Severity scores, exploitation forecasts, and incident reports answer different questions. Use them to build a queue around the systems you run and the consequences you need to prevent.
The infrastructure team has one change window left this week. An inventory scan flags a critical vulnerability in a library installed on an isolated reporting service. Meanwhile, a vendor reports active exploitation of a lower-scored issue in the gateway your partners use to submit orders. The gateway owner can restrict access today, although several partners will need help keeping orders flowing. In this fictional scenario, spend the next engineering slot on the gateway and put a named owner on confirming the assumptions about the reporting service.
The useful unit of work is a decision about a deployment. A tracker can group every instance of a vulnerability under one ID. It still has to tell an owner which instance needs action first, what interruption the business can tolerate, and what evidence would change the plan.
Read the scores for what they measure
A Common Vulnerability Scoring System (CVSS) base score describes technical severity using standardized characteristics of the weakness. CVSS v4.0 also defines threat and environmental metrics that can add context. CISA’s Known Exploited Vulnerabilities catalog records vulnerabilities with evidence of exploitation in the wild. An entry tells you that exploitation has occurred somewhere, so check your own deployment and telemetry. The Exploit Prediction Scoring System (EPSS) estimates the probability that exploitation activity for a published CVE will be observed by its data partners over the next 30 days. Its probability describes the observed population and can inform a decision when local evidence is thin. FIRST’s CVSS specification · CISA’s catalog · FIRST’s EPSS FAQ
A historical evaluation shows how the choice of signal can change what makes the cut. In a 2023 study, Jay Jacobs and colleagues ranked vulnerabilities using data available on December 1, 2022, and compared those rankings with exploitation activity observed over the following 30 days. At roughly equal selection sizes, about 15 percent of published CVEs, EPSS version 3 selected 90.4 percent of the vulnerabilities with observed exploitation activity. A ranking based on CVSS v3.x base scores selected 33.5 percent. Figure 4 in the EPSS research paper
That comparison used a historical test period and telemetry from the researchers’ data partners. The unit of effort was a distinct CVE selected, and coverage meant a match with observed exploitation activity. An enterprise budget deals in staff hours and disruption, and the outcomes that matter include compromises and losses avoided. A forecast can make a long list more useful, and the organization still has to decide whether the work at the top protects anything it values. EPSS study methodology and limitations
Give the asset owner a decision they can defend
Start with presence and access. Identify the version actually deployed and check whether the affected feature or component is used. Then establish who can reach it. Public internet access is one path. Partner connectivity, ordinary customer accounts, and access from another service can matter just as much when the target has broad privileges. A version match can justify investigation. An owner who wants to defer a finding needs evidence about the deployment, such as the effective network policy or configuration. The UK National Cyber Security Centre likewise uses exposure to discriminate urgency after confirming an actively exploited vulnerability is present. NCSC guidance on responding to active exploitation
Translate the technical effect into what the system can do for an attacker. The partner gateway can interrupt order intake and has access to the order queue. The reporting service might become just as urgent if its isolation is uncertain or its service identity can modify production data. Give credit to controls you can verify. Revisit a deferral when exposure, threat activity, or the business role changes.
Give private application findings the same hearing. A test might establish that a user with ordinary access can approve a refund outside their assigned authority. There may be no public identifier, no EPSS score, and no catalog entry. The team already has evidence of reachability and a business consequence. If the issue affects the refund workflow in production, it belongs in the same priority conversation as vendor advisories.
Account for work that reduces several exposures at once. If three findings affect the same reachable deployment and one supported upgrade resolves all three, schedule the upgrade as one change and track the coverage it buys. If a temporary control only protects a subset of partner traffic, estimate how much urgent work remains. This is where engineering judgment belongs. The options have different time to benefit, outage costs, and coverage, even when every option carries the same vulnerability identifier in the tracker.
Make an urgent slot available for the right work
When an affected deployment is reachable and the consequence is serious, assign an owner to reduce exposure while a durable fix is prepared. In the gateway example, operations might coordinate a temporary restriction with the partner team and confirm that orders still arrive through an approved channel. If there is credible reason to suspect prior exploitation, investigate in parallel. An update can change future behavior while leaving the effects of past compromise to be dealt with separately. NCSC guidance on containment, investigation, and remediation
For each item that competes for the next slot, record four things in plain English. Which deployment is affected? Who can reach it? What could happen to the business? What action will change that, and by when? Add the strongest evidence for the decision and mark any assumption that could reverse the order. For findings deferred behind more urgent work, name the owner and the event or date that brings them back for review.
The metric to take to a leadership meeting is how long your most consequential, reachable exposures stayed open. Review that alongside a sample of recent deferrals to see whether the underlying assumptions held. If the team spends half its time finding owners and deployed versions, fixing that gap may free more engineering capacity than changing the score threshold again.
Sources and further reading
- FIRST, CVSS v4.0 Specification Document, version 4.0 first published November 1, 2023.
- CISA, Known Exploited Vulnerabilities Catalog, living dataset, accessed November 20, 2026.
- FIRST EPSS Special Interest Group, EPSS Frequently Asked Questions, accessed November 20, 2026.
- Jay Jacobs and colleagues, Enhancing Vulnerability Prioritization: Data-Driven Exploit Predictions with Community-Driven Insights, Workshop on the Economics of Information Security, 2023.
- UK National Cyber Security Centre, Responding to active exploitation of vulnerabilities, May 1, 2026.
Want a second look at your own applications?
We find and verify what’s exposed, fix what matters, and modernize what can’t be secured piece by piece.

