Skip to content

CVSS Explained: Vulnerability Severity Ratings, CVSS 4.0, EPSS and KEV

How CVSS 3.1 and 4.0 score severity, with a worked CVE example, plus EPSS, CISA KEV, SSVC and Bugcrowd's VRT, from FIRST, NVD and CISA sources.

Mark KissBy , Founder & CEOUpdated 18 min read

Penetration Testing ServicesExternal network from $4,200, web app from $5,200Fixed scope, free retest, no hourly billing.

Get my fixed price

This guide explains how the Common Vulnerability Scoring System (CVSS) rates severity in versions 3.1 and 4.0, works through the arithmetic for a real CVE, and covers the signals that answer a different question, whether a flaw will be exploited: FIRST’s EPSS, CISA’s Known Exploited Vulnerabilities (KEV) catalog and SSVC. It also compares Bugcrowd’s bug bounty taxonomy and shows how penetration test reports rate findings. Every figure links to FIRST, NVD, CISA or the publisher’s own documentation.

How to cite: link to this page or to the primary source beside each figure.

Key points at a glance

  • Five bands, same in CVSS 3.1 and 4.0: None 0.0, Low 0.1 to 3.9, Medium 4.0 to 6.9, High 7.0 to 8.9, Critical 9.0 to 10.0 (FIRST CVSS v4.0 Specification).
  • CVSS 4.0 scores by lookup, not formula: FIRST grouped 15 million possible vectors into 270 equivalence sets ranked by experts (FIRST CVSS v4.0 Specification).
  • A base score is not risk: FIRST says CVSS-B scores “should not be used alone for patch prioritization” (FIRST CVSS v4.0 FAQ, 2026).
  • CVSS Critical is about 10% of new CVEs: of roughly 61,000 CVEs published in the 12 months before FIRST’s analysis, just over 10% were rated Critical (FIRST, Using EPSS, 2026).
  • Exploitation is rare: 1.5% to 3% of published vulnerabilities show exploitation activity in any 30-day period (FIRST, Using EPSS, 2026), and KEV lists roughly 0.5% of published CVEs (FIRST EPSS FAQ, 2026).
  • KEV held 1,781 vulnerabilities in catalog version 2026.11.13 (CISA KEV feed, November 2026).
  • Federal fix deadlines now run from 3 days to the next upgrade, set by exposure, KEV status, automation and technical impact under BOD 26-04, issued June 10, 2026 (CISA BOD 26-04).

1. What CVSS is and who maintains it

CVSS is an open framework for describing a vulnerability’s technical characteristics and turning them into a 0 to 10 score. It is owned and managed by FIRST (the Forum of Incident Response and Security Teams), licensed free of charge, and anyone who publishes CVSS data is required to publish the vector string with the score so others can see how it was derived (FIRST CVSS v4.0 Specification).

Product maintainers and the National Vulnerability Database (NVD) “typically provide only the Base Scores”; threat and environmental metrics are left to the organization running the software (FIRST CVSS v4.0 Specification). CVSS does not need a CVE ID, so it also works for internal findings that are never published (FIRST CVSS v4.0 FAQ).

CVSS v4.0 was first published on November 1, 2023; the current specification is version 1.2 of June 18, 2024 (FIRST CVSS v4.0 Specification). Version 3.1 is still common: NVD’s record for the CVE in section 3 carries only a v3.1 vector.

2. CVSS 3.1 vs 4.0: the metric groups

CVSS v3.1 CVSS v4.0
Metric groups Base, Temporal, Environmental Base, Threat, Environmental, Supplemental
Impact model Scope (Unchanged/Changed) plus C, I, A Separate impacts on the vulnerable system (VC, VI, VA) and subsequent systems (SC, SI, SA)
Exploitability AV, AC, PR, UI AV, AC, AT (Attack Requirements, new), PR, UI
User Interaction None or Required None, Passive or Active
Time-varying metrics Exploit Code Maturity, Remediation Level, Report Confidence Exploit Maturity only; the other two were retired
Score method Algebraic formula Expert-ranked MacroVectors, lookup table and interpolation
Labels Base, Temporal, Environmental scores One score labelled CVSS-B, CVSS-BT, CVSS-BE or CVSS-BTE

Sources: FIRST CVSS v3.1 Specification, FIRST CVSS v4.0 Specification, FIRST CVSS v4.0 User Guide.

Three changes matter most in practice:

  • Defaults assume the worst. In v4.0, an unset Exploit Maturity is scored as Attacked, and unset Confidentiality, Integrity and Availability Requirements are scored as High (FIRST CVSS v4.0 Specification). Threat data can therefore only lower a vendor’s score, which FIRST says is intended because vendors must assume the reasonable worst case (FIRST CVSS v4.0 FAQ).
  • Supplemental metrics never change the number. Safety, Automatable, Provider Urgency (Clear, Green, Amber, Red), Recovery, Value Density and Vulnerability Response Effort add context only (FIRST CVSS v4.0 Specification).
  • Scores are coarser by design. There are over 15 million vector combinations but only 101 possible scores from 0.0 to 10.0, so distinct vectors share scores, and the calculator’s adjustment between neighboring vectors is limited to 0.1 (FIRST CVSS v4.0 FAQ).

3. Worked example: scoring CVE-2024-3400

CVE-2024-3400 is an OS command injection flaw in the GlobalProtect feature of Palo Alto Networks PAN-OS, published on April 12, 2024 (CVE record). It makes a good example because CVSS 3.1, CVSS 4.0, KEV, SSVC and EPSS have all rated it.

CVSS v3.1. NVD and Palo Alto both publish CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H, a base score of 10.0 (Critical), with an exploitability sub-score of 3.9 and impact sub-score of 6.0 (NVD, CVE-2024-3400). Using the equations and weights in section 7 of the v3.1 specification:

Step Calculation Result
Impact Sub-Score (ISS) 1 - (1 - 0.56) × (1 - 0.56) × (1 - 0.56) 0.9148
Impact (Scope Changed) 7.52 × (0.9148 - 0.029) - 3.25 × (0.9148 - 0.02)^15 6.05
Exploitability 8.22 × 0.85 (AV:N) × 0.77 (AC:L) × 0.85 (PR:N) × 0.85 (UI:N) 3.89
Base score (Scope Changed) Roundup(Minimum(1.08 × (6.05 + 3.89), 10)) = Roundup(Minimum(10.73, 10)) 10.0

Our sub-scores match NVD’s 6.0 and 3.9 once rounded. Changing one metric shows how sensitive the formula is (our calculations with the same equations): with Scope Unchanged the vector scores 9.8; with Privileges Required set to High it scores 9.1. Adding v3.1 temporal values for a functional exploit with an official fix (E:F/RL:O/RC:C) gives 10.0 × 0.97 × 0.95 × 1 = 9.215, which rounds up to 9.3.

CVSS v4.0. Palo Alto’s advisory publishes CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H/AU:Y/R:U/V:C/RE:M/U:Red, scored 10 (Critical) (Palo Alto Networks advisory). The supplemental values (Automatable: Yes, Recovery: User, Value Density: Concentrated, Response Effort: Moderate, Provider Urgency: Red) do not affect that score. Running the vector through FIRST’s reference calculator code (FIRST CVSS v4.0 calculator) shows what a consumer’s own data would do (the threat rows are hypothetical, since this CVE is actively exploited):

Consumer adds Resulting score (our calculation)
Nothing (CVSS-B, threat defaults to Attacked) 10
E:P (proof-of-concept only, no attacks) 9.3
E:U (no known exploit or attacks) 9.1
E:A plus low CR, IR, AR (an unimportant system) 9.6
E:A plus MAV:A (reachable only from an adjacent network) 9.4

Every variant stays Critical. The other systems agree: CISA added the CVE to KEV on April 12, 2024 with a due date of April 19, 2024 (NVD, CVE-2024-3400); CISA’s SSVC assessment in the CVE record reads Exploitation: active, Automatable: yes, Technical Impact: total (CVE record, CISA-ADP); and its EPSS score on November 15, 2026 was 0.99999, the 100th percentile (FIRST EPSS API).

Same vulnerability, two versions. FIRST’s own examples show how the move to v4.0 shifts scores in both directions (FIRST CVSS v4.0 Examples, document version 1.9):

Vulnerability CVSS v3.1 CVSS v4.0
Log4Shell, CVE-2021-44228 10.0 (Base) 9.3 (Base + Threat)
Log4Shell in immutable containers Not given 8.2 (Base), 5.5 (Base + Threat + Environmental)
Heartbleed, CVE-2014-0160 7.5 (Base), 7.0 (Base + Temporal) 8.7 (Base + Threat)
NGINX mp4 module, CVE-2022-41741 7.0 7.3 (Base)

Log4Shell drops because FIRST now scores its impact on the vulnerable system only, rather than through v3.1’s Scope Changed. The container variation is the clearest illustration of environmental scoring: the same CVE falls from Critical to Medium when read-only filesystems limit what an attacker can do.

See what a real pentest finds. Read our sample report, then get a fixed quote in writing.

4. The qualitative severity bands

Rating CVSS score (v3.1 and v4.0)
None 0.0
Low 0.1 to 3.9
Medium 4.0 to 6.9
High 7.0 to 8.9
Critical 9.0 to 10.0

Sources: FIRST CVSS v3.1 Specification, Table 14; FIRST CVSS v4.0 Specification, Table 22.

The bands apply to any CVSS score, whether base only or base plus threat and environmental metrics, and FIRST describes their use as optional (FIRST CVSS v4.0 Specification). When FIRST built v4.0 it set the score boundaries to be backwards compatible with the v3.x bands, so the labels stay comparable across versions (FIRST CVSS v4.0 Specification, section 8.1). The bands are not evenly filled: just over 10% of about 61,000 recent CVEs were Critical, and High or above covers roughly the top 48% of published vulnerabilities (FIRST, Using EPSS, 2026). A policy that says “fix every High and Critical in 30 days” is therefore a policy to fix about half of everything.

5. Limitations of CVSS

  • It measures severity, not risk. FIRST’s v3.1 user guide says CVSS “should not be used alone to assess risk” (FIRST CVSS v3.1 User Guide, section 2.1), and the v4.0 FAQ repeats that base scores “are not risk” (FIRST CVSS v4.0 FAQ). FIRST adds that a fully populated CVSS-BTE score “can be considered much closer to ‘Risk’” (FIRST CVSS v4.0 User Guide). Our threat vs vulnerability vs risk explainer covers the vocabulary.
  • Published scores are base scores. Threat and environmental metrics are the consumer’s job, so a vendor or NVD score carries the worst-case defaults until you add your own data.
  • One CVE can carry several scores. FIRST acknowledges that a vendor score and a database score can differ legitimately (different platforms) or through error, and leaves the choice to the consumer (FIRST CVSS v4.0 FAQ).
  • It rates one flaw at a time. CVSS has no formal metric for chains; the v3.1 user guide asks analysts to list each vulnerability’s score next to any chained score (FIRST CVSS v3.1 User Guide, section 3.4).
  • It predicts exploitation poorly. The EPSS team describes CVSS and EPSS as “empirically uncorrelated” and says high severity scores are “slightly better than random” at predicting exploitation activity (FIRST EPSS FAQ).

6. EPSS: the probability of exploitation

The Exploit Prediction Scoring System estimates the probability that a published CVE will be exploited in the wild in the next 30 days, as a number from 0 to 1 with a percentile rank, published daily and free via CSV and API (FIRST EPSS). It is run by the EPSS Special Interest Group at FIRST, and the scores are generated by Empirical Security (FIRST EPSS FAQ). The current model, EPSS v5, has published scores since June 15, 2026; v4 began on March 17, 2025 (FIRST EPSS, Get the Data).

The model uses about 2,850 features per vulnerability per day (exploit code, chatter across security sources, the individual CVSS metrics rather than the score, vendor and age), trained on about 120 million rows paired with more than 1.6 million recorded daily exploitation events over 12 months, from nearly a dozen data partners running honeypots, IDS/IPS, endpoint telemetry and threat intelligence feeds. One finding surprises practitioners: vulnerabilities older than two years are exploited at roughly five times the rate of those less than a month old (FIRST EPSS, How It Works).

How to read the number:

  • Exploitation is rare, so most scores are small. The mean EPSS score is around 2.8% and the median around 0.7% (FIRST, Using EPSS). A probability of 0.10 sits around the 95th percentile (FIRST EPSS FAQ).
  • Small probabilities add up: across 100 vulnerabilities scoring 0.05 each, the chance that at least one is exploited within 30 days is about 99.4% (FIRST, Using EPSS).
  • EPSS does not define Critical or High bands; the threshold is yours (FIRST EPSS FAQ). For a team moving off CVSS, FIRST suggests that acting on CVSS Critical is about the same workload as acting on the 90th EPSS percentile (a score of 0.04 or more), and acting on High and above corresponds to a threshold of around 0.008 (FIRST, Using EPSS).
  • Do not multiply EPSS by CVSS. FIRST calls it “never a good idea” because a calibrated probability times an ordinal rank has no meaning (FIRST EPSS FAQ).
CVSS EPSS CISA KEV SSVC
Question answered How bad is the flaw if exploited? How likely is exploitation in 30 days? Has it been exploited? What should we do about it?
Output 0.0 to 10.0 and a band Probability 0 to 1 and percentile Listed or not, with a due date Track, Track*, Attend or Act
Changes over time Rarely (base) Daily When entries are added When inputs change
Published by FIRST (standard); vendors and NVD (scores) FIRST EPSS SIG CISA CMU SEI and CISA

7. CISA KEV and the BOD 26-04 deadlines

The KEV catalog is CISA’s list of vulnerabilities exploited in the wild. An entry needs a CVE ID, reliable evidence of active exploitation (attempted or successful), and a clear remediation action such as a vendor update (CISA, KEV criteria). A public proof-of-concept alone does not qualify. The catalog held 1,781 entries in version 2026.11.13 (CISA KEV feed).

CISA created the catalog in 2021 under BOD 22-01. On June 10, 2026, BOD 26-04 revoked BOD 22-01 and BOD 19-02 and replaced their fixed deadlines with timelines based on four questions: is the asset publicly exposed, is the CVE in KEV, can an adversary automate exploitation, and does exploitation give partial or total control (CISA BOD 26-04). The directive binds federal civilian agencies, but CISA “strongly recommends” every organization prioritize KEV entries (CISA, KEV criteria). Selected rows of Table 1:

Publicly exposed In KEV Automatable Technical impact Deadline (calendar days)
Yes Yes Yes or No Total control 3 days and forensic triage
Yes Yes Yes Partial control 3 days
Yes No Yes Total control 3 days
Yes Yes No Partial control 14 days
Yes No No Partial control 60 days
No Yes Yes Total control 3 days and forensic triage
No No No Partial or total Fix on system upgrade

Source: CISA BOD 26-04, Table 1 (16 rows in full).

The KEV feed now carries a forensic triage flag. For example, CVE-2026-114713 in UnrealIRCd was added on November 13, 2026 with a due date of November 16, 2026 and forensic triage marked “Yes” (CISA KEV feed). The directive itself notes that technical impact “is similar to” the CVSS base score concept of severity (CISA BOD 26-04, Appendix A), so CVSS still feeds the decision; it just no longer decides it alone. Our October 2026 KEV roundup ranks a month of entries this way.

8. SSVC: a decision tree instead of a score

Stakeholder-Specific Vulnerability Categorization was created in 2019 by Carnegie Mellon University’s Software Engineering Institute with CISA, and CISA built its own tree in 2020 (CISA SSVC). Instead of a number it returns a decision: Track and Track* (remediate within standard update timelines, the second with closer monitoring), Attend (remediate sooner than standard timelines) or Act (remediate as soon as possible, with leadership involved). CISA’s tree uses five inputs: exploitation status, technical impact, automatable, mission prevalence and public well-being impact (CISA SSVC). CISA publishes the first three for CVEs through its Vulnrichment program, which is where the “active / yes / total” assessment of CVE-2024-3400 above comes from. FIRST’s position is that none of CVSS, EPSS and SSVC replaces the others; they are meant to be used together (FIRST CVSS v4.0 FAQ).

9. How bug bounty taxonomies differ: Bugcrowd’s VRT

Bug bounty platforms rate vulnerability classes rather than individual vectors, so thousands of reports can be triaged consistently. Bugcrowd’s Vulnerability Rating Taxonomy (VRT) sets a “baseline priority rating” for common vulnerabilities, including edge cases (Bugcrowd VRT), on a scale of P1 (Critical) to P5 (Informational). Bugcrowd released it at the beginning of 2016 and open-sourced it in April 2017; its baseline starts from industry guidance and is adjusted for each class’s average acceptance rate, average priority and how often programs exclude it (Bugcrowd VRT README). Version 1.19 was released on June 24, 2026 and added an Active Directory category (VRT changelog).

Sample baseline priorities from VRT 1.19 (Bugcrowd VRT 1.19 JSON):

Vulnerability VRT priority
SQL injection; remote code execution P1
Stored XSS, non-privileged user to anyone P2
Kerberoasting (insecure service account management) P2
Reflected XSS, non-self P3
Reflected XSS, self; missing X-Frame-Options header P5
Insecure direct object reference Varies by variant: P1 for viewing and modifying sensitive data through guessable IDs, P5 for non-sensitive data

The VRT publishes mappings to CVSS v3 and v4 and to CWE (Bugcrowd VRT README), but the two scales do not line up one to one. Scored with the v3.1 formula, the VRT’s own CVSS v3 vector for remote code execution comes to 9.8 (Critical, matching P1), and its vector for non-self reflected XSS to 6.1 (Medium, matching P3). Its vector for stored XSS from a non-privileged user to anyone comes to 10.0, while the VRT rates that entry P2. The VRT reflects how bug bounty programs have historically treated a class of report; CVSS describes the technical characteristics of one instance. The two answer different questions. Our bug bounty vs penetration testing comparison covers when each model fits.

10. How penetration test reports rate findings

Many findings from a web application penetration test have no CVE: a broken authorization check in your API exists only in your code. CVSS still applies; EPSS and KEV do not, because both are keyed to CVEs. A useful report therefore does four things:

  1. Prints the vector with every score, as FIRST requires of anyone publishing CVSS data, so a reader can check the number.
  2. States the version once (v3.1 or v4.0) and uses it for every finding. The bands are identical, so a severity summary reads the same either way.
  3. Adds an adjusted severity with a reason when exposure, the data reached or a chain makes a finding worse or better than its base score. PCI DSS supports this: the PCI Security Standards Council confirms that Requirement 6.3.1 “does not require that an entity accepts risk rankings assigned by external sources” (PCI SSC FAQ 1597, May 2025).
  4. Scores chains as chains, listing each finding’s score beside the combined result, per the CVSS user guide.

For findings that do match a published CVE, such as an unpatched VPN appliance found on an external test, check KEV first and EPSS second before reaching for the base score. Invadel reports use CVSS v3.1 scores and vectors with an adjusted severity and the reason for any change. Our penetration testing report template includes the severity definitions table and sample wording, and how to read penetration test results explains how to dispute and re-rank findings. To turn ratings into a fix order, see vulnerability remediation and risk-based vulnerability management.

Sources

Frequently asked questions

What is a CVSS score? A 0.0 to 10.0 rating of a vulnerability’s technical severity, defined by FIRST and published with a vector string. It maps to five bands: None, Low (0.1 to 3.9), Medium (4.0 to 6.9), High (7.0 to 8.9) and Critical (9.0 to 10.0).

What is the difference between CVSS 3.1 and CVSS 4.0? CVSS 4.0 replaces Scope with separate impacts on vulnerable and subsequent systems, adds Attack Requirements and Supplemental metrics, renames Temporal to Threat, and calculates scores from expert-ranked groups (270 equivalence sets of 15 million vectors) instead of a formula. The severity bands are unchanged.

Is a CVSS 10.0 always the top priority? No. FIRST says base scores are not risk and should not be used alone for patch prioritization. Check whether the vulnerability is in CISA’s KEV catalog, what its EPSS probability is, and whether the affected asset is exposed. Under BOD 26-04, an exposed, KEV-listed flaw giving total control gets 3 days; an internal flaw that is not in KEV and cannot be automated can wait for the next system upgrade.

How is EPSS different from CVSS? CVSS describes how severe a flaw is; EPSS estimates the probability it will be exploited in the next 30 days. Only 1.5% to 3% of published vulnerabilities show exploitation activity in a given 30-day period, so most EPSS scores are low, and a score of 0.10 already sits around the 95th percentile.

Does the CISA KEV catalog apply to private companies? BOD 26-04 binds only federal civilian agencies, but CISA strongly recommends every organization prioritize the 1,781 KEV vulnerabilities (catalog version 2026.11.13), and FIRST advises following KEV even when the EPSS score is low.

How does Bugcrowd’s VRT compare with CVSS? The VRT rates vulnerability classes from P1 to P5 based on industry guidance and Bugcrowd’s program history, which suits bug bounty triage. CVSS scores the specific instance. They usually agree on the band but not always: the VRT’s own CVSS v3 vector for stored XSS from a non-privileged user scores 10.0, while the VRT rates it P2.

Mark Kiss

Written by

Mark Kiss, Founder & CEO

15+ years in the field. Focus: Offensive security & strategy. Part of the team that scopes, tests, and reports for our clients. Full profile →

Find out what an attacker sees.

Tell us what to test and see your fixed price.

Prefer the full scoping questionnaire? 
Start the conversation

Get a quick proposal

We are online