EPSS
Exploit Prediction Scoring System
Definition
The Exploit Prediction Scoring System (EPSS) estimates the probability that a published vulnerability (a CVE) will be exploited in the wild within the next 30 days. Each score is a number between 0 and 1, recalculated daily by FIRST, the organization that also maintains CVSS.
Security teams use EPSS to decide which of thousands of known vulnerabilities to fix first, based on how likely attackers are to use them rather than how severe they look on paper.
How EPSS works
EPSS is a machine-learning model trained on observed exploitation activity and on what is known about each vulnerability: its description, its CVSS details, whether exploit code has been published, how often it's discussed, and how vulnerabilities like it have been exploited before. Every day it produces two numbers for each CVE:
- Probability (0 to 1): the estimated chance of exploitation activity in the next 30 days. A score of 0.42 means roughly a 42% chance.
- Percentile: where that score ranks against every other scored CVE. A CVE in the 95th percentile scores higher than 95% of the others.
The two are easy to confuse. Most CVEs have a very low probability, so a CVE can sit in a high percentile with a probability of only a few percent. Use the probability for decisions and the percentile for context.
Scores are free. FIRST publishes them as a daily CSV of every scored CVE and through a public API. The current model was released in March 2025.
EPSS vs CVSS vs CISA KEV
The three answer different questions, which is why most teams use them together:
| Answers | Scale | Changes | |
|---|---|---|---|
| CVSS | How bad would it be if this were exploited? | 0 to 10 severity | Rarely, once scored |
| EPSS | How likely is it to be exploited soon? | 0 to 1 probability | Daily |
| CISA KEV | Has it actually been exploited? | On the list or not | When CISA adds an entry |
CVSS alone over-prioritizes: a large share of CVEs score "high" or "critical", but only a small fraction are ever exploited. EPSS narrows that list by likelihood. KEV is confirmed evidence, so it overrides both.
How to use EPSS
A common pattern:
- Fix anything on CISA KEV first. Exploitation is confirmed.
- Rank the rest by EPSS and CVSS together. High likelihood and high severity first. Set a probability threshold that your team can actually keep up with, then adjust it as you see results.
- Apply your own context. Is the vulnerable component deployed? Is it reachable from the internet? Does it handle sensitive data?
- Re-check daily. EPSS scores move as exploit code appears or attacks begin, so a low score today is not a low score forever.
What EPSS can't tell you
- Whether it affects you. EPSS is a global likelihood. It says nothing about whether the vulnerable component is in your software, which version you run, or where it runs.
- Anything without a CVE. EPSS only scores published CVEs. A malicious package slipped into a registry has no CVE and no EPSS score, and it is often the more urgent problem.
- Targeted attacks. The model learns from observed exploitation, so attacks against a few organizations can stay invisible to it.
- Certainty. It's a probability. Some vulnerabilities later added to KEV had low EPSS scores beforehand.
EPSS and Crash Override
EPSS tells you which vulnerabilities attackers are likely to use. It can't tell you where they are sitting in your environment. That is the first question when something changes: where is it running right now?
Crash Override is AppSec for the AI era. The build sensor records what every artifact is and what's inside it, and that record follows the artifact into production, so the data plane knows where each component actually runs. That's Exposure response: in minutes you see every service running an affected component, every one that ever ran it, and when. Only truly malicious packages get blocked; everything else is prioritized by where it actually runs, so teams fix what's in production first. Less work, not more.
EPSS: frequently asked questions
Exploit Prediction Scoring System. It's maintained by FIRST (the Forum of Incident Response and Security Teams), which also maintains CVSS.
Lower is better: it's the estimated probability of exploitation in the next 30 days. There's no universal cut-off. Teams pick a threshold they can keep up with, often fixing everything above a set probability alongside anything on CISA KEV.
CVSS rates how severe a vulnerability would be if exploited, on a 0 to 10 scale. EPSS estimates how likely it is to be exploited soon, as a 0 to 1 probability. A critical CVSS score with a low EPSS score is often less urgent than a medium one that attackers are actively using.
Daily. FIRST recalculates every score each day and publishes them as a CSV and through a free API.
No. It means exploitation looks unlikely in the next 30 days based on what the model can see. Scores change, targeted attacks can be missed, and anything on CISA KEV should be fixed regardless of its EPSS score.
Related Terms
Can this ship? Are we exposed? Can you prove it?
Every CISO is asking these three questions. We answer them with facts, not guesses, from repository to runtime.