Building a patch-priority queue
Turning scanner exports into a ranked, owned queue: deduplicating findings, adding asset criticality and exposure, a transparent scoring rule built from KEV, EPSS and CVSS, SLAs and exceptions, and writing to system owners in terms they act on.
01Inputs: findings, assets and threat data
- Normalise and deduplicate scanner findings into fixable units
- Join asset context: owner, exposure, criticality
- Build a transparent priority rule from KEV, EPSS, CVSS and context
- Set SLAs and exceptions, and communicate risk to owners
A raw scanner export is the wrong unit of work. One missing cumulative update can appear as forty CVEs on four hundred hosts; one outdated Java runtime can appear in three products. Owners fix packages on systems, so reshape findings into that unit before you rank anything.
| Input | Source | Used for |
|---|---|---|
| Findings | Qualys, Tenable, Rapid7, Defender VM, container and cloud scanners | CVE, host, package, fixed version |
| Asset inventory / CMDB | CMDB, cloud tags, EDR device inventory | Owner, environment, business service, criticality |
| Exposure | External attack-surface scans, firewall rules, load balancer config | Internet-facing or reachable from untrusted networks |
| Threat | CISA KEV, EPSS, vendor advisories, intelligence | Likelihood and urgency |
| Controls | WAF virtual patches, EDR coverage, segmentation | Compensating factors that can justify a later date |
- ACTOR
- SERVICE
- DATA
- STREAM
- Live flow
- Lookup / async
Findings on assets with no owner are where breaches live. Track the share of findings without an owner as a metric of its own, and route them to whoever runs the CMDB, not to a shared mailbox.
02A transparent priority rule
Use a rule people can read and argue with, not a vendor's opaque risk number. A tiered rule is easier to defend than a formula, because every finding's priority comes with a reason.
from dataclasses import dataclass
@dataclass
class Finding:
cve: str
host: str
cvss: float | None # base score from your chosen source
epss: float | None # probability, not percentile
in_kev: bool
ransomware: bool # KEV knownRansomwareCampaignUse == "Known"
internet_facing: bool
criticality: str # "high" | "medium" | "low" from the CMDB
def priority(f: Finding) -> tuple[str, str]:
if f.in_kev and f.internet_facing:
return "P1", "Known exploited and reachable from the internet"
if f.in_kev and (f.ransomware or f.criticality == "high"):
return "P1", "Known exploited on a critical asset or used by ransomware"
if f.in_kev:
return "P2", "Known exploited"
if (f.epss or 0) >= 0.1 and (f.internet_facing or f.criticality == "high"):
return "P2", "Likely to be exploited soon on an exposed or critical asset"
if (f.cvss or 0) >= 9.0 and f.internet_facing:
return "P2", "Critical severity on an internet-facing asset"
if (f.cvss or 0) >= 7.0 or (f.epss or 0) >= 0.1:
return "P3", "High severity or elevated likelihood"
return "P4", "Routine: fix with standard updates"| Priority | Target time to remediate | Route |
|---|---|---|
| P1 | 72 hours (internet-facing KEV: 24 to 72 hours) | Emergency change, named owner, daily check-in |
| P2 | 14 days | Next change window, tracked weekly |
| P3 | 30 to 60 days | Normal patch cycle |
| P4 | 90 days or next platform refresh | Routine updates and image rebuilds |
The times above are typical, not a standard. Agree them with risk owners in advance and write them into policy, so that a P1 is not renegotiated in every incident call.
03From findings to a queue
index=vuln sourcetype=tenable:sc:vuln state=open
| lookup kev_catalog.csv cveID AS cve OUTPUT dateAdded knownRansomwareCampaignUse
| lookup epss_current.csv cve OUTPUT epss
| lookup cmdb_assets.csv host OUTPUT owner service criticality internet_facing
| eval in_kev=if(isnotnull(dateAdded), 1, 0)
| eval priority=case(
in_kev=1 AND internet_facing="yes", "P1",
in_kev=1 AND (knownRansomwareCampaignUse="Known" OR criticality="high"), "P1",
in_kev=1, "P2",
epss>=0.1 AND (internet_facing="yes" OR criticality="high"), "P2",
cvss>=9.0 AND internet_facing="yes", "P2",
cvss>=7.0 OR epss>=0.1, "P3",
true(), "P4")
| stats dc(host) AS hosts values(cve) AS cves max(cvss) AS max_cvss max(epss) AS max_epss min(priority) AS priority BY owner service plugin_name fixed_version
| sort priority - hostsThe final stats is the important step: it turns thousands of rows into one line per fix per owner, such as "Payments team: upgrade OpenSSL to 3.0.15 on 38 hosts, covers 4 CVEs, P2". min(priority) works because the labels sort P1 before P4. That line is something a team can put in a sprint.
- Measure outcomes: mean time to remediate by priority, share of P1s closed within SLA, and KEV exposure (open KEV findings on internet-facing assets). Count of open findings on its own mostly measures scanner coverage.
- Exceptions need an owner, a reason, a compensating control (WAF rule, isolation, disabled feature) and an expiry date. An exception without an expiry is a permanent acceptance nobody signed.
- Verify fixes with a rescan or agent data, not with the change ticket's status.
04Communicating risk to owners
System owners are busy and outnumbered by findings. A message they act on names the asset, the risk in business terms, the fix, the deadline and the reason for the deadline. It does not paste a CVSS vector.
Subject: [P1, due Thu 19 Nov 18:00 UTC] Patch the GlobalProtect firewalls (fw-edge-01, fw-edge-02)
What: Both internet-facing firewalls run a PAN-OS version with a flaw (CVE-2024-3400) that lets anyone
on the internet run commands on the firewall without logging in.
Why now: It is being actively exploited (listed by CISA as known exploited) and the devices are exposed.
A compromised firewall gives an attacker a foothold inside our network and our VPN credentials.
Fix: Install the vendor hotfix listed in the advisory, in tonight's emergency window if possible.
Change CHG-20431 is pre-approved.
Also: Security will run the vendor's compromise checks before and after patching; please keep the
current tech-support file.
Contact: SOC on-call, ext 4400. If you cannot meet the date, reply with the blocker so we can agree a
compensating control today.05Practice
Two findings: A is CVSS 9.8, EPSS 0.004, not in KEV, on an internal test server. B is CVSS 7.5, in KEV with known ransomware use, on an internet-facing file transfer server. Which goes first and why?
"I reshape findings into fixes per owner, enrich them with KEV, EPSS, exposure and criticality, and rank them with a tiered rule where every priority carries a reason. Owners get one message per fix that names the asset, the business impact, the deadline and why. I measure time to remediate and exposed KEV findings, not raw counts."
Interviewers often hand over a small table of findings and ask you to order them. Say your rule out loud (KEV, then exposure and criticality, then EPSS, then CVSS), justify the top item in business terms, and mention exceptions with expiry dates.
Q1What if an owner refuses to patch?+
Escalate through the risk process: document the risk, a compensating control and an expiry, and get the accountable executive to accept it in writing. Security advises; the business owns the risk.
Q2How do you handle findings with no CVE?+
Configuration findings and end-of-life software get priority from exposure and impact, plus vendor guidance. End-of-life with no fix is a replacement project, not a patch.
Q3Why not let the scanner's risk score decide?+
You cannot explain it to an owner or auditor, it may not use your asset context, and it changes when the vendor changes the model. A readable rule is defensible.
Related lessons
- Design a vulnerability management platformHLD · Design problems: Infrastructure & dataPro
- Design a CVE intelligence feedHLD · Design problems: vulnerability and exposurePro
- CISA KEV and what "known exploited" meansAnalyst · Vulnerability triage & CVE prioritisationFree
- Incident response engineeringSecEng · Detection & Response EngineeringPro