OneNym.idDiscuss deployment
English

THE HARDENED STACK / Chapter 13

Vulnerability Management

A first authenticated scan can return thousands of findings, and treating them all as equally urgent exhausts a small team. This chapter asks how to find the few that matter this week, fix them quickly, and prove the fix held.

This page summarises the chapter. The full text is in the book.

A scan report is raw material, not a to-do list

The chapter opens with Ben Carter at Meridian returning from his first authenticated scan of the server estate with 11,482 findings. Aiko Tanaka's response frames the chapter: the scanner found eleven thousand things, and the team's job is to find the few dozen that actually matter. The author argues that finding vulnerabilities is the easy part, because scanners do it; the skill is ranking them, fixing what matters first, verifying the fix, and reporting progress honestly.

The scope is strictly defensive: your own estate, scanned with your own authorisation. The book notes that most real breaches begin with a known vulnerability for which a fix existed but had not been applied, which makes this loop of find, prioritise, fix and verify one of the highest-value routines a small IT team runs.

Finding: authenticated, complete and regular scans

A scanner compares installed operating systems, software, services and configuration with a database of known flaws. Options include the scanning built into cloud and endpoint tools, the open-source OpenVAS, and commercial products. The most important choice is authenticated scanning, which logs in and sees exact package versions and missing patches, over unauthenticated scanning, which only sees what a host reveals on the network and is better suited to mapping external exposure. Scanner credentials are privileged, so they belong in a secrets vault, scoped to read-only where possible, rotated and monitored.

Coverage decides whether the programme works. Breaches tend to involve forgotten systems such as an old test server, an appliance nobody thinks of as a computer, or a cloud VM built outside the managed environment. The book ties scanning to the asset inventory and treats the comparison between the two, including what the scanner failed to reach, as one of the most useful outputs. Scans should run continuously or at least weekly, should be authorised in writing, and should cover only assets you own or are contracted to assess, with the shared-responsibility boundary checked first for managed services.

Prioritising: CVSS, EPSS, KEV and your own context

CVSS, scored from 0 to 10, describes how severe a flaw could be if exploited. The author argues it answers the wrong question when used alone, since a critical score on an unreachable, unused component can be less risky than a lower score on an internet-facing server under active attack. EPSS adds a frequently updated probability that a vulnerability will be exploited soon, and the CISA Known Exploited Vulnerabilities (KEV) catalogue records flaws confirmed to be exploited in the wild. Anything in your estate that appears on KEV goes to the front of the queue almost regardless of CVSS.

These scores describe the vulnerability, not your exposure to it, so the book adds context only you hold: whether the host is internet-facing and how its data is classified. Its example query weights a KEV listing at 100 points, EPSS multiplied by 40, CVSS multiplied by 3, internet exposure at 20 and restricted or confidential data at 15, then takes the top fifty. The exact weights matter less than the shape. Because KEV and EPSS change, prioritisation is a standing process to re-run, and an absence of known exploitation is not proof of safety.

Fixing: risk-based SLAs, an emergency path and documented exceptions

Patch SLAs give fixing a shared clock. The book's example: KEV findings within 24 to 72 hours, critical or high-EPSS exposed findings within 7 days, high severity without exploitation within 30 days, medium within 60 to 90 days, and low items at the next maintenance cycle or accepted with sign-off. Because a weekly change board is too slow for active exploitation, the emergency path should be agreed in advance: what counts as an emergency, who authorises an out-of-band change, what reduced testing is acceptable, and how to roll back. When a patch is not yet possible, compensating controls such as restricting network access, adding authentication, extra monitoring or taking the service offline shrink the exposure. Staging environments, configuration as code and pre-change backups make prompt patching less frightening.

Unfixed findings need exceptions that record the flaw, affected assets, the reason, the compensating control, a named owner with authority to accept the risk, and a real review date. The author stresses that risk acceptance is a business decision, not something IT should absorb quietly. At Meridian, a VPN appliance that had sat mid-queue appeared on KEV on a Thursday afternoon; the daily ranking query moved it to the top, review showed it was more exposed than the inventory recorded, and the pre-agreed emergency process had it patched within eighteen hours, with a backup taken first.

Verifying and reporting the loop

A finding is closed only when a later authenticated scan confirms it is gone. The book lists common ways a ticket says patched while systems are not: containers still built from an old base image, configuration management reverting a package, and auto-scaling launching instances from an outdated template.

Raw finding counts rise whenever the scanner learns new signatures, so reporting them rewards scanning less. The suggested measures are open KEV findings and the age of the oldest, the share of critical and high findings fixed within SLA, mean time to remediate by risk band, open exceptions past their review date, and the share of the estate scanned in the last seven days. Reported this way, the work becomes visible to the business and supports requests for maintenance windows or staging environments.

KEY POINTS

Key points

  • Run authenticated scans across endpoints, servers, cloud, containers and network devices, with scanner credentials kept in a vault.
  • Compare scan results with the asset inventory and investigate anything the scanner did not reach.
  • Enrich findings with EPSS and the CISA KEV flag and check open findings against KEV today.
  • Write risk-based patch SLAs and agree an emergency change process for actively exploited flaws in advance.
  • Give every unfixed finding a time-boxed exception with a named business owner and review date.
  • Close findings only after a re-scan confirms them gone, and report time to remediate and open-KEV age instead of raw counts.

QUESTIONS

Common questions

What is the difference between authenticated and unauthenticated vulnerability scanning?
An unauthenticated scan looks at a system from the network and sees only what it exposes, which is useful for mapping external attack surface. An authenticated scan logs in with credentials and sees exact installed versions, missing patches and configuration, so it is far more accurate for your own estate. The scanning credentials are privileged and should be kept in a vault, scoped to read-only where possible, rotated and monitored.
Why is CVSS not enough to prioritise vulnerabilities?
CVSS describes how severe a flaw could be in theory, not how likely it is to be exploited against you. A high-scoring flaw on an isolated, unused component can be lower risk than a medium-scoring one being actively exploited on an internet-facing server. The Hardened Stack recommends adding EPSS, which estimates the probability of exploitation, the CISA KEV catalogue of confirmed exploited flaws, and your own context such as exposure and data classification.
What should we do when a vulnerability in our estate appears on the CISA KEV list?
Treat it as an emergency regardless of its CVSS score, because KEV lists flaws confirmed to be exploited in the wild. The book's example SLA is a fix within 24 to 72 hours through a pre-agreed out-of-band change process, with a backup and rollback plan ready. If a patch cannot be applied yet, use compensating controls such as restricting network access or adding monitoring until it can.
What patch SLAs are reasonable for a small IT team?
The Hardened Stack gives an example to adapt: actively exploited (KEV) flaws within 24 to 72 hours, critical or high-EPSS exposed flaws within 7 days, high severity without known exploitation within 30 days, medium within 60 to 90 days, and low or informational items at the next maintenance cycle or accepted with sign-off. The key is tying SLAs to real risk rather than CVSS alone.
How should vulnerability exceptions be handled?
Each exception should record the finding, affected assets, why it is not being fixed now, the compensating control, the named person who accepted the risk, and a real review date. Accepting risk is a business decision made by someone with authority, not something IT should do quietly. Exceptions past their review date should appear in regular reporting, since a KEV listing can make an accepted risk urgent overnight.
Which vulnerability management metrics should we report?
Report open KEV findings and the age of the oldest one, the percentage of critical and high findings remediated within SLA, mean time to remediate by risk band, open exceptions past their review date, and the share of the estate scanned in the last seven days. Avoid reporting the raw total of findings, because it rises whenever scanning improves and so rewards scanning less.