COMPARISON, read 10 Nov 2026
EPSS is an input and SSVC is a decision: use EPSS to answer the exploitation question inside an SSVC tree
- Maintained by
- SSVC: CERT/CC, with CISA publishing data. EPSS: FIRST EPSS SIG
- Scale
- SSVC: action categories. EPSS: probability 0 to 1 and percentile
- Current version
- SSVC: cisa:BOD2604:1.0.0. EPSS: model v2026.06.15
- Updated
- SSVC: per CVE. EPSS: daily
- Answers
- SSVC: what to do and how fast. EPSS: the chance of exploitation in the next 30 days
How they relate
The two are different kinds of object. EPSS is a measurement that produces a probability for every published CVE. SSVC is a method that takes several measurements, including an exploitation measurement, and returns an action. The SSVC documentation has a page called Using EPSS with SSVC, with sections titled EPSS Probability as input to Exploitation and EPSS Percentiles as an Amplifier.
This makes the comparison one of position in a process. EPSS sits upstream; SSVC sits at the decision. A program does not choose between them. It chooses how to map an EPSS value to an SSVC Exploitation answer, and writes the mapping down.
| Property | SSVC | EPSS |
|---|---|---|
| Role | Decision method | Input signal |
| Output | Action category | Probability and percentile |
| Needs asset data | Yes, for exposure and impact | No: population-level |
| Method published | Decision tables and code | Model described; pipeline not published |
| Refresh | Per decision | Daily |
The mapping problem
The Exploitation decision point in SSVC takes values such as none, proof of concept and active. EPSS is a continuous probability. A mapping needs thresholds, and FIRST warns that thresholds are subjective: bins compress scores that differ by a tiny amount into different categories, and there is no universal right cutoff.
The EPSS SIG's own guidance is to communicate the probability with its percentile. A mapping in an SSVC tree should therefore name the probability cutoff, the percentile cutoff if used, the date it was chosen and the reviewer.
The percentile deserves its own rule. FIRST explains that percentiles are computed over every published CVE, so the percentile of a CVE within your inventory can differ from its published percentile. A rule that uses percentiles as an amplifier should state which population it uses. A published percentile of 0.9 means a score above 90 percent of all scored CVEs, not above 90 percent of yours.
What the ledger says about the mapping
On listing day, EPSS was a weak indicator of confirmed exploitation in this window. Of 260 scored KEV additions, 197 (75.8 percent) were below 0.1 and the median was 0.0066. A mapping that required EPSS at 0.1 to set Exploitation to active would have set it wrongly for three in four of the confirmed entries.
For 58 additions there was no score on that day. In an SSVC tree, a missing EPSS value must have a rule, such as defaulting to the Exploitation answer from Vulnrichment or from KEV. CISA's tree solves this by making KEV membership a separate decision point, In KEV, rather than a derived value.
The missing-value group is not random. The 58 additions without a score on listing day had a median gap from NVD publication of 0 days, so they are the newest CVEs, the ones for which the exploitation question is most urgent. A tree that defaults to none when EPSS is absent would route exactly those CVEs to a slower category.
What each contributes that the other cannot
SSVC contributes the decision. EPSS says nothing about a specific asset, and FIRST states that the score is not about your environment. SSVC takes exposure and mission impact into account and returns a time to act. EPSS contributes breadth: it provides a score for every CVE, without anyone answering questions.
CISA's data illustrates the split. CISA publishes SSVC answers for CVEs it assesses; EPSS publishes a number for all of them. A program with thousands of open CVEs will have many that CISA has not assessed, and EPSS is the fallback signal for those.
Timing differs too. EPSS publishes a new score every day, typically within a minute or two after 13:30 UTC. An SSVC decision is made at an event: when a CVE is published, when it is listed, when an asset changes exposure. A program needs a rule for when to rerun the decision as the EPSS value moves, such as at every KEV addition and at each weekly review.
A practical design
Use the CISA tree for KEV-listed CVEs. For the rest, use EPSS to set the Exploitation input: below a documented low threshold, none; above a documented high threshold, treat as active for planning; in between, a reviewer decides. Record the thresholds in the policy.
Test the design on the ledger's CVE list. Run each KEV addition through your mapping as of its listing day, using the EPSS value from the FIRST API for that date, and count how many your design would have routed below your fast category. That count is the miss rate.
Keep the design small at the start, and record the first quarter of results before adding branches. Three Exploitation answers and two thresholds are enough to run. A tree with many branches and many thresholds produces results nobody can explain, and the CERT/CC documentation itself describes SSVC as a human-scale decision aid, with a section titled SSVC as a Human-Scale Bottleneck.
Reporting
Report the SSVC outcome as the priority and show the EPSS value, with its date, as supporting evidence. Do not publish a number that combines them. FIRST calls multiplying EPSS by an ordinal score laundering, and the same objection applies to any blend.
The audit trail is then simple: outcome, decision points, EPSS value and date, source for each.
Verdict by situation
- Federal agencies and contractors: Rank by the SSVC outcome first because BOD 26-04 prescribes it, and bring EPSS in only to help answer the exploitation question for CVEs outside the catalog.
- Large enterprises with ten or more scanners: Rank by the SSVC outcome first for tiered assets and feed EPSS into its Exploitation input, because the decision has to carry asset context EPSS cannot.
- Managed security providers: Rank by EPSS first because it needs no per-client asset answers and is comparable across clients, then add an SSVC outcome where a client supplies exposure data.
- Mid-size teams replacing Kenna or a spreadsheet: Rank by EPSS first because a small team can adopt it in a day, and move to an SSVC decision when asset exposure data is reliable.
- Application security and infrastructure in one queue: Rank by the SSVC outcome first where assets span code and infrastructure, since a decision table can hold both, and use EPSS for the CVE-bearing items only.
Other comparisons
Sources
- CERT/CC, SSVC documentation (Using EPSS with SSVC), read 10 Nov 2026
- CERT/CC, SSVC overview, read 10 Nov 2026
- CERT/CC, CISA BOD 26-04 Response Model, read 10 Nov 2026
- FIRST, EPSS home, read 10 Nov 2026
- FIRST, How EPSS works, read 10 Nov 2026
- FIRST, Probability, percentile and bins, read 10 Nov 2026
- FIRST, Using EPSS, read 10 Nov 2026
Last reviewed 10 Nov 2026.