact forecast. Currency, valuation period, gross versus net basis, and uncertainty are essential metadata. Converting heterogeneous monetary observations into a common currency does not make their underlying financial definitions equivalent. Missing values should be retained as missing and modeled only with stated assumptions.

Translating economic loss into covered loss

For a simplified per-event insurance contract with a deductible d and indemnity limit l, let covered gross loss be C and indemnity be Y = min(max(C − d, 0), l). This expression assumes a single deductible and limit, no additional exclusions, no coinsurance, no waiting period and no other applicable conditions. The contract actualities dominate the estimate; the formula is an illustration of familiar insurance mechanics.

Assume a firm-year incident probability of 4%, with conditional covered gross severity equal to $250,000 in 80% of incidents and $2 million in 20%. Let the deductible be $50,000 and limit $1 million. The two indemnities are $200,000 and $1 million. Conditional mean indemnity is $360,000, and annual expected indemnity is $14,400. Gross conditional covered loss before deductible and limit averages $600,000, or $24,000 annually. A premium reflecting the indemnity need not equal either figure; operating costs, uncertainty, capital, reinsurance, expenses, profit, and market conditions may also matter.

Suppose the public source captures 20% of smaller incidents and all larger incidents. Of the observed incidents, the expected larger-event share is 0.20 / (0.80 × 0.20 + 0.20), approximately 55.56%. The observed event probability is 1.44% rather than 4%. Multiplying the observed frequency by the observed mean indemnity yields $9,280 per exposure year, which is below the true $14,400 in the constructed example. Although the public severity distribution is skewed upward, lost smaller claims lead to underestimation of annual expected indemnity. The separate biases do not reliably cancel.

There is a further bias when the information pertains to gross economic damage rather than covered loss. A securities issuer's impact on earnings may include opportunity costs, reputational effects, or disruption outside the coverage period. An insurer may indemnify only explicitly defined categories after deductibles, waiting periods, sublimits, and proof conditions. Some legal expenses may fall under different coverage grants from interruption loss. Limits may apply across related claims or policy periods. The financial amount in a public report must retain its original definition until a defensible mapping to insured costs exists.

Bayesian interpretation of a historical report

A recent event can be a useful signal even when its selection is imperfect. The following hypothetical calculation illustrates why the same signal can have different meaning under different observation conditions. Suppose 20% of applicants have a persistent elevated-risk state H, and the remainder have ordinary risk L. During a prior observation period, event probabilities are 30% in H and 5% in L. Assume initially that reporting is equally likely conditional on an event, at 50% for both groups, and reports have no false positives.

The probabilities of a report are 15% in H and 2.5% in L. Among reported applicants the posterior probability of H is (0.20 × 0.15) / (0.20 × 0.15 + 0.80 × 0.025) = 60%. Among applicants without a report it is (0.20 × 0.85) / (0.20 × 0.85 + 0.80 × 0.975), approximately 17.89%. Silence reduces the hypothetical elevated-risk probability only moderately because many events are unreported.

If high-quality monitoring makes ordinary-risk organizations more likely to report, the interpretation changes. With conditional reporting probability 20% in H and 90% in L, the probability of seeing a public report is 6% in H and 4.5% in L. The posterior probability of H given a report becomes 25%. The report remains potentially valuable but has much weaker discriminatory power than in the first setting.

The numbers are not estimates of insured firms; they isolate the consequence of nonuniform observation. The elevated-risk state is only a convenient hypothetical category. A real pricing study must estimate or validate its predictors, address confounding, separate observation and outcome periods, and consider remediation. Any use of prior public records in risk classification requires awareness that transparent organizations may be overrepresented.

Longitudinal and predictive research design

A public filing corpus and insurer decisions can support a predictive study if matched using entity identity and decision time. At a minimum the analyst needs the risk's underwriting date, coverage and term, source availability date, event dates or intervals, policy exposure, and claims observed over a consistent outcome window. Later case development may be used to construct mature outcome labels but not to construct features for an earlier decision.

An appropriate comparison is the incremental usefulness of incident information over a baseline using lawful and relevant exposure attributes. Separate validation cohorts by calendar time, avoid leakage across sister entities, and hold cluster membership fixed when testing generalization. An apparent improvement in a random train-test split can disappear when truly new events or providers arise, because multiple disclosures of one root event appear in both training and testing sets.

Predictive discrimination alone does not establish fair or effective pricing. Calibration across relevant segments and periods matters because an error in expected loss can affect rates and capacity. A report may predict an outcome precisely in one segment only because that segment is subject to a specific legal obligation. Transfer to unlisted entities or another jurisdiction can fail. Researchers should document feature availability, missingness, legal permissions, changes in reporting rules, and model drift.

A causal study is a separate research undertaking. An observed association between reporting and premiums might reflect more severe events, insurer market cycles, applicant size, improved security, policy term changes, or the decision to seek insurance. An analysis of the introduction of a mandate should examine comparable affected and unaffected groups, prior trends, differences in implementation timing, and changes to reporting probabilities. Even then, claims and coverage may change at the same time. Without credible identification and decision-level data, it is unjustified to infer that mandatory public reporting caused a particular premium change.

Integrated Scientific Synthesis and Research Agenda

Integrated interpretation of the evidence

The previous chapters establish that public reporting has three separable underwriting roles: identifying events and dependencies that merit investigation; corroborating, qualifying, or challenging information obtained privately; and supplying observable records for statistical research subject to selection and timing limitations. These are practical analytical uses of existing institutional and actuarial methods, rather than a claim to a novel general theory.

The central problem is one of correspondence. A source must correspond to the legal entity, relevant period, event category, and financial question. Without that correspondence, greater volume of public records yields an apparent precision detached from the insurance object. A large dataset can be consistently biased if its observations originate mainly from organizations subject to a particular mandate or from a narrow class of severe outcomes.

For a specific underwriting decision, an incident record gains usefulness when it is connected to a current mechanism of exposure. A prior attack that led to verified changes in authentication, containment, recovery, or supplier arrangements may represent a different prospective risk from an unresolved incident. The underwriter can use the historical record to ask about the continuing mechanism rather than simply count adverse events. The analysis should distinguish measures reported by the applicant from those verified by independent or technical evidence.

For portfolio research, incident records gain value from correct grouping. Reporting systems should preserve the root event, impacted entities, source documents, exposure associations, and claim observations separately. One event may have multiple notices, one organization may have multiple events, and a claim may involve costs from multiple circumstances. A many-to-many data relationship is often more faithful to the evidence than a flat row for each announcement.

Minimum evidence for comparison

A research dataset suitable for investigation of public reporting and underwriting needs several categories of metadata. These are ordinary documentary and statistical requirements arising from the research questions, not a newly proposed standard.

Dimension Essential distinctions
Source provenance Author or reporting body; exact document; version and date; source audience
Incident identity Root event; affected entity; type; onset, discovery and uncertainty
Observation Reporting regime; threshold; first notice; public availability; amendments
Exposure Entity and policy identities; underwriting and coverage periods; risk denominator
Financial measurement Currency, valuation date, economic/covered/paid basis, estimate versus realized value
Contract response Applicable grant, conditions, retention, limit, relevant related-claim treatment
Dependency Provider or service identity, reliance, alternatives, linked events, uncertainty
Governance Lawful use, access constraints, source corrections, human review and contestability

The evidence ledger should retain divergent observations when sources differ in timing or scope. Replacing the earlier value with the latest may be appropriate for a current estimate, but not for historical reproduction of a decision. A versioned dataset should support both uses. For quantitative research, every derived number should state which version and denominator it uses.

The source-to-claim distinction is equally important in machine-assisted extraction. A filing can be automatically parsed to identify dates and quantities, but a sentence containing an anticipated insurance recovery must preserve its prospective status. A statement that a company experienced no material operational impact should not be mapped to zero financial cost. An insurer payment made for litigation about two incidents cannot be automatically allocated to the larger or more recent event. Review should distinguish explicit evidence from inference.

Reproducible methods and their limits

The illustrative calculations in this study show four scientific risks: conflation of reporting and incidence, severity selection, mapping from gross costs to indemnities, and failure to model dependence. Each is reproducible through elementary arithmetic or standard probability identities. They establish consequences of specified assumptions only.

A credible empirical extension would acquire a frame of organizations over time, link public notices to legal identities, classify affected operations, and estimate ascertainment with an independent or partially independent source. In a prospective study, using at least two information channels can increase coverage, but they may not be conditionally independent. Capture-recapture estimators can be misleading when organizations appearing in one source are more likely to appear in another, when the definition of a case changes, or when the population is open. The analyst would need to test those assumptions or report sensitivity analyses.

Insurance outcome research should separately track underwriting decisions, terms offered, policy acceptance and mature claims. Without unsuccessful applications or declinations, selection by insurer acceptance cannot easily be distinguished from applicant demand. Without coverage and limit data, changes in incurred loss per policy may reflect changes in protection rather than operational risk. Without time-consistent exposure information, productivity and market growth may be mistaken for changes in event frequency.

A study of disclosure laws would ideally identify implementation dates and applicability at the entity level, retain the original publication and amendment dates, and evaluate changes in both observed events and decision processes. A public event series spanning a new reporting law should be annotated as a series with a potentially changing observation process. The study must not treat its break as evidence of changed attack incidence without corroborating information.

Inferences about underwriting fairness require data about similarly situated risks and the consequences of different classification procedures. Legal fairness constraints depend on the jurisdiction and product. At a scientific level, differences in observation probabilities can be studied by comparing coverage of sources, rates of missingness, error distributions, and model calibration across relevant segments. Exclusions from publicly observable populations, such as small or unlisted firms, must be described rather than interpreted as an absence of cyber risk.

Research agenda and conclusions for institutional analysis

Three lines of empirical investigation follow. First, research should measure the availability and lag distributions of particular reporting channels. A matched source corpus could distinguish event development from dissemination delay and examine whether changes in legal reporting regimes alter the observed composition of events. This work requires stable entity identifiers, known implementation dates and carefully bounded populations.

Second, studies should evaluate whether event history improves prospective loss prediction conditional on current controls, firm characteristics, policy design, and portfolio selection. The incremental contribution of public information should be compared with private application data and other lawful sources. Estimates should report calibration and out-of-time validation, alongside the practical availability of records at the decision date.

Third, research should examine common dependencies across insured populations. Public disclosures can generate candidate links, but actual exposure data and policy terms are required to test the effects on aggregate loss. An empirical study should address ambiguity in root-event attribution and varying degrees of operational dependence, instead of equating every affected organization with a separate independent attack.

The scientific conclusions of this monograph are grounded in accessible official texts, documented corporate case histories, selected empirical literature, and transparent examples. A fully representative international database of public and nonpublic incidents linked to actual insurance contracts was not available. Certain legal source pages yielded indexed operative passages but restricted direct retrieval, and journal evidence explicitly described as abstract-level remains limited to that scope. These limitations constrain generalization and motivate future empirical work. They do not change the demonstrated arithmetic or authorize an assertion of current universal effects on premium levels.

Conclusion

Public cyber incident reporting is part of the infrastructure of underwriting knowledge. Its significance derives from the information it supplies about affected organizations, operations, financial consequences, and common dependencies, conditioned by the reporting process through which those facts become observable. The information should retain its original audience, legal threshold, documentary identity, timing, and financial meaning when incorporated into insurance analysis.

The comparative study establishes that investor disclosure, privacy notification, operational reporting, ransom payment reporting, and insurance notice have different institutional objects. Their timing requirements cannot be treated as a universal event reporting clock. Early notices and amended reports describe stages in the development of an event; an insurer's decision must be reconstructed from the information available at the relevant time.

The probability and severity of insured losses require an exposure denominator, a defined observation mechanism, and the actual contract response. Hypothetical calculations establish that unequal observation probabilities can reverse comparisons of risk, inflate observed mean severity, and alter expected indemnity. Correlated technology dependencies can dramatically change aggregate risk without changing the portfolio's marginal expected loss. These findings explain why transparent records and disciplined actuarial interpretation belong together.

The documented cases reinforce the boundary between corporate economic information and claim evidence. Estimated earnings effects, response costs, financing assistance, legal settlements, and actual insurer payments have distinct meanings. No public corporate disclosure analyzed here establishes a complete single-event insurance valuation by itself.

The principal contribution is the integrated application of established documentary, legal, actuarial, and underwriting methods to the problem of publicly observable incident histories. Public reporting can improve risk assessment and portfolio awareness when its evidence is matched to the insured entity, coverage, timing, and mechanism of loss. An insurer or researcher who preserves these distinctions gains a more reliable basis for analyzing future liabilities, while reducing the risk of penalizing disclosure merely because it makes events visible. The actual market-wide pricing effect of reporting remains an empirical question requiring appropriate decision and claims data.

References

  • U.S. Securities and Exchange Commission. (2023). Cybersecurity risk management, strategy, governance, and incident disclosure. Final rule, Release Nos. 33-11216; 34-97989; 88 Fed. Reg. 51896 (August 4, 2023). SEC cybersecurity disclosure rule.

  • Gerding, E. (2024, June 20). Selective disclosure of information regarding cybersecurity incidents. U.S. Securities and Exchange Commission, Division of Corporation Finance. Selective disclosure guidance.

  • European Parliament & Council of the European Union. (2022). Directive (EU) 2022/2555 of 14 December 2022 on measures for a high common level of cybersecurity across the Union (NIS 2 Directive). OJ L 333, 27 December 2022, 80–152. NIS 2 Directive.

  • European Parliament & Council of the European Union. (2016). Regulation (EU) 2016/679 of 27 April 2016 (General Data Protection Regulation). OJ L 119, 4 May 2016, 1–88. General Data Protection Regulation.

  • European Parliament & Council of the European Union. (2022). Regulation (EU) 2022/2554 of 14 December 2022 on digital operational resilience for the financial sector. OJ L 333, 27 December 2022, 1–79. Digital Operational Resilience Act.

  • European Commission. (2025). Commission Delegated Regulation (EU) 2025/301 of 23 October 2024 supplementing Regulation (EU) 2022/2554 with regard to the content and time limits for reports on major ICT-related incidents and notifications of significant cyber threats. OJ L, 2025/301, 20 February 2025. DORA incident reporting time limits.

  • U.S. Department of Health and Human Services, Office for Civil Rights. (n.d.). Breach notification rule. Guidance on 45 C.F.R. §§ 164.400–414. HIPAA Breach Notification Rule.

  • Parliament of Canada. (2000, as amended). Personal Information Protection and Electronic Documents Act, S.C. 2000, c. 5, ss. 10.1–10.3. Canadian PIPEDA.

  • Parliament of Australia. (2024). Cyber Security Act 2024, No. 98, 2024, Part 3, ss. 26–30. As made 29 November 2024. Australian Cyber Security Act.

  • Australian Department of Home Affairs. (n.d.). Cyber Security Act. Implementation information and links to instruments. Cyber Security Act implementation.

  • United States Congress. (2022). Cyber Incident Reporting for Critical Infrastructure Act of 2022, codified at 6 U.S.C. § 681b. 2024 edition of the United States Code. CIRCIA statutory text.

  • Office of Information and Regulatory Affairs. (2026, November 18). Executive order submissions under review: Department of Homeland Security, RIN 1670-AA04. OIRA rule review record.

  • American International Group, Netherlands. (n.d.). Cyber Insurance Application Form. English-language questionnaire, Prior Claims, Circumstances & Warranties Section, p. 8. AIG cyber insurance application.

  • Romanosky, S., Ablon, L., Kuehn, A., & Jones, T. (2019). Content analysis of cyber insurance policies: how do carriers price cyber risk? Journal of Cybersecurity, 5(1), tyz002. DOI: 10.1093/cybsec/tyz002. Cyber insurance policy content analysis.

  • Romanosky, S. (2016). Examining the costs and causes of cyber incidents. Journal of Cybersecurity, 2(2), 121–135. DOI: 10.1093/cybsec/tyw001. Costs and causes of cyber incidents.

  • Eling, M., & Wirfs, J. (2019). What are the actual costs of cyber risk events? European Journal of Operational Research, 272(3), 1109–1119. DOI: 10.1016/j.ejor.2018.07.021. Actual costs of cyber risk events.

  • U.S. Government Accountability Office. (2021). Cyber insurance: Insurers and policyholders face challenges in an evolving market (GAO-21-477). GAO cyber insurance market report.

  • National Association of Insurance Commissioners. (2026). Report on the Cybersecurity Insurance Market. Calendar-year 2025 data. NAIC. NAIC cybersecurity insurance market report.

  • Pascoe, C., Quinn, S., & Scarfone, K. (2024). The NIST Cybersecurity Framework (CSF) 2.0. NIST Cybersecurity White Paper 29. DOI: 10.6028/NIST.CSWP.29. NIST Cybersecurity Framework 2.0.

  • Microsoft Corporation. (2024, March 8). Current Report, Form 8-K/A, Item 1.05. SEC accession 0001193125-24-062997. Microsoft amended Form 8-K.

  • MGM Resorts International. (2023, October 5). Current Report, Form 8-K. SEC accession 0001193125-23-251667. MGM October 2023 Form 8-K.

  • MGM Resorts International. (2026). Annual Report on Form 10-K for the year ended December 31, 2025. SEC accession 0000789570-26-000018, Note 12. MGM 2025 Form 10-K.

  • UnitedHealth Group Incorporated. (2024). Quarterly Report on Form 10-Q for the quarter ended September 30, 2024. SEC accession 0000731766-24-000323, Item 2. UnitedHealth Form 10-Q.

  • U.S. Department of Health and Human Services, Office for Civil Rights. (n.d.). Change Healthcare cybersecurity incident frequently asked questions. Change Healthcare incident FAQs.

  • DaVita Inc. (2024). Quarterly Report on Form 10-Q for quarter ended September 30, 2024. SEC accession 0000927066-24-000179. DaVita Form 10-Q.