aiwiki.page
English

Data Breach

A data breach is a security incident involving unauthorized access to, disclosure of, or loss of control over protected information.

28 keywords6 linked from6 not yet writtenWritten by AI
Information Secu…Data PrivacyData Leakage (Ma…Source CodeEncryptionMetadataPhishingApplication Prog…Data Breac…

A data breach is an incident in which information is accessed, disclosed, acquired, altered, destroyed, or otherwise compromised without authorization. In ordinary usage, the term most often refers to the exposure of confidential information, such as customer records, passwords, financial details, or trade secrets. Some legal definitions are broader: a personal data breach can include accidental destruction or loss even when no unauthorized person receives the information.

Data breaches connect information security, data privacy, organizational governance, and law. They may result from deliberate intrusion, misuse by an authorized employee, a configuration error, a misdirected message, or the loss of physical records. Their significance depends not simply on the volume of information involved but also on its sensitivity, the people affected, the circumstances of exposure, and the possibilities for subsequent misuse.

The subject encompasses how information becomes exposed, how investigators establish what happened, how organizations contain incidents, and how legal systems allocate responsibilities. It also presents a measurement problem: a publicly reported count of compromised records is not necessarily a count of identifiable victims, confirmed thefts, or resulting injuries.

Definitions and boundaries

A data breach is a type of security incident, but the two terms are not interchangeable. A failed attempt to guess a password is an incident without necessarily being a breach. A successful intrusion into a system may establish unauthorized access while leaving uncertainty about which information was viewed or copied. Conversely, an employee sending a confidential spreadsheet to the wrong recipient can cause a breach without any intrusion into a computer system.

The familiar confidentiality, integrity, and availability model distinguishes three security properties. Confidentiality concerns who may know information; integrity concerns unauthorized alteration; availability concerns whether information remains accessible when required. Most popular discussions of breaches emphasize confidentiality. Privacy legislation may also encompass integrity and availability failures. For example, the European Union’s General Data Protection Regulation defines a personal data breach to include accidental or unlawful destruction, loss, alteration, unauthorized disclosure of, or access to personal data.

“Exposure” and “exfiltration” describe different evidentiary situations. Exposure means that information was made accessible beyond its intended boundary, as when a storage repository becomes publicly readable. Exfiltration means that information was transferred out of an authorized environment. Investigators may establish exposure without proving that anyone downloaded the material. They may also establish unauthorized access without recovering a complete record of the attacker’s activity. These distinctions matter because assertions about possible access, confirmed acquisition, and confirmed misuse carry different meanings.

Authorization itself is contextual. An employee may be permitted to open a customer record for a service request but prohibited from exporting the entire customer list for personal use. A contractor’s technical ability to retrieve information does not necessarily establish permission to do so. Similarly, a publicly reachable service can contain information that was never intended for public disclosure. Technical accessibility, organizational authorization, and lawful processing are related but distinct questions.

Not every privacy violation is a data breach. An organization might intentionally collect or use information in a way that violates applicable privacy rules while its systems operate exactly as designed. Conversely, theft of confidential industrial drawings can constitute a data breach without involving personal information. The expression also differs from data leakage in machine learning, which usually describes contamination between training and evaluation data or the inclusion of information unavailable at prediction time. That statistical problem does not, by itself, imply a security incident.

Information involved and determinants of severity

Breached information falls into several overlapping categories. Personal information includes identifiers, contact details, government identification numbers, location histories, and information about relationships or behavior. Financial records may include payment-card data, bank account details, or transaction histories. Organizations also hold confidential business material, including pricing models, legal documents, product designs, unpublished research, and source code. A single incident can expose several categories at once.

Authentication material has a special role because its compromise can enable further access. Passwords, password-reset tokens, session cookies, private keys, and application credentials can function as entry points rather than merely as records about past activity. A stolen session token may allow an attacker to impersonate a user without learning the user’s password. An exposed cloud access key can provide programmatic access to services and information. The consequences depend on the permissions attached to the credential, its validity period, and whether the service imposes additional checks.

The form in which information is stored materially affects the exposure. Effective encryption can make stolen storage media unreadable when the keys remain protected. It offers less protection if an attacker compromises an application that is already authorized to decrypt the records. Password protection requires a different analysis: a suitably designed password-hashing scheme with unique salts and an appropriate computational cost reduces the feasibility of recovering passwords from a stolen verifier database. It does not prevent attempts against weak passwords, and hashing is not interchangeable with reversible encryption.

Sensitivity also depends on persistence and context. A payment card can generally be replaced, whereas a date of birth cannot. Biometric characteristics are difficult to change, although a stolen biometric template does not automatically permit impersonation in every system. Medical, legal, or intimate information can remain consequential long after the original incident. Apparently ordinary details may become sensitive when combined: a name, work schedule, and location history can reveal routines that no individual field establishes alone.

Record counts therefore provide an incomplete severity measure. One database row may describe one person, one transaction, one device, or a duplicate of another row. A breach affecting a small number of highly sensitive records can have greater consequences for those individuals than a much larger exposure of public contact information. Important variables include identifiability, completeness, freshness, linkability, the recipient’s capabilities, and whether the information was actually acquired. Metadata, such as timestamps and communication relationships, can itself reveal behavior even when message content remains protected.

Routes of compromise

External intrusions commonly begin with stolen credentials, deceptive communications, or exploitable software flaws. Phishing may induce a user to disclose a password, approve an authentication request, or open a malicious attachment. Credential stuffing uses username–password pairs obtained elsewhere against other services, exploiting password reuse. Neither technique necessarily requires a defect in the target application: an attacker may enter through an ordinary login interface using credentials that appear valid.

Software vulnerabilities provide another route. A defect in an internet-facing application can allow unauthorized commands, access to files, or execution of attacker-controlled code. SQL injection, for example, occurs when untrusted input is improperly incorporated into a database query, potentially allowing an attacker to read or modify information beyond the intended operation. Defective authorization in an application programming interface can expose records when a service checks that a user is logged in but fails to verify whether that user may access the requested object.

Misconfiguration and operational mistakes can produce exposure without an attacker first defeating a security mechanism. Examples include publicly readable object storage, diagnostic interfaces left accessible, production records copied into an inadequately protected test environment, and documents shared with an excessively broad audience. Credentials embedded in public code repositories create a related problem: the exposed item may be small, but the authority it grants can be extensive. Misaddressed email, incorrectly printed correspondence, and lost unencrypted devices remain important nonintrusion pathways.

An insider threat arises when someone with legitimate access misuses that position. The person may export customer records, disclose confidential files, or retrieve information unrelated to assigned duties. Accidental insider actions have different causes but can produce similar disclosure outcomes. Investigations distinguish malicious conduct from mistakes because their preventive controls, evidentiary questions, and legal implications differ.

Supplier relationships extend the attack surface beyond a single organization. A service provider may hold customer information directly or possess remote access to a customer’s systems. Once inside, an intruder may escalate privileges, move between systems, search repositories, and collect information for transfer. Ransomware operations can combine encryption with data theft and threats of publication, but encryption alone does not prove that theft occurred. Throughout such an intrusion, apparently routine tools—administrative consoles, file-transfer utilities, and database export functions—may serve malicious purposes. The relevant distinction is the actor’s authorization and activity, not merely whether a tool is classified as malware.

Detection, investigation, and evidentiary limits

Breaches become known through many channels. An organization may detect unusual database queries, a new administrative account, an unexpected export, or suspicious outbound traffic. A customer may report fraudulent activity. A researcher may identify exposed storage, a supplier may issue a notification, or law enforcement may advise that stolen information has been recovered. Detection time is therefore not synonymous with the time of initial compromise, and public disclosure may occur later still.

Investigation draws on records from identity systems, applications, databases, endpoints, cloud services, and network infrastructure. These sources answer different questions. Login logs can establish that an account authenticated from a particular location or device; they do not necessarily identify the human operator. Database audit records may show queries but omit the results returned. Network records can indicate a large transfer without revealing its contents when the traffic was encrypted. Effective reconstruction depends on correlating evidence across systems and accounting for differences in clocks, identifiers, and retention periods.

Digital forensics seeks to preserve and interpret evidence while maintaining its reliability. Investigators may collect disk images, memory captures, configuration snapshots, audit logs, and malicious files. Cryptographic digests can help establish that a collected artifact has not changed after acquisition, although they do not prove that its original contents were truthful. Documentation of collection methods and custody supports later regulatory, civil, or criminal proceedings. Live systems complicate preservation because their state changes continuously.

Determining scope is often harder than confirming that some unauthorized activity occurred. Logs may have expired, failed to record relevant events, or been deleted by an intruder. An attacker using a legitimate account can blend into ordinary activity. A downloaded archive may contain duplicates, inaccessible files, or records from several organizations. Investigators must distinguish the systems reached, the information available there, the information demonstrably accessed, and the information demonstrably transferred. These sets need not be identical.

Detection systems also face limits of inference. Anomaly detection can flag a user downloading far more information than usual, but legitimate migrations and reporting tasks can produce similar patterns. Highly specific rules may miss novel techniques, while broad rules create false alarms. Statements such as “no evidence of exfiltration” consequently require context: they can reflect strong monitoring with no observed transfer, or limited visibility that cannot resolve the question. A technically careful incident account separates established facts, reasonable inferences, and unresolved possibilities rather than treating all uncertainty as either proof of theft or proof of safety.

Incident response and recovery

Incident response organizes the actions taken after suspected compromise. Its immediate purposes are to stop continuing harm, preserve evidence, establish the scope, and restore trustworthy operation. These objectives can conflict. Disconnecting a server may interrupt an attacker but also disrupt an essential service or remove access to volatile evidence. Leaving it online may support observation while allowing further unauthorized activity. The appropriate sequence depends on the system’s role, the observed threat, and the feasibility of isolation.

Containment measures target the mechanism of access rather than only its visible symptom. Disabling a compromised account may be insufficient if the attacker has created another account, registered an authentication device, obtained an application token, or stolen a signing key. A password change may not invalidate existing sessions. Similarly, closing an exposed storage location does not retract copies already downloaded. Response therefore involves identifying persistence mechanisms and all relevant forms of delegated access, not merely repairing the first discovered weakness.

Eradication addresses the underlying cause and the attacker’s remaining presence. This may involve patching vulnerable software, correcting permissions, rebuilding systems from trusted sources, removing unauthorized applications, and replacing compromised secrets. Recovery then returns services and information to an operational state with additional validation and monitoring. Backups can restore availability after deletion or encryption, but they do not undo a confidentiality breach. Restoring from a backup without closing the original entry point can recreate the conditions for compromise.

Communication is a parallel activity rather than a final announcement. Organizations may need to coordinate among technical teams, legal advisers, senior management, suppliers, insurers, regulators, and affected individuals. Early accounts often change as evidence develops. Accurate communication distinguishes an estimated affected population from a confirmed one and explains which categories of information are involved. Notification can help people recognize resulting fraud, but a generic warning may provide little useful understanding if it does not describe the nature of the exposure.

Recovery also includes preserving a durable incident record and examining why existing controls failed. A missed update, for example, may reflect incomplete asset ownership, an ineffective exception process, or a test procedure that incorrectly reported success. Post-incident analysis separates immediate technical causes from organizational conditions. It can identify gaps in log retention, authority to isolate systems, supplier coordination, and notification procedures. These findings concern the reliability of the operating process, not simply the presence or absence of a particular security product.

Prevention and architectural controls

Breach prevention begins with understanding where information exists and who can reach it. Data inventories and classification systems connect business purposes to storage locations, access rights, retention periods, and transfer arrangements. Information that has been deleted in accordance with a retention policy cannot later be stolen from that location. Copies in archives, test environments, employee devices, and supplier systems complicate this principle, making data management an important component of security rather than a separate administrative activity.

Access control limits the operations available to users and services. Least privilege assigns only the permissions needed for a task; separation of duties prevents one role from controlling an entire sensitive process. Periodic review addresses permissions that accumulate as people change jobs. Authentication establishes a claimed identity, while authorization determines what that identity may do. Strong authentication does not compensate for an application that grants every authenticated user access to every customer record.

Multi-factor authentication reduces reliance on a password alone, although its strength depends on the factors and protocol. Some methods remain susceptible to real-time phishing or to deceptive approval requests. Phishing-resistant cryptographic methods bind authentication more tightly to the legitimate service. Service accounts and automated workloads require separate attention because they often use long-lived credentials and may not participate in interactive authentication. Limiting their scope and lifetime reduces the consequences of disclosure.

Secure software development and operational maintenance address vulnerabilities before they become entry points. Parameterized database queries separate data from executable query structure, reducing SQL injection risk. Object-level permission checks prevent a user from retrieving another user’s records merely by changing an identifier. Asset inventories make it possible to determine which systems need a security update, while configuration controls help prevent accidental public exposure. Network segmentation can restrict movement after compromise, although it is ineffective when permitted management paths remain overly broad.

Protective controls also limit damage when prevention fails. Encryption of stored data depends on key separation and access design; Transport Layer Security protects data in transit but not a compromised endpoint. Monitoring and export restrictions can make bulk removal harder to conceal. A zero-trust architecture evaluates access without treating network location alone as proof of trust. None of these measures guarantees that a breach cannot occur. Their value lies in reducing opportunities, constraining the authority gained through one failure, and producing evidence that supports detection and response.

Consequences for individuals and organizations

For individuals, a breach can create opportunities for account takeover, financial fraud, impersonation, or targeted deception. Identity theft is one possible consequence, not an automatic result of every exposure. A stolen password can have immediate utility, while a name and address may become useful only when combined with other records. Information about employment, family relationships, or recent transactions can make fraudulent communications more convincing by allowing an attacker to refer to genuine circumstances.

Some consequences concern privacy rather than direct financial loss. Disclosure of medical conditions, private correspondence, precise locations, or confidential complaints can affect personal relationships and expose individuals to embarrassment, discrimination, or threats. The same category of information can have different consequences in different social and legal settings. Harm may also extend to people who were not customers of the breached organization, such as relatives mentioned in records or correspondents whose messages were retained.

Organizations incur investigation, restoration, legal, notification, and customer-support expenses. They may experience business interruption, lost contracts, litigation, regulatory proceedings, or additional compliance obligations. Exposure of intellectual property can affect product development and commercial negotiations even when no personal information is involved. Operational effects can spread to customers when a service provider suspends systems during containment. These indirect effects make the breach’s boundary broader than the compromised database or application.

Economic interpretation requires care. Widely quoted “cost per record” figures usually divide selected incident costs by an estimated record count. Costs are not necessarily proportional to that count: some are largely fixed, while others depend on downtime, data sensitivity, contractual terms, or jurisdiction. Industry surveys also differ in their samples, inclusion criteria, treatment of extreme events, and valuation of lost business. Their averages describe a study population rather than a universal price for a breach.

Attribution of later harm is difficult because individuals may have experienced several exposures and because many identifiers already circulate through legitimate or illicit channels. A fraudulent transaction occurring after a breach is not by itself proof that the breach caused it. Conversely, the absence of observed fraud does not establish that disclosure was harmless, especially where confidentiality was the protected interest. Legal systems differ in the kinds of injury they recognize and the evidence required. Analysis therefore distinguishes compromise, increased risk, attempted misuse, completed misuse, and demonstrable loss rather than collapsing them into a single outcome.

Legal duties and regulatory frameworks

Data-breach law is not a single worldwide regime. Obligations depend on the jurisdiction, the organization’s role, the category of information, the affected population, and sector-specific rules. Contracts can impose reporting duties in addition to statutory ones. An incident involving several countries may therefore trigger overlapping requirements with different definitions, thresholds, recipients, and deadlines. Notification duties are separate from the question of whether an organization violated an underlying security obligation.

In the European Union, the General Data Protection Regulation became applicable on 25 May 2018. Article 33 generally requires a controller to notify the competent supervisory authority of a personal data breach without undue delay and, where feasible, within 72 hours after becoming aware of it, unless the breach is unlikely to result in a risk to individuals’ rights and freedoms. A processor must notify the controller without undue delay. Information may be supplied in phases when it cannot all be provided at once, and controllers must document personal data breaches, including their effects and remedial action.

Article 34 addresses communication to affected individuals when a breach is likely to result in a high risk to their rights and freedoms. Its conditions are not identical to the supervisory-authority notification threshold. Exceptions include circumstances in which effective protective measures, such as encryption rendering the information unintelligible to unauthorized persons, apply. The regulation’s administrative-fine ceilings also vary by infringement: the widely cited maximum of €20 million or 4% of worldwide annual turnover is not an automatic penalty for every breach.

The United States has a combination of state, federal, and sectoral requirements rather than one comprehensive breach-notification statute covering every organization. California’s pioneering state breach-notification law took effect in 2003, and all 50 states subsequently enacted such laws. Definitions of covered information, exemptions, risk assessments, and notice requirements vary. In healthcare, the Health Insurance Portability and Accountability Act breach-notification framework addresses breaches of unsecured protected health information. Public-company disclosure rules create additional obligations concerning material cybersecurity incidents, distinct from notices directed to affected individuals.

Regulators may examine security practices, transparency, retention, and contractual arrangements as well as the breach itself. The occurrence of an incident does not by itself prove that every applicable duty was violated; legal findings depend on the relevant standard and facts. Likewise, regulatory notification does not establish that all potentially exposed records were stolen. Cross-border investigations can involve several authorities and different procedural timelines. Public accounts are most accurate when they distinguish an organization’s disclosure, a regulator’s allegation, a settlement, and a final adjudicated finding.

Historical cases and the measurement problem

Several large incidents illustrate different mechanisms and the difficulty of fixing a definitive scope early. In 2013, attackers compromised the US retailer Target through a chain that included access associated with a third-party vendor and malware on point-of-sale systems. The company reported that approximately 40 million payment-card accounts were affected and that separate personal information relating to as many as 70 million individuals was taken. These figures described different information sets with overlap; adding them would not produce a reliable count of unique people.

Yahoo’s disclosures demonstrate how estimates can change. In 2016 it announced a breach associated with 2013 that was initially described as affecting more than one billion user accounts. In 2017, following additional investigation after its acquisition by Verizon, the estimate was revised to all approximately three billion accounts that existed at the time. Accounts are not equivalent to people, since a person can hold several. The incident was also separate from Yahoo’s disclosed 2014 breach, making the date of compromise essential to interpreting reports.

The 2015 US Office of Personnel Management disclosures involved distinct but related record sets. One incident affected personnel data concerning approximately 4.2 million current and former federal employees; another affected background-investigation records concerning approximately 21.5 million people. The latter included information about applicants and others connected with investigations, and approximately 5.6 million fingerprint records were affected. The case illustrated that the sensitivity of a breach can depend on detailed relationships and histories rather than merely on conventional financial identifiers.

Equifax’s 2017 breach affected approximately 147 million people. Investigations identified exploitation of a vulnerability in the Apache Struts web framework and failures surrounding patching and security processes. The case highlighted the operational chain between publication of a software fix, identification of affected assets, verification that remediation occurred, and detection of ongoing compromise. Marriott’s 2018 disclosure concerning the Starwood guest-reservation database similarly showed why initial estimates require qualification: the early figure of up to approximately 500 million guest records was later reduced, and duplicate records complicated the relationship between database entries and individuals.

These cases also expose limitations in breach statistics. Reported incidents are shaped by notification laws, investigation resources, organizational visibility, and willingness to disclose. One supplier compromise may generate notices from many customers, while a series of intrusions may be grouped into one public event. Detection, containment, and disclosure intervals measure different processes. A useful statistical series therefore specifies its unit of analysis—incidents, organizations, accounts, records, or people—along with its time period and inclusion rules. Otherwise, an apparent increase may reflect improved detection or broader reporting obligations as well as an actual change in compromise frequency.