CVSS Score Explained: Metrics, Severity Bands & v4.0 Changes

CVSS score explained properly: the 0–10 severity bands, every base metric, a worked 9.8 calculation by hand, and what changed in v4.0 — including why a CVSS…

A CVSS score is a number from 0.0 to 10.0 that describes how damaging a vulnerability would be if someone exploited it. It shows up on every CVE, every scanner report and every patch advisory — and it is routinely misread as a risk rating. This page explains what the score actually measures, how it is calculated, and how to read one without the common mistakes.

The severity bands

Every CVSS score maps to one of five qualitative ratings. The bands are identical in v3.1 and v4.0:

RatingScore
None0.0
Low0.1 – 3.9
Medium4.0 – 6.9
High7.0 – 8.9
Critical9.0 – 10.0

When someone says "this CVE is a 9.8", they mean Critical — and almost always mean the base score, which is the severity of the flaw in a vacuum.

What the score is made of

CVSS groups its metrics. In v3.1 there are three groups; v4.0 reshuffles them into four:

GroupWho sets itChanges the score?
BaseVendor or CNA, at disclosureYes
Temporal (v3.1) / Threat (v4.0)Consumer, from threat intelYes
EnvironmentalConsumer, about their own deploymentYes
Supplemental (v4.0 only)EitherNo — context only

The base metrics split into two questions. Exploitability — how easy is it to exploit? — covers Attack Vector (network, adjacent, local, physical), Attack Complexity, Privileges Required and User Interaction. Impact — what does the exploit damage? — rates Confidentiality, Integrity and Availability. In v3.1 a Scope metric captures whether the exploit escapes the vulnerable component; v4.0 replaces Scope with separate impact ratings for the vulnerable system (VC, VI, VA) and for downstream systems (SC, SI, SA).

A worked base score, by hand (v3.1)

The classic unauthenticated remote-code-execution shape — think a network service that runs attacker input — is this vector:

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

Each metric has a fixed weight: AV:N = 0.85, AC:L = 0.77, PR:N = 0.85, UI:N = 0.85, and each High impact = 0.56. The v3.1 specification then computes:

Exploitability = 8.22 x 0.85 x 0.77 x 0.85 x 0.85 = 3.887
Impact         = 6.42 x [1 - (1-0.56)(1-0.56)(1-0.56)] = 5.873
Base Score     = Roundup(min(5.873 + 3.887, 10)) = Roundup(9.760) = 9.8

That 9.8 — Critical — is the score most unauthenticated RCEs receive when Scope is Unchanged. You can score your own vectors with the CVSS Score Calculator — the full v3.1 base, temporal and environmental formula runs in your browser, and this worked example is one of its presets. With Scope Changed the same metrics score 10.0, which is exactly what happened with Log4Shell (CVE-2021-44228): NVD published CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H — 10.0 — because a crafted log line escapes the logging library and takes over the host process and often everything downstream of it.

What changed in v4.0

CVSS 4.0 was published by FIRST in November 2023. The important differences:

Areav3.1v4.0
Attack conditionsinside Attack Complexitysplit out as Attack Requirements (AT)
User InteractionNone / RequiredNone / Passive / Active
ScopeS:U / S:Cremoved — replaced by VC/VI/VA + SC/SI/SA
Time-based groupTemporal: E, RL, RCThreat: exploit maturity only, and Attacked is the default
Scoringpublished formulamacrovectors — 270 equivalence sets, expert-ranked, then interpolated
Labels—CVSS-B, CVSS-BT, CVSS-BE, CVSS-BTE

Two consequences matter in practice. First, you cannot hand-compute a 4.0 score in a spreadsheet the way you can with the v3.1 formula — use a calculator built on the FIRST reference implementation. Second, the Threat group defaults to "attacked": a Base score is effectively computed as if exploitation is already happening, and only evidence that it is not (proof-of-concept only, or unreported) lowers it. That is the mechanism by which 4.0 expects you to feed in sources like CISA KEV and EPSS.

Severity is not risk

First things first: FIRST themselves say the Base score measures severity, not risk. A 9.8 in an internal library that nothing exposes and no one attacks is not the same emergency as a 7.5 that is in CISA’s Known Exploited Vulnerabilities list with public exploits. A sane priority queue combines:

  • CVSS — how bad the flaw is in principle.
  • Exploit maturity — is anyone actually exploiting it (CISA KEV, EPSS probability).
  • Exposure — is the vulnerable component internet-facing, and what is downstream of it.

The same logic is our standing advice in the CVE & CVSS glossary entry: combine severity with exploitation intelligence rather than patching strictly by base score.

Frequently asked

What is a CVSS score, in one sentence?

A 0.0–10.0 rating of how severe a security vulnerability is, maintained by FIRST, built from metrics covering how easy it is to exploit and how much damage it does.

What does a CVSS score of 9.8 mean?

Critical (the 9.0–10.0 band). In practice 9.8 is the signature of unauthenticated remote code execution with no user interaction — the worked example above is exactly that shape.

Is CVSS a risk score?

No. It is severity only: the flaw’s intrinsic technical impact, with no account of whether your systems are exposed or whether attackers are active. Risk comes from layering exploit intelligence and exposure on top.

What is the difference between CVSS v3.1 and v4.0?

The metric set (Attack Requirements, split system impacts, richer User Interaction), the time-based group (Threat with an "attacked" default replaces Temporal), and the scoring method (macrovector lookup and interpolation instead of a formula). The severity bands are unchanged.

Can I calculate a CVSS score by hand?

For v3.1, yes — the base score is a published formula and the worked example above is the whole computation. For v4.0, no — scores come from a macrovector lookup table, so use a calculator that implements the FIRST reference.

Who maintains CVSS?

FIRST (the Forum of Incident Response and Security Teams). v4.0.0 was released in November 2023; v3.1 remains widely published — NVD, for instance, still scores with v3.1 alongside v4.0.

See also: the CVE & CVSS glossary entry for the identifier side, Nuclei for severity-aware scanning, and zero-day vulnerabilities for the case where no score exists yet.


Keep reading

Related articles

Browse all articles