The Exploit Prediction Scoring System (EPSS) is a machine learning model that estimates the probability a published vulnerability will be exploited in the wild within the next 30 days. Maintained by FIRST, EPSS produces a daily score between 0 and 1 (expressed as a percentage) for every known CVE, helping security teams focus remediation on the vulnerabilities most likely to be used in real attacks rather than treating every high-severity finding as equally urgent.
The Exploit Prediction Scoring System, or EPSS, is a data-driven model that answers a question CVSS can't: how likely is this vulnerability to actually be exploited?
Developed and maintained by FIRST (Forum of Incident Response and Security Teams), EPSS uses machine learning trained on real-world exploitation data to assign each published CVE a probability score between 0 and 1. A score of 0.95 means EPSS estimates a 95% chance the vulnerability will be exploited in the next 30 days. A score of 0.02 means a 2% chance. Scores are updated daily as new threat intelligence flows in.
The current version, EPSS v4, was released in March 2025. It covers all public CVEs and draws on signals including:
EPSS also provides a percentile ranking that places each vulnerability relative to all other CVEs. A vulnerability at the 95th percentile has a higher exploitation probability than 95% of all known vulnerabilities.
Startups face the same vulnerability volume as large enterprises but with a fraction of the staff to address them. The Exploit Prediction Scoring System provides the exploitation intelligence needed to focus your limited remediation capacity where it counts.
Here's why it matters:
For startups building a vulnerability management program, here's how to integrate the Exploit Prediction Scoring System:
EPSS v4 achieves an ROC AUC of 0.838, meaning it correctly distinguishes between exploited and non-exploited vulnerabilities about 84% of the time. In practical terms, vulnerabilities scored in the top 10% by EPSS account for a disproportionately large share of observed exploitation activity. No prediction model is perfect, as EPSS can't predict zero-day exploitation or attacks using vulnerabilities without CVE identifiers. But as a prioritization tool, it significantly outperforms static severity scoring alone.
EPSS scores are recalculated and published daily. Each day's scores reflect the latest threat intelligence, exploit availability data, and exploitation evidence. This means a vulnerability's EPSS score can change significantly over time as new exploit code emerges, threat actors begin targeting it, or exploitation activity subsides. You can access current scores and historical data through the FIRST EPSS API or download the daily dataset from first.org/epss.
No, and it's not designed to. CVSS and EPSS measure fundamentally different things. CVSS measures technical severity (how bad a vulnerability is if exploited), while EPSS measures exploitation probability (how likely it is to be exploited). A vulnerability can be technically severe (CVSS 9.8) but rarely exploited (EPSS 0.01), or moderately severe (CVSS 5.0) but frequently exploited (EPSS 0.8). The best practice is to use both together: CVSS to understand the potential impact, and EPSS to understand the likelihood. Add CISA's KEV catalog for confirmed active exploitation, and you have a well-rounded prioritization framework.