Every vulnerability advisory arrives with a number between 0 and 10. Security teams sort spreadsheets by it, auditors ask about it, and patch SLAs are often written around it. That number is the Common Vulnerability Scoring System score, or CVSS, maintained by FIRST (the Forum of Incident Response and Security Teams). It is useful, but it answers a narrower question than most people assume. Understanding exactly what it measures is the difference between a vulnerability programme that reduces risk and one that just closes tickets.

What the base score actually describes

The score you see on most advisories is the base score. It describes the intrinsic characteristics of a flaw, the things that stay true no matter who runs the software. In CVSS v3.1, the base metrics are grouped into two sets.

The exploitability metrics describe how hard the flaw is to use: Attack Vector (can it be reached over the network, only from an adjacent network, only with local access, or only with physical access), Attack Complexity (does the attacker need special conditions to line up), Privileges Required (none, low or high), and User Interaction (does a victim have to click or open something).

The impact metrics describe what happens if it works: the effect on Confidentiality, Integrity and Availability, each rated None, Low or High. A Scope metric captures whether a compromise of one component spills over into another security boundary, for example escaping from a container into the host.

These choices are written as a vector string, a compact line such as AV:N/AC:L/PR:N/UI:N followed by the scope and impact values. The vector is more informative than the number. Two flaws can both score 7.5 for completely different reasons: one is a network-reachable flaw that only leaks data, the other a local flaw that hands over full control. Always read the vector, not just the score.

Why 9.8 shows up so often

A flaw that is reachable over the network, needs no special conditions, no credentials and no user interaction, and gives high impact on all three of confidentiality, integrity and availability without changing scope, scores 9.8 under v3.1. Remote code execution in an internet-facing service typically lands exactly there. If scope changes as well, the score reaches 10.0. That is why so many critical advisories cluster at 9.8: it is simply the arithmetic outcome of the worst unscoped combination.

The qualitative bands in v3.x are: 0.0 None, 0.1 to 3.9 Low, 4.0 to 6.9 Medium, 7.0 to 8.9 High, and 9.0 to 10.0 Critical. Many organisations attach patch deadlines to these bands, such as critical within 15 days and high within 30. Those deadlines are internal policy, not part of CVSS itself, and they vary widely between companies and regulators.

The part most teams skip: context

The base score assumes a generic deployment. It cannot know that your vulnerable server sits on an isolated lab network, or that the affected feature is switched off in your configuration, or that the machine holds your entire customer database. CVSS was designed with extra metric groups to account for that.

In v3.1 these are the temporal metrics (is exploit code available, is an official fix out, how confident is the report) and the environmental metrics (how important confidentiality, integrity and availability are for this particular asset, plus the ability to override base values to match your setup). CVSS v4.0, published by FIRST in late 2023, renamed the temporal group to Threat and simplified it around exploit maturity, kept Environmental, and added an optional Supplemental group for extra context such as whether the attack can be automated. It also introduced naming conventions like CVSS-B (base only), CVSS-BT (base plus threat) and CVSS-BTE (base, threat and environmental) so that people stop comparing a raw base score with a context-adjusted one as if they meant the same thing.

In practice, vendor advisories and the US National Vulnerability Database publish base scores. Adjusting for threat and environment is the consumer’s job, and most organisations never do it.

A worked comparison

Consider a mid-sized Pune logistics firm with two findings from the same weekly scan.

  • Finding A: a 9.8 remote code execution flaw in a reporting component installed on an internal test server. The server is reachable only from a jump host, holds synthetic data and is rebuilt weekly. No public exploit exists yet.
  • Finding B: a 7.5 information disclosure flaw in the customer-facing shipment tracking portal. It leaks other customers’ addresses and phone numbers. A working exploit has been published and is being used in the wild.

Sorted by base score, A goes first. Reasoned through context, B is far more urgent: it is internet-exposed, it touches personal data that the company is obliged to protect under India’s Digital Personal Data Protection Act, and attackers are already using it. A still needs fixing, but its effective attack vector is closer to adjacent than network, its confidentiality requirement is low, and there is no exploit pressure. A team that patched A first and B a fortnight later would have followed the spreadsheet and still made the wrong call.

What CVSS is not

FIRST itself states that CVSS measures severity, not risk. Risk depends on the likelihood that someone will actually exploit the flaw against you and on what the affected asset is worth. That is why other signals have grown up around CVSS:

  • EPSS (Exploit Prediction Scoring System, also from FIRST) estimates the probability that a CVE will be exploited in the wild over the next 30 days, based on observed data.
  • CISA’s Known Exploited Vulnerabilities catalogue lists flaws with confirmed real-world exploitation. Presence on that list is strong evidence that something deserves to jump the queue.
  • Your own asset inventory: internet exposure, data sensitivity and business criticality. No external feed knows these.

A sensible prioritisation model treats the CVSS base score as one input among these, not as the verdict.

Common ways the score gets misused

Treating scores from different versions as comparable. A v2 score, a v3.1 score and a v4.0 score for the same flaw can differ, because the formulas and metrics differ. Older CVEs may only carry v2 scores, which used a different scale for some concepts. Note the version alongside the number.

Trusting a single source. The vendor, NVD and a scanner plug-in sometimes score the same CVE differently because they interpreted a metric differently. When the numbers disagree, read the vectors and decide which reading matches your deployment.

Using the score to decide whether to patch at all. Low and medium findings can chain together. A medium information leak that reveals internal hostnames plus a medium authentication weakness may add up to a serious path in. Severity bands should set the pace of remediation, not permission to ignore.

Writing SLAs that cannot be met. If policy says every critical must be patched in seven days but the change window for a core banking system is monthly, the policy will be breached constantly and soon ignored. Better to define compensating controls, such as blocking the vulnerable service at the firewall or disabling the affected feature, that count as interim remediation with a documented exception.

Read well, CVSS is a shared vocabulary that lets a vendor in California and a security analyst in Chennai describe the same flaw consistently. Read lazily, it becomes a sorting key that pushes the wrong work to the top. The habit worth building is simple: look at the vector, ask whether each metric is true in your environment, and then ask whether anyone is actually exploiting it.

Written by the RP Strativa Team. The company and findings in the example are illustrative.

Leave a Reply

Trending