Blog/Vulnerabilities

Vulnerability Management: Turning a Scan Report Into a Process

Why vulnerability counts never go down, how to prioritize with CISA KEV and EPSS instead of raw CVSS, and how to verify a fix actually landed.

CyberWatch AI18 November 2026 · 7 min read
A spreadsheet of numbers open on a laptop screen beside a notebook and pen

A scanner report lands in an inbox: 14,206 findings, 1,840 of them rated critical or high. The person who receives it has a day job. Nothing in the report says which of those 1,840 an attacker could actually reach, which ones are being exploited this week, or which server matters to the business. So the report gets filed, and three months later the next one arrives with 15,900 findings.

This is the normal state of vulnerability management in most organizations, and it is a prioritization failure rather than a tooling failure. Scanning is solved. Deciding what to do with the output is not.

What follows is the process in five stages, with most of the attention on the stage that actually determines whether any of it works.

Stage one: know what you own

Every vulnerability management program is capped by its asset inventory. A scanner can only report on what it is pointed at, and the systems nobody remembers are consistently the ones that get compromised: the test server from a project that ended, the appliance a departed administrator installed, the forgotten subdomain still pointing at live infrastructure.

A workable inventory does not need to be a formal asset database. It needs to answer four questions for everything you run: what is it, who owns it, is it reachable from the internet, and what data does it hold or touch. Build it from several sources, because no single source is complete. Cloud provider APIs list what exists in your accounts. Your identity provider lists what people log into. DNS records reveal names pointing at infrastructure. Network discovery finds devices nobody registered. External attack surface scanning shows what the internet can see, which is often a longer list than anyone expected.

The fastest useful version of this is to enumerate internet-facing services first, since that is where indiscriminate exploitation lands, and the list is usually short enough to review by hand.

Stage two: find the weaknesses

Several different activities get called scanning, and they find different things.

  • Network and host scanning identifies software versions and known vulnerable configurations. Authenticated scans, where the scanner logs in, are substantially more accurate than unauthenticated ones, which infer versions from banners and guess wrong in both directions.
  • Software composition analysis inspects the dependencies your applications pull in. This is where most application vulnerabilities actually live, and it is covered further in our guide to supply chain security.
  • Container and image scanning checks base images and installed packages. Findings here are often fixed by rebuilding from a current base image rather than patching anything.
  • Cloud configuration assessment finds misconfigurations rather than software flaws, and those are a larger cause of cloud incidents than vulnerable software. See the cloud security guide.
  • Penetration testing finds logic flaws, chained weaknesses and business-process problems that no scanner detects. It answers a different question and does not replace continuous scanning.

Coverage beats precision at this stage. A tool that covers everything you own with some false positives is more useful than an excellent tool covering two thirds of it.

Stage three: prioritization, where the work is decided

This is the stage that separates a program from a report, and the common approach of sorting by CVSS score is the reason so many programs stall.

CVSS measures severity from the technical properties of the vulnerability itself. The specification is explicit that the base score is not a risk score and is not intended to be used alone for prioritization. It knows nothing about whether you run the software, whether it is reachable, whether anyone is exploiting it, or what the affected system does. Sorting 1,840 high-severity findings by a number that ignores all four of those produces a work queue with no relationship to risk.

Four signals, combined, give a defensible order.

SignalQuestion it answersWhere it comes from
Exploitation evidenceIs this being used against real systems?CISA KEV catalog, vendor advisories
Exploitation likelihoodHow probable is exploitation soon?EPSS daily scores
ExposureCan an attacker reach the affected system?Your own asset inventory
Impact if exploitedWhat does this system do and hold?Business context, not the scanner

The KEV catalog is the strongest free signal available. Its inclusion criteria are strict: an assigned CVE ID, reliable evidence of active exploitation in the wild, and a clear remediation action. Notably, a publicly available proof-of-concept is not sufficient for inclusion, and neither is scanning activity. If something you run appears in KEV, it goes to the front of the queue regardless of its CVSS score.

EPSS fills the gap for everything not yet in KEV. It publishes a probability between 0 and 1, updated daily for every published CVE, that the vulnerability will be exploited in the wild within the next 30 days. Because it is derived from observed activity rather than expert judgment, it changes as the real world changes: a CVE can sit at a low score for months and climb sharply when exploit code circulates. The practical use is as a filter. The overwhelming majority of CVEs score very low, which means a threshold lets you defer most of the backlog with a defensible reason rather than ignoring it silently.

Exposure is the signal only you can supply, and it is often the largest discriminator. The same CVE on a public API gateway and on a laptop behind a VPN is not the same problem.

If you want a published framework rather than a home-made scoring sheet, CISA's Stakeholder-Specific Vulnerability Categorization model turns these inputs into a decision tree producing one of four actions: track, track closely, attend or act. The federal directive on prioritizing security updates uses the same logic, setting remediation timelines from whether a flaw is in KEV, how automatable exploitation is, the technical impact, and whether the affected asset is publicly exposed. The most urgent category requires action within three days plus a forensic check of whether the asset has already been compromised. The specific deadlines bind US federal agencies, but the structure transfers to any organization.

Stage four: fix it

Remediation has four possible outcomes, and only one of them is patching.

  1. Patch. The default, and the only option that removes the flaw.
  2. Mitigate. Apply a vendor workaround, disable the affected feature, or restrict who can reach the service. Treat every mitigation as temporary and give it an expiry date, because workarounds get bypassed and then forgotten.
  3. Remove. Often the cheapest fix. Software nobody uses, services exposed for no current reason and systems kept "just in case" can be decommissioned instead of maintained.
  4. Accept. Legitimate when a fix would break something essential, and only with a named owner, a documented reason, a compensating control and a review date.

NIST's guide to enterprise patch management planning makes a point worth repeating to anyone treating patching as a purely technical task: most of the difficulty is organizational. Maintenance windows, change approval, systems nobody will authorize downtime for, and software whose vendor no longer ships updates are what actually block remediation. Deciding those policies in advance, including a standing emergency path for exploited vulnerabilities, removes the negotiation from the moment it is least affordable.

Two practical notes. First, automate the routine cases. Operating system and browser updates on end-user devices should flow automatically with a rollback plan, because the risk of an occasional bad patch is lower than the risk of a three-month lag; our endpoint security guide covers the device side. Second, patch at the source where you can. Rebuilding container images from a current base, or bumping a dependency version, fixes every instance at once instead of each deployed copy individually.

Stage five: verify and measure

A deployment tool reporting success is not verification. Patches fail silently, services do not restart, containers get redeployed from a cached old image, and a reboot nobody performed means the running kernel is still the vulnerable one. Rescan the specific asset and confirm the finding is gone.

Then measure the few things that reflect reality. Total open findings is not one of them; it mostly tracks how much you have expanded scanning coverage.

  • Time to remediate vulnerabilities known to be exploited, measured from the date they appeared in KEV. This is the number that correlates with not getting compromised.
  • Percentage of the estate actually scanned, which tells you how much of your confidence is justified.
  • Age of the oldest unremediated finding on an internet-facing system. A single uncomfortable number that is hard to argue with.
  • Number of accepted risks past their review date, which is where exceptions go to become permanent.

Our guide to measuring security progress covers how to present these without turning them into targets that get gamed.

Patch notifications are a phishing theme. Fake "critical update required" emails spike after major vulnerability news. Paste anything suspicious into CyberWatch AI Scan for a free check, and download updates only from the vendor's own site.

What this looks like when it works

CISA's advisories on routinely exploited vulnerabilities keep returning the same finding year after year: attackers get most of their results from vulnerabilities that have had patches available for a considerable time. That is a reassuring conclusion, because it means the organizations doing this well are not the ones with the best tooling. They are the ones who know what they own, can say which systems face the internet, check exploitation evidence before arguing about severity scores, have a route to deploy an emergency patch within a day, and track their exceptions honestly.

None of that requires a large team. It does require accepting that the backlog will never reach zero, and that deciding deliberately what to leave undone is the actual skill. For how this connects to flaws with no patch available yet, see our guide to zero-day vulnerabilities.

Frequently asked questions

What is vulnerability management?

The ongoing process of finding weaknesses across the systems you run, deciding which ones matter most, fixing them, and confirming the fix worked. It is distinct from vulnerability scanning, which is only the first step and the easiest one.

Why does my vulnerability count never go down?

Because new findings arrive faster than most teams close old ones. Every patch cycle from every vendor adds new records, and the scanner finds more as you expand coverage. A flat or rising total is normal, which is why the count is a poor measure of progress.

Should I just patch everything with a CVSS score above 7?

That approach generates far more work than it prevents harm. CVSS describes the severity of a flaw in the abstract, not your exposure to it. A high-scoring vulnerability in software that is not reachable can matter much less than a medium-scoring one on your public login page.

What is the CISA KEV catalog?

A list maintained by the US Cybersecurity and Infrastructure Security Agency of vulnerabilities with reliable evidence of exploitation in the wild. Entries need an assigned CVE ID, that evidence, and a clear remediation action. It is the strongest free signal that something is actually being used against real systems.

What is EPSS and how is it different from CVSS?

The Exploit Prediction Scoring System publishes a daily probability between 0 and 1 that a published CVE will be exploited in the wild within the next 30 days. CVSS tells you how bad exploitation would be; EPSS estimates how likely it is. They answer different questions and work best together.

How fast should we patch?

Set different targets by exposure and evidence rather than one number. Internet-facing systems with a vulnerability known to be exploited need action in days at most; internal systems with no exploitation evidence can follow a normal monthly cycle. Published federal timelines use exactly this logic.

What if we cannot patch something?

Record it as an accepted risk with a named owner, an expiry date and a compensating control such as restricting network access, removing the system from the internet, or increasing monitoring on it. An untracked exception is the same as having no process.

Do we need a commercial scanner?

Eventually, but asset inventory matters more. A perfect scanner pointed at 60% of your estate produces a confident, incomplete picture, which is worse than an imperfect scanner that covers everything you own.

How do we know a patch actually worked?

Rescan the specific asset and confirm the finding is gone, rather than trusting that the deployment tool reported success. Patches fail to apply, services do not restart, and containers get rebuilt from an old base image, so verification is a separate step.

Sources

  1. SP 800-40 Rev. 4, Guide to Enterprise Patch Management Planning, US National Institute of Standards and Technology
  2. Known Exploited Vulnerabilities Catalog, US Cybersecurity and Infrastructure Security Agency
  3. Reducing the Significant Risk of Known Exploited Vulnerabilities, US Cybersecurity and Infrastructure Security Agency
  4. BOD 26-04: Prioritizing Security Updates Based on Risk, US Cybersecurity and Infrastructure Security Agency
  5. Stakeholder-Specific Vulnerability Categorization (SSVC), US Cybersecurity and Infrastructure Security Agency
  6. Using EPSS, FIRST
  7. CVSS v4.0 Specification Document, FIRST
  8. Vulnerability management, UK National Cyber Security Centre
  9. CIS Critical Security Control 7: Continuous Vulnerability Management, Center for Internet Security
  10. 2023 Top Routinely Exploited Vulnerabilities, US Cybersecurity and Infrastructure Security Agency
Share this article
LinkedIn WhatsApp X
For organizations

Your people get these messages at work too.

CyberWatch AI sends your staff realistic practice attacks, trains the gaps it finds, and shows management exactly where the organization stands.