CVSS vs EPSS vs KEV: prioritising vulnerabilities under the CRA
CVSS rates how severe a vulnerability's technical impact would be, EPSS estimates the probability it will be exploited in the next 30 days, and CISA's KEV catalogue lists vulnerabilities with evidence of exploitation in the wild. Together they rank triage; none decides CRA reportability alone.
By the Vellaci teamPublished Updated 7 min readOperational guidance, not legal advice
Key facts
- CVSS
- FIRST standard: a 0.0–10.0 severity score; v4.0 (November 2023) has Base, Threat, Environmental and Supplemental metrics
- CVSS ratings
- None 0.0 · Low 0.1–3.9 · Medium 4.0–6.9 · High 7.0–8.9 · Critical 9.0–10.0
- EPSS
- FIRST model: a daily probability (0–1) that a published CVE is exploited in the wild in the next 30 days, with a percentile
- KEV
- CISA's catalogue of CVEs with reliable evidence of exploitation — 1,748 entries in catalogue version 2026.10.16
- CRA relevance
- Exploitation evidence feeds the Article 14 'actively exploited' review; severity and likelihood set remediation priority (Annex I Part II(2))
- Decides reportability?
- No signal does on its own — a named person does, on the evidence for the specific product
Three different questions
| Aspect | CVSS | EPSS | CISA KEV |
|---|---|---|---|
| Question | How severe would exploitation be? | How likely is exploitation activity in the next 30 days? | Is it being exploited in the wild? |
| Publisher | FIRST (scores assigned by NVD, CNAs and vendors) | FIRST (data-driven model) | US Cybersecurity and Infrastructure Security Agency |
| Output | Score 0.0–10.0 and a vector string | Probability 0–1 and percentile | Catalogue entry with dates and required action |
| Updates | When the CVE record is (re)scored | Daily, for every scored CVE | As CISA adds entries |
| Strength | Describes technical impact consistently | Ranks likelihood across the whole CVE population | Evidence, not prediction |
| Limit | Says nothing about likelihood or your configuration | A probability, not evidence; only for published CVEs | US-centred and incomplete; absence is not proof of safety |
CVSS: severity, not likelihood
The Common Vulnerability Scoring System, maintained by FIRST, describes a vulnerability's characteristics — attack vector, complexity, privileges and user interaction required, and impact on confidentiality, integrity and availability — and turns them into a score from 0.0 to 10.0. CVSS v4.0, first published on 1 November 2023, organises metrics into Base, Threat, Environmental and Supplemental groups; scores should be labelled with the groups used (for example CVSS-B for base only). Most public records still carry v3.1 base scores.
Base scores describe the worst case in a generic environment. Your product's configuration, exposure and mitigations can make the real severity much lower or higher, which is what the environmental metrics and your own triage are for. The qualitative scale is the same in v3.1 and v4.0:
| Rating | Score |
|---|---|
| None | 0.0 |
| Low | 0.1 – 3.9 |
| Medium | 4.0 – 6.9 |
| High | 7.0 – 8.9 |
| Critical | 9.0 – 10.0 |
EPSS: the probability of exploitation
The Exploit Prediction Scoring System, also from FIRST, estimates for every published CVE the probability that it will be exploited in the wild in the next 30 days, published daily as a value between 0 and 1 with a percentile. A percentile of 0.95 means the CVE scores at least as high as 95 % of all scored CVEs. EPSS is useful precisely where CVSS is weak — telling apart the few critical-rated vulnerabilities that attackers actually use from the many that they do not — but it is a model output. A high EPSS score is a reason to look for exploitation evidence and to remediate sooner, not evidence that exploitation has happened.
KEV: evidence of exploitation in the wild
CISA's Known Exploited Vulnerabilities catalogue lists CVEs for which CISA has reliable evidence of active exploitation. Each entry carries the CVE identifier, vendor and product, a vulnerability name and short description, the date added, the required action and a due date that applies to US federal civilian agencies, whether use in ransomware campaigns is known, notes with references and the related CWE identifiers. The catalogue is published as CSV and JSON feeds; the version of 16 October 2026 listed 1,748 CVEs. For a CRA manufacturer, a KEV listing of a component it ships is strong evidence that the 'actively exploited' threshold may be met for its own product — and a reason to open an Article 14 reportability review immediately.
How the signals map to CRA obligations
Annex I Part II(2) requires manufacturers to address and remediate vulnerabilities without delay 'in relation to the risks posed'. Severity (CVSS, adjusted for the product) and likelihood (EPSS) are the natural inputs to that risk judgement and to remediation targets; Annex I Part II(4) also expects published advisories for fixed vulnerabilities to state their severity.
Article 14 is a different question. An actively exploited vulnerability is one for which there is reliable evidence that malicious code was executed on a system without the owner's permission. KEV entries, vendor and CSIRT advisories, customer reports and your own telemetry are possible sources of that evidence. The determination is per product: a KEV-listed CVE in a component you ship but whose vulnerable code is absent or unreachable is a candidate for a documented 'not reportable' decision — and ideally a VEX statement — rather than a notification.
A triage policy that combines them
An example policy, to adapt and document; the thresholds are choices to calibrate to your capacity, not regulatory values.
| Tier | Condition | Action | Example target |
|---|---|---|---|
| 0 | KEV-listed, or other reliable exploitation evidence, and the component ships in a supported version | Open the Article 14 reportability review now; record awareness when evidence is confirmed; remediate first | Review within hours |
| 1 | CVSS 9.0 or higher, or EPSS above your chosen threshold (for example 0.10), and the vulnerable code is reachable | Fix in the next security release; publish mitigations meanwhile | 7 days |
| 2 | CVSS 7.0–8.9 and reachable | Scheduled security release | 30 days |
| 3 | Everything else | Normal release cycle, or a not-affected decision with a VEX justification | 90 days or documented |
Worked example: CVE-2023-44487
The HTTP/2 'Rapid Reset' vulnerability carries an NVD CVSS v3.1 base score of 7.5 (High) — availability impact only. CISA added it to KEV on 10 October 2023, and its EPSS probability on 20 October 2026 was above 0.99. CVSS alone would put it in an ordinary high-severity queue; KEV and EPSS move it to the top. For each product that ships an affected HTTP/2 implementation, the review asks whether the vulnerable server code is reachable: if yes, remediate first and assess reportability on the evidence; if not, record a not-affected decision with the reason.
Common mistakes
- Using the CVSS base score as the priority, so internet-facing exploited issues wait behind theoretical criticals.
- Treating a high EPSS score as proof of exploitation — or a low one as proof of safety.
- Reading 'not in KEV' as 'not exploited'; KEV is one catalogue, not the whole picture.
- Ignoring version ranges and reachability, so the queue fills with components that are present but not affected.
- Not recording which score, version and date informed a decision, so it cannot be explained later.
Frequently asked questions
- What is the difference between CVSS and EPSS?
- CVSS rates how severe a vulnerability's impact would be if exploited, on a 0–10 scale. EPSS estimates how likely it is to be exploited in the next 30 days, as a probability between 0 and 1. Use CVSS for impact and EPSS for urgency.
- What is a KEV?
- A KEV is a known exploited vulnerability: a CVE that CISA has added to its Known Exploited Vulnerabilities catalogue because there is reliable evidence it is being exploited in the wild. The catalogue gives the date added, the required action and a remediation due date for US federal agencies.
- What is a good EPSS threshold?
- There is no universal one. Choose a threshold your team can remediate within your target time, check how many open findings it selects, and review it periodically. Many teams combine a probability threshold with a percentile and always treat KEV-listed findings as top priority regardless of EPSS.
- Does a KEV listing mean we must report under the CRA?
- Not automatically. It is strong evidence that a vulnerability is actively exploited, so it should trigger an immediate reportability review for every product that ships the component. Whether your product's vulnerability is actively exploited — and so reportable under Article 14 — depends on whether the vulnerable code is present and exposed in that product.
- Should we use CVSS v3.1 or v4.0?
- Use what your sources provide and label it. Most advisories still carry v3.1 scores; v4.0 adds threat and supplemental metrics and clearer nomenclature. For CRA advisories, state the version and the vector with the score so customers can interpret it.
From requirement to evidence
What the obligations in this guide mean for an engineering, security or compliance team, the Vellaci workflow that supports each one, and the records it leaves. The full map for every CRA obligation is on the CRA compliance software page.
How to read this. Official requirement summarises the text of Regulation (EU) 2024/2847 with its reference. What your team does is Vellaci’s operational interpretation, not legal advice. Vellaci workflow and Evidence describe what the product does today. Methodology · reviewed 2026-10-05.
- Official requirement
- In relation to the risks posed, address and remediate vulnerabilities without delay, including by providing security updates; where technically feasible, provide security updates separately from functionality updates.
- What your team does (interpretation)
- Every matched vulnerability gets a human decision — affected, not affected, fixed, accepted — with a reason, an owner and, if affected, a remediation target and the version that fixes it.
- Vellaci workflow
- Triage with mandatory reasons, explainable priority (CVSS, EPSS, KEV, exposure), remediation owner and due date, Jira and Linear ticket links. How it works
- Evidence produced
- Decision history per finding with reviewer and timestamp
- Remediation owner, due date and fixed version
- Linked tickets
- Official requirement
- Notify any actively exploited vulnerability contained in the product to the CSIRT designated as coordinator and to ENISA via the single reporting platform: early warning within 24 hours of becoming aware, notification within 72 hours, final report no later than 14 days after a corrective or mitigating measure is available.
- What your team does (interpretation)
- Decide who confirms awareness and reportability, record when they did it and in which timezone, and have the content of each stage ready before the clock runs out.
- Vellaci workflow
- Explicit CRA review on every finding, reporting case with server-computed 24 h / 72 h / final-report deadlines, staged and prefilled forms. How it works
- Evidence produced
- Reportability decision with rationale and reviewer
- Confirmed awareness time and timezone
- Frozen revision of each submitted stage with SRP reference