EPSS (Exploit Prediction Scoring System)
EPSS, the Exploit Prediction Scoring System, is a probability score maintained by the FIRST.org special interest group that estimates how likely a given CVE is to be exploited in the wild within the next 30 days. Each score runs from 0 to 1 (0% to 100%) and updates daily as new exploitation evidence arrives. Security teams read EPSS alongside CVSS severity to set patch order. CVSS rates how damaging a vulnerability would be if exploited. EPSS estimates how likely that exploitation is. For AI infrastructure, EPSS helps rank which model server, inference endpoint, and MCP server CVEs to remediate first.
How EPSS scoring works
The FIRST.org EPSS group trains a model on observed exploitation data (honeypot activity, intrusion-detection telemetry, and public exploit publication) and publishes a daily score for every CVE with a published identifier. A score of 0.90 means the model estimates a 90% probability of exploitation activity in the next 30 days. Because the score moves as evidence accumulates, a CVE that looked low-priority at disclosure can climb sharply once a working exploit circulates.
Where EPSS fits in AI security operations
An AI platform runs on a stack of model servers, inference endpoints, agent frameworks, and MCP servers, each carrying its own CVEs. EPSS gives the team a defensible order for that patch queue when the list is longer than the maintenance window. DeepInspect does not patch those systems. What it contributes is the per-decision audit record of which identity called which endpoint, so when a high-EPSS CVE affects a component in the AI request path, responders can see exactly what AI traffic traversed it.
Related reading
- EPSS and KEV Patch Prioritization for the AI Stack: A Working Guide
CVSS severity alone tells you a vulnerability is bad, not whether it is being exploited. This guide shows how to combine the FIRST EPSS exploitation probability score with the CISA Known Exploited Vulnerabilities catalog to rank patch work across your AI inference stack, with a scoring rule, a worked example on real 2026 AI-tool CVEs, and an honest line on what patch prioritization does and does not cover.
- AI Bug-Hunting Drove ~1,500 High-Severity CVEs in June 2026: Prioritizing the Patch Window
AI models are now finding software flaws at scale. Epoch AI counted roughly 1,500 high-severity and critical CVEs reported in June 2026, about 3.5x the previous monthly record, while real exploitation still arrives within hours of disclosure. This walks through why patch prioritization by exploitability, plus identity-bound inline policy at the AI request boundary, is the control that holds between disclosure and patch.