Skip to content
StratQuad

Exploit Prediction Scoring System (EPSS)

3 min read

A daily, data driven estimate of the probability that a published CVE will see exploitation activity in the next 30 days.

At a glance

Owner
Forum of Incident Response and Security Teams (FIRST), EPSS Special Interest Group
Version
EPSS v5
Published
15 Jun 2026
Applies to
Any organisation prioritising vulnerability remediation
Mandatory
Voluntary
Assessed by
Model output published by FIRST; no assessment
Cost model
Free, including the public API

Key points

  • EPSS gives each published CVE a probability between 0 and 1 that exploitation activity will be observed in the next 30 days.
  • Scores are recalculated daily. A score is a snapshot, not a property of the vulnerability.
  • The percentile shows how a CVE ranks against all others: a percentile of 0.95 means 95 per cent of scored CVEs have an equal or lower probability.
  • The current model is version 5, released on 15 June 2026; version 4 had been in use since 17 March 2025.
  • EPSS predicts likelihood. CVSS describes severity. They answer different questions and are used together.

What it is

EPSS is a machine learning model maintained by a FIRST special interest group. Each day it combines information about every published CVE, such as its description, the vendor, CVSS characteristics, references to exploit code and whether it appears in lists such as the CISA KEV catalogue, with observed exploitation activity reported by data partners from honeypots, intrusion detection sensors and endpoint telemetry. It outputs a probability for every CVE.

The prediction is specific: exploitation activity (attempts observed in the wild, not confirmed compromises) within the next 30 days.

Reading a score

FieldMeaningExample
EPSS scoreProbability of exploitation activity in the next 30 days0.94 (94 per cent)
PercentileProportion of scored CVEs with an equal or lower score0.999 (higher than 99.9 per cent of CVEs)

Most CVEs score very low; a small minority score high. Because the distribution is so skewed, a modest looking probability such as 0.10 can already sit in a high percentile.

Note

Scores move. A CVE can score 0.02 on the day it is published and 0.90 a week later when exploit code appears. Store the date with any score you record, and re-pull scores rather than caching them for long.

Version history

VersionReleased
v17 January 2021
v24 February 2022
v37 March 2023
v417 March 2025
v5June 2026

Historic daily scores are available through the API from 14 April 2021. Scores from different model versions are not directly comparable.

Getting the data

FIRST publishes the full daily score set as a downloadable file and through a free API, for example:

Code
https://api.first.org/data/v1/epss?cve=CVE-2024-3400

Using it well

Practitioner note

EPSS is most useful for the long tail. Your scanner reports thousands of findings scored High or Critical by CVSS; EPSS tells you which handful attackers are actually likely to use. A common pattern is: fix everything in the CISA KEV catalogue first, then everything above a chosen EPSS threshold, then the rest by CVSS and asset value. FIRST presents thresholds as a trade off between effort and coverage rather than recommending a single value; choose one by checking how much work it creates against how much exploited activity it would have caught in your own environment.

Limitations

Limitations

  • EPSS predicts activity across the internet, not against your organisation. It knows nothing about your exposure, compensating controls or asset value.
  • It scores published CVEs only. Vulnerabilities without a CVE, misconfigurations and zero days before disclosure are out of scope.
  • Exploitation data records attempts, and exploitation is bursty, so short term scores can swing.
  • A low score is not a guarantee. Targeted attacks against niche products may never appear in the telemetry EPSS learns from.
  • Already exploited vulnerabilities belong in KEV based workflows regardless of their EPSS score.

Sources

Suggest a correction