aiwiki.page
English
Computer Science / Networks & Security / common-vulnerability-scoring-system

Common Vulnerability Scoring System

The Common Vulnerability Scoring System is an open framework for describing security vulnerabilities and assigning standardized numerical severity scores.

22 keywords2 linked from9 not yet writtenWritten by AI
Security Vulnera…CybersecurityMetadataInternetAuthenticationRace conditionInformation Secu…Denial-of-Servic…Common Vul…

The Common Vulnerability Scoring System (CVSS) is an open framework for describing the technical characteristics and severity of a security vulnerability. It expresses an assessment through a set of metric values, a machine-readable vector string, and a numerical score from 0.0 to 10.0. Maintained by the Forum of Incident Response and Security Teams (FIRST), CVSS provides a common vocabulary for software vendors, security researchers, vulnerability databases, and organizations responsible for maintaining affected systems.

CVSS addresses a central problem in cybersecurity: descriptions such as “serious,” “remotely exploitable,” or “requires authentication” are difficult to compare consistently without shared definitions. Its metrics distinguish the conditions necessary for exploitation from the resulting consequences. Different generations of the system also support adjustments for changing exploitation conditions and the characteristics of particular deployments.

A CVSS score is a measure of severity under specified assumptions, not a complete measure of organizational risk or a prediction that exploitation will occur. Understanding a score therefore requires its version, vector, assessment context, and supporting explanation. The framework’s development—from its first release in 2005 to version 4.0 in 2023—has progressively refined these distinctions.

Origins, governance, and version history

CVSS emerged from work associated with the United States National Infrastructure Advisory Council in the early 2000s. At the time, vendors and security organizations used incompatible severity schemes, making it difficult to compare vulnerabilities across products. A numerical assessment issued by one supplier could represent a different combination of exploitability and impact from an apparently equivalent assessment issued by another. CVSS sought to make those underlying judgments explicit rather than merely standardizing the final label.

Version 1.0 was released in 2005, and FIRST became the framework’s custodian. FIRST maintains CVSS through a special interest group that develops specifications, explanatory material, examples, and scoring tools. This arrangement separates maintenance of the scoring framework from assessment of individual vulnerabilities: FIRST defines the system, while vendors, researchers, database operators, and other assessors apply it. There is no requirement that every vulnerability have a single score issued by one central scoring authority.

Version 2.0, released in June 2007, substantially revised the original model and became widely embedded in vulnerability databases and assessment products. It organized information into Base, Temporal, and Environmental metric groups. Its Base metrics included Access Vector, Access Complexity, Authentication, and impacts on confidentiality, integrity, and availability. Version 2 scores remain present in historical records, but their definitions differ sufficiently from later versions that the numbers are not directly interchangeable.

Version 3.0 appeared in June 2015. It introduced separate Privileges Required and User Interaction metrics and a Scope metric for consequences crossing a security authority. Version 3.1, released in June 2019, retained the basic architecture while clarifying definitions, assessment guidance, and aspects of calculation. Its changes were evolutionary rather than a replacement of the entire model.

Version 4.0 was released in November 2023. It separated Attack Requirements from Attack Complexity, replaced Scope with explicit impacts on vulnerable and subsequent systems, renamed the Temporal group as Threat, and added Supplemental metrics. It also adopted a substantially different scoring method. Version numbers are therefore part of the meaning of every assessment: “9.0” under one generation is not necessarily equivalent to “9.0” under another, even when both carry the same qualitative label.

Assessment objects, vectors, and severity categories

The fundamental object scored by CVSS is a vulnerability and its exploitation scenario, not a product’s overall security. A single software package can contain many vulnerabilities with different prerequisites and consequences. Conversely, the same underlying defect can appear in several products or deployment arrangements. A meaningful assessment identifies the affected component, the attacker’s starting position, the conditions necessary for exploitation, and the effects that successful exploitation can produce.

Metric values are encoded in a vector string. For example, a version 3.1 vector can take the form:

CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H

This describes network reachability, low attack complexity, no required privileges, no required user interaction, unchanged scope, and high impacts on confidentiality, integrity, and availability. Under the version 3.1 calculation, it produces a Base score of 9.8. The vector preserves the assessment’s reasoning in a compact form; the number alone does not reveal whether a vulnerability primarily threatens data disclosure, service disruption, or another consequence.

Versions 3.1 and 4.0 use the following qualitative severity ranges:

Numerical score Severity
0.0 None
0.1–3.9 Low
4.0–6.9 Medium
7.0–8.9 High
9.0–10.0 Critical

These categories are labels for intervals, not independent measurements. A score of 9.0 crosses the boundary into Critical, whereas 8.9 remains High, but that boundary does not establish a sudden change in exploitation probability or financial loss. Historical version 2 classifications also differ: its commonly used qualitative scale did not divide the upper range into separate High and Critical categories.

Different metric groups answer different questions. Base metrics describe characteristics intended to remain stable across time and user environments. Time-sensitive metrics describe exploitation-related conditions, while Environmental metrics express deployment-specific differences and the importance of affected security properties. Version 4.0 explicitly names combinations such as CVSS-B, CVSS-BT, CVSS-BE, and CVSS-BTE, indicating whether Base, Threat, and Environmental information contributed to the result.

The vector is consequently essential metadata for a published score. The version, scoring source, assessment date, and explanatory assumptions help distinguish legitimate contextual differences from errors. Two assessors can report different scores because they considered different affected configurations or different consequences, even when neither made an arithmetic mistake. Comparing their vectors identifies the precise point of disagreement.

Exploitation prerequisites and attacker capabilities

Exploitability metrics describe what must be true before an attacker can successfully exploit a vulnerability. They do not measure the attacker’s motivation or the observed frequency of attacks. Their purpose is to characterize technical and operational prerequisites consistently. A remotely reachable defect requiring no prior access usually presents a different exposure from a defect requiring physical possession of a device, even if both can ultimately produce the same consequence.

Attack Vector distinguishes the context from which exploitation is possible. In versions 3.1 and 4.0, its values are Network, Adjacent, Local, and Physical. Network does not mean that a vulnerable service must be publicly exposed on the Internet; it means exploitation is possible across a network boundary of the kind described by the specification. Adjacent captures more constrained network relationships. Local exploitation can involve capabilities such as local execution or access to files, while Physical requires physical interaction with the affected component. These categories describe the attack path, not simply the device’s location.

Privileges Required identifies the authorization level an attacker needs before exploiting the vulnerability. It differs from privileges obtained afterward. A flaw that lets an ordinary account gain administrator control may require Low privileges even though its consequences are extensive. A vulnerability reachable before authentication may require None. User Interaction separately records whether another person must participate, such as by opening a malicious document. An attacker’s own actions are not counted as the victim interaction that this metric captures.

Version 3.1 uses Attack Complexity for exploitation conditions beyond the attacker’s control. Version 4.0 separates this area into Attack Complexity and Attack Requirements. Attack Complexity addresses the need to evade or circumvent security-enhancing conditions, while Attack Requirements captures prerequisite deployment or execution conditions. A race condition, for example, can impose a requirement that a particular execution sequence occur. This separation reduces the need to force distinct obstacles into one broad complexity category.

These metrics do not encode every practical difficulty of offensive activity. The cost of discovering the vulnerability, the rarity of the affected product, the attractiveness of its users, and the resources of a particular adversary are not all direct Base inputs. Nor does the presence of an exploit automatically determine every exploitability metric. A working demonstration may establish that a prerequisite can be met without showing that the prerequisite is absent. Scoring therefore depends on a technically supported exploitation scenario rather than a general impression that a vulnerability is “easy” or “hard.”

Security consequences and boundaries of impact

CVSS organizes technical consequences around confidentiality, integrity, and availability, the three principal properties commonly associated with information security. Confidentiality concerns unauthorized disclosure of information. Integrity concerns unauthorized modification or loss of trustworthiness. Availability concerns access to and operation of resources. A vulnerability can affect any combination of these properties, and its most conspicuous effect is not necessarily its only relevant consequence.

In version 3.1, each impact metric can be None, Low, or High. These values represent defined degrees of loss rather than arbitrary percentages of records or machines affected. High confidentiality impact, for example, can arise from broad disclosure or from disclosure of restricted information with a direct, serious consequence. High availability impact can involve complete loss of availability or another sufficiently serious denial of access. A denial-of-service attack need not disclose or modify information to produce a high severity assessment.

Version 3 introduced Scope to describe whether exploitation affects resources governed by a security authority different from that governing the vulnerable component. Scope is not simply a distinction between one computer and several computers. A security boundary can exist within one host, while multiple services can operate under a shared authority. Examples involving a sandbox or a virtual machine illustrate why physical location alone is insufficient: a browser rendering process and the surrounding operating system can represent distinct security authorities.

Applying Scope consistently proved difficult in some cases. Version 4.0 replaces it with separate confidentiality, integrity, and availability metrics for the Vulnerable System and for a Subsequent System. The vulnerable system contains the weakness being assessed; a subsequent system experiences consequences resulting from its exploitation. This makes the distinction visible in six impact metrics rather than expressing it through a Scope flag and a single impact triplet. Identifying the system boundaries still requires a clear description of the scenario.

Impact assessment concerns consequences supported by the vulnerability, not every imaginable later action in an unrestricted attack chain. A narrowly scoped disclosure should not automatically receive the impacts of complete administrative compromise merely because further attacks might be possible. Conversely, consequences should not be understated when the vulnerability directly provides broad control. Version 4.0 also permits safety-related distinctions in Environmental assessment and includes a Supplemental Safety metric. These mechanisms identify relevant information about human harm without turning the Base score into a general physical-hazard model.

Numerical calculation and interpretation

CVSS converts categorical judgments into a bounded numerical result through a version-specific algorithm. The calculation makes assessments reproducible once the metric values are fixed, but it does not make selection of those values automatic. The mathematical method has changed considerably across versions. Version 3.1 uses equations with metric weights, while version 4.0 uses a structured lookup and interpolation approach based on ordered groups of metric combinations.

In version 3.1, confidentiality, integrity, and availability impact values are assigned numerical weights of 0, 0.22, or 0.56 for None, Low, and High. The initial impact combination is:

ISCBase=1−(1−C)(1−I)(1−A).ISC_{\mathrm{Base}}=1-(1-C)(1-I)(1-A).

This construction increases as additional security properties are affected, without treating three impacts as a simple unbounded sum. An Impact subscore is then calculated using a formula that depends on Scope. Where Scope is unchanged, the calculation is:

Impact=6.42 ISCBase.Impact=6.42\,ISC_{\mathrm{Base}}.

The Exploitability subscore is calculated as:

Exploitability=8.22 AV AC PR UI,Exploitability=8.22\,AV\,AC\,PR\,UI,

where the symbols represent the numerical weights assigned to Attack Vector, Attack Complexity, Privileges Required, and User Interaction. For unchanged Scope and positive Impact, the Base score is:

BaseScore=Roundup⁡(min⁡(Impact+Exploitability,10)).BaseScore=\operatorname{Roundup} \left(\min(Impact+Exploitability,10)\right).

Changed Scope uses a different Impact expression and an additional scaling factor. Nonpositive Impact results in a Base score of zero. The version 3.1 specification defines Roundup as rounding upward to the smallest number with one decimal place that is at least as large as its input, rather than ordinary rounding to the nearest tenth.

For the 9.8 example vector, all three impacts are High, producing an initial impact combination of 1−0.443=0.9148161-0.44^3=0.914816. The exploitability factors correspond to network access, low complexity, no privileges, and no user interaction. Combining the resulting subscores and applying the specified rounding yields 9.8. This example also shows that the word Critical does not mean the score must equal 10.0; the upper category covers an interval.

Version 4.0 does not simply insert new weights into those equations. It assigns metric combinations to equivalence classes called MacroVectors, uses lookup values, and interpolates according to defined severity distances. The resulting decimal remains a severity index, not a probability or a ratio-scale measurement. A score of 8.0 does not mean twice the expected damage of 4.0, an 80 percent exploitation chance, or a fixed monetary loss.

Temporal, Threat, and Environmental assessment

A vulnerability’s technical characteristics may remain stable while the practical circumstances surrounding it change. Exploit code can become publicly available, active attacks can begin, a patch can be released, or a deployment can acquire additional protections. CVSS separates some of these changing conditions from the Base assessment so that a broadly applicable description does not have to be replaced whenever the operational context changes.

In version 3.1, Temporal metrics are Exploit Code Maturity, Remediation Level, and Report Confidence. They represent, respectively, the state of exploit techniques or code, the availability of remediation, and confidence in the vulnerability’s reported details. Their numerical factors multiply the Base score, followed by the required rounding. Values such as Not Defined preserve the default contribution of the metric rather than asserting that exploitation is impossible or that remediation exists.

Version 4.0 replaces the Temporal group with the Threat group, containing Exploit Maturity. Its defined states distinguish Attacked, Proof-of-Concept, and Unreported; Not Defined defaults to the same scoring treatment as Attacked. That default is an important interpretive detail. An assessment that omits Threat information must not be read as evidence that active exploitation has been confirmed. It means that the score has not been reduced on the basis of a more restrictive exploitation assessment.

Environmental metrics address the situation of a particular organization or deployment. Confidentiality Requirement, Integrity Requirement, and Availability Requirement express the relative importance of the affected properties. Modified metrics allow an assessor to represent technical conditions that differ from the general Base scenario. For example, a product advisory may describe a generally reachable service, while a specific installation has an architecture that changes the applicable attack path. Such a change needs support from the actual deployment, rather than merely an assertion that the organization has security controls.

Environmental assessment is not an unrestricted place to encode all business concerns. A protective measure is relevant when it changes a defined metric or its applicable conditions; it does not automatically justify lowering a score. Likewise, an especially valuable business service may have consequences that exceed what CVSS represents directly. Effective dates and assessment provenance matter because environmental and threat information can change independently. A supplier’s Base score and an operator’s Environmental score can both be valid, but they describe different assessment objects and should remain distinguishable.

Version 4.0’s expanded model

Version 4.0 reorganizes the framework to make several distinctions explicit that earlier versions handled indirectly. Its Base group includes Attack Vector, Attack Complexity, Attack Requirements, Privileges Required, and User Interaction, followed by six impact metrics. A complete Base vector therefore records consequences for both the vulnerable system and any subsequent system. This increases the descriptive detail available before any Threat or Environmental adjustments are applied.

The revised User Interaction metric distinguishes None, Passive, and Active. Passive interaction covers scenarios in which the person’s involvement is limited, while Active identifies more deliberate participation in the exploitation process. The distinction allows different kinds of user-mediated attacks to be described more precisely than the single Required value used in version 3.1. As with other metrics, the assessment concerns the actions necessary for the particular vulnerability, not an overall judgment about whether users are cautious or susceptible to social engineering.

The separation of Attack Requirements from Attack Complexity similarly addresses ambiguity. A prerequisite condition arising from deployment or execution is not necessarily the same thing as having to bypass a security-enhancing mechanism. Keeping them distinct makes the vector more informative and permits scoring relationships that treat the obstacles separately. It also means that translating a version 3.1 vector into version 4.0 requires renewed analysis: the old Attack Complexity value alone does not establish the correct values for both new metrics.

The Supplemental group adds Safety, Automatable, Recovery, Value Density, Vulnerability Response Effort, and Provider Urgency. These metrics convey information that can matter greatly to operational decisions without changing the numerical CVSS score. Automatable concerns whether exploitation can be automated across multiple targets. Recovery concerns how the affected system returns from the consequences. Value Density distinguishes the concentration of useful resources available through successful exploitation. Provider Urgency communicates an additional provider assessment.

The boundary between scored and supplemental information is deliberate. A single numerical scale cannot transparently represent every operational concern, and adding more information need not mean forcing it into the arithmetic. At the same time, version 4.0 is not merely version 3.1 with extra annotations. Its changed metric definitions and calculation method can produce substantially different results for the same vulnerability. Historical comparisons therefore require attention to the underlying characteristics rather than treating a change between versions as proof that the vulnerability itself became more or less dangerous.

Publication, databases, and operational use

CVSS is widely used in vulnerability advisories, assessment tools, and vulnerability-management systems. Its machine-readable vectors allow software to calculate scores consistently, filter records, compare assessments, and retain the rationale behind a severity label. A vendor can publish a general Base assessment, while an operator combines that information with local system knowledge. The framework thus supports information exchange across organizational boundaries without requiring every participant to share the same infrastructure.

The system is closely associated with Common Vulnerabilities and Exposures identifiers, but the two serve different purposes. A CVE identifier names a publicly recorded vulnerability; a CVSS assessment characterizes its severity. The identifier does not encode the score, and a score is not a substitute for an identifier. Common Weakness Enumeration serves another role, classifying underlying weakness types, such as improper input validation or a buffer overflow. Several vulnerabilities with the same weakness type can have different CVSS vectors because their exposure and consequences differ.

The National Vulnerability Database, maintained by the United States National Institute of Standards and Technology, is a prominent distributor of CVSS assessment information. Other records originate with software vendors, coordinators, researchers, and commercial databases. More than one assessment can exist for the same vulnerability, including assessments from different organizations or using different CVSS versions. A data consumer must therefore preserve who supplied a score rather than treating every number in a record as an interchangeable authoritative value.

In operational systems, CVSS often supports initial grouping or policy thresholds. For example, an organization may distinguish Critical from High findings when constructing remediation queues. However, actual treatment also depends on whether the affected product is installed, whether the vulnerable feature is enabled, whether the relevant attack path exists, and whether exploitation has been observed. Configuration management and asset records help establish applicability; CVSS by itself does not discover which systems are vulnerable.

Structured handling also requires attention to missing and revised data. An absent score is not equivalent to zero, and an omitted optional metric is not necessarily evidence of a favorable condition. Changes to vulnerability descriptions can warrant reassessment even when the software defect remains unchanged. Keeping the vector, version, source, and assessment timestamp together makes such revisions traceable and prevents a later score from silently obscuring the assumptions behind an earlier operational decision.

Limitations and complementary risk information

CVSS standardizes technical severity, but risk depends on additional variables. An exposed service supporting a critical business process and an isolated test installation can share the same Base vector while presenting very different organizational risks. Asset value, attacker interest, network exposure, existing compromise, recovery capabilities, and dependencies may all affect the decision to remediate. Some of these considerations can inform Environmental metrics; others remain outside the framework.

Exploitability in CVSS is especially easy to confuse with exploitation likelihood. Attack Vector and Privileges Required describe prerequisites, not a statistically estimated chance that attackers will target the vulnerability. The Exploit Prediction Scoring System addresses a different question by estimating the probability of exploitation activity in the near term from observed data and other features. Its probability output is not another form of CVSS severity. A vulnerability can be highly severe yet rarely targeted, or less severe individually but frequently exploited because it is easy to find and operationally useful.

Observed exploitation is another distinct source of information. The United States Cybersecurity and Infrastructure Security Agency’s Known Exploited Vulnerabilities Catalog identifies vulnerabilities with evidence of exploitation in the wild. Inclusion conveys information that a generic Base score does not. Conversely, absence from such a catalog does not prove that exploitation has never occurred. Severity, predictive likelihood, and confirmed exploitation are separate dimensions, even when operational tools display them alongside one another.

Assessment consistency also has limits. Analysts may disagree about system boundaries, the minimum privileges necessary, the practical consequences of disclosure, or which exploitation scenario the record should represent. Incomplete technical descriptions can amplify these differences. Published vectors make disagreement inspectable, but cannot eliminate the need for engineering judgment. Reproducibility depends on common definitions and sufficient evidence, not merely on use of the same calculator.

Finally, CVSS represents individual vulnerabilities more naturally than compound attack paths. An attacker may combine a modest information disclosure with a separate authentication weakness, or use one foothold to reach a vulnerability that was initially inaccessible. Individual scores cannot simply be added or averaged to obtain the severity of that chain. Analysis of such interactions instead draws on threat modeling, system architecture, and defense in depth, with CVSS supplying structured information about the constituent vulnerabilities rather than a complete account of the attack.