← all systems · see it in The System
Vulnerability management
The cyclical practice of identifying, classifying, prioritising, remediating and mitigating software vulnerabilities across an organisation (src:wp:vulnerability-management); in the research literature, security patch management: identifying, acquiring, testing, installing and verifying security patches (src:dissanayake-2022-slr), framed by NIST as preventive maintenance for technology (src:nist-sp800-40r4). Walked as practitioners and researchers describe it: scanning and inventories, advisory and exploitation feeds, scoring and prioritisation (CVSS, EPSS, KEV, SSVC), the decision and coordination of patching, the programme's cycles, the people who do it, its economics and the measurements of how well it works.
Properties published scienceour proposal
Sourced fields, and our own readings.
| born at | 1999-2003 (CVE public 1999; earliest studies 2002; Patch Tuesday 2003) |
| born precision | year |
| boundary | Inside: the programme's sensing (scanners, inventories, feeds), its findings and ranked queue, the decisions, tests, schedules and deployments of fixes and mitigations, the verification after, and the people and cycles that run it. Outside: the vulnerabilities themselves (their own system), vendors who write patches, attackers, the SOC and incident responders, regulators, and the rest of the organisation it serves. |
| nested level | A process inside an organisation's security function, next to the SOC and incident response; fed by the shared vulnerability record and vendors' patches. |
| scale | social |
Measures (29) published science
| characteristic clock: patch cycle (Patch Tuesday, second Tuesday each month) | 1 month |
| characteristic clock: patch coordination meeting interval | 14 (about 1.5 hours each) days |
| standard patch window | 4 hours |
| standard phase time frames (retrieval, scan/assess, test, deploy, verify) | 2 days; 1 week; 1 week; 2 weeks; by next cycle time |
| delayed patching tasks (healthcare, four years) | 132 of 232 tasks |
| share of delays at deployment | 54 % |
| longest average delay per phase | 3.6 months |
| organisations unable to patch critical flaws within 72 hours | more than 50 (about 15% unpatched after 30 days) |
| share of patch-management solutions rigorously evaluated in industry | 20.8 % |
| share of CVEs exploited in the wild within 12 months (2016-2018) | 3.7 (921 of 25,159) % |
| remediation efficiency of a CVSS 7+ rule | 6.2 (coverage 62.7) % |
| effort saved at equal coverage (EPSS vs CVSS 7+) | 60.7 (2,773 vs 1,091 fixes) % |
| median time to remediate (almost 300 companies) | 100 (75% within 392) days |
| known vulnerabilities exploited in the wild | about 5 % |
| EPSS v3 improvement over past models | 82 % |
| CVEs published per year | 25,059 (2022); 28,961 (2023); 29,004 (2024) CVEs |
| CVEs whose EPSS exceeded 50% before KEV listing | fewer than 20 % |
| enterprise time to patch 90% of client hosts (average) | over 6 (avg 67 days to 50%, 200 to 90%) months |
| Chrome enterprise patch time | 18 days to 50%; 78 days to 90% |
| server application vulnerability window (average) | 7.5 (over 9 months to patch 90%) |
| servers fixed one week after the OpenSSL announcement (2002) | 23 % |
| share of operators who patch immediately | about 35 (65 take no action) % |
| knees in bad-patch risk after release | 10 and 30 days |
| revised or pulled security patches (1999-2001) | 20 of 112 patches (136 CVE entries, 24 without a patch) |
| time to generate an exploit from a patch | a few minutes |
| Heartbleed: increase in patching after notification | 47 (150,000 hosts) % |
| Heartbleed: vulnerable Top-1M HTTPS sites two months after disclosure | 3 (plateau after about two weeks) % |
| Microsoft updates, first quarter 2013 | over 300 updates |
| home users who misunderstood their update settings | 28 of 37 users |
Open contentions (4) open contention
Where sources or readings disagree. Both stay on the map.
- C-1632 Do companies patch faster than home users?
booked: The map books the reported difference and notes that the two data sets come from different years.
the other reading: the difference comes from study periods (2008-2013 vs 2015-2017), not enterprises - C-1633 When did vulnerability management begin?
booked: The map dates it to 1999-2003 (shared flaw IDs, first studies, Patch Tuesday), with earlier patching and the 2005 scoring system kept alongside.
the other reading: patches on paper tape and punched cards before networked security; CVSS 2005 as the start of systematic prioritisation - C-1634 Should you patch at once or wait?
booked: The map keeps both pressures, and books the best time as where the risk of attack and the risk of a bad patch cross.
the other reading: patch early and often (72-hour recommendation); wait for the bad-patch risk to fall (10 or 30 days, Beattie 2002) - C-1635 Which score best picks the flaws to fix first?
booked: The map books each score with its study's result and ranks none, since the main scoring systems barely agree.
the other reading: EPSS beats CVSS: same coverage with 60.7% less effort (Jacobs 2019), 82% better in v3 (Jacobs 2023); EPSS rarely above 50% before KEV listing, CVSS higher for later-exploited CVEs; four systems barely agree (Koscinski 2025)
Partners and parts published science
Built from / builds (7)
- Competition for control of a running program (exploit vs patch)
- Human (Homo sapiens)
- Common Vulnerability Scoring System
- Security patch
- Software bill of materials
- Vulnerability scanner
- Patching against uptime inside the organisation
Sits inside / holds (2)
Trades with (valence) (14)
- Risk management via sys:software-vulnerability · Vulnerability management gain
- Human (Homo sapiens) via sys:computer · Vulnerability management gain
- Government via sys:law · Vulnerability management mixed
- Operating system via sys:software · Vulnerability management gain
- Insurance via sys:money · Vulnerability management mixed
- Security operations center (SOC) via sys:organization · Vulnerability management gain
- Incident response via sys:organization · Vulnerability management gain
- Threat actor via sys:internet · Vulnerability management mixed
- Organization via sys:organization · Vulnerability management gain
- Software vulnerability via sys:software · Vulnerability management gain
- Secure software development via sys:software · Vulnerability management gain
- Security information and event management (SIEM) via sys:organization · Vulnerability management gain
- Cyber threat intelligence via sys:internet · Vulnerability management gain
- Penetration testing via sys:organization · Vulnerability management gain
expressed by (17)
- Templating / Standardization
- Amplification/Scale-Up (Manufacture & Distribution)
- Internal Coordination/Regulation
- Sensing/Scanning
- Uncertainty Reduction/Learning
- Active Experimentation / Probing (Optimal Experimental Design)
- Multi-Scale Data Fusion / Triangulation
- Trail-Marking / Externalized Memory (Stigmergy)
- Social Learning / Public Information Use
- Reactive Response
- Proactive Defense
- Physical/Virtual Barrier
- Self-Monitoring
- Validation of Effectiveness
- Enclave/Segregation
- Triage & Prioritized Allocation
- Quality Control & Error Correction
floor filled by (11)
- Risk management
- Human (Homo sapiens)
- Government
- Software
- Computer
- Organization
- Common Vulnerability Scoring System
- National Vulnerability Database
- Fuzzing
- Exploit Prediction Scoring System
- Vulnerability scanner
Kinds (6)
- Risk management
- Security patch management
- Risk-based (exploit-driven) vulnerability management
- Compliance-driven patching
- End-user software updating
- Patching in critical infrastructure and healthcare
Notes published science
- Patch everything scored CVSS 7 or above and only about 6 of every 100 fixes land on a flaw that attackers actually use (Jacobs et al. 2019: efficiency 6.2%, coverage 62.7%). An exploit-prediction score matched that coverage with 60.7% less work, though a 2025 study found the four main scoring systems barely agree on the same CVEs (src:jacobs-2019-epss, src:koscinski-2025-scores).
- A patch is also a map of the hole: researchers built working exploits from Windows patches in a few minutes, which is why the day after Patch Tuesday is called Exploit Wednesday (src:brumley-2008-apeg, src:wp:patch-tuesday).
- Your cells do vulnerability management too: one repair system scans the whole genome, while another fixes damage on the genes being read first, as patch teams scan everything but fix exposed, in-use systems first. Emergency medicine runs the same trade, accepting up to 35% over-triage to keep missed serious injuries at 5% or less (src:rastogi-2010-uv, src:newgard-2022-triage).
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
the boardroom (shared floors)
- L4: Assessment and prioritisation: findings ranked by severity (CVSS), exploitation (EPSS, KEV), weakness type (CWE) and asset context; the ranked queue
- L5: Decision and coordination: the risk response chosen (accept, mitigate, transfer, avoid), scheduled through change management, tested, approved; patch meetings across teams, customers and vendors
- L6: The programme: maintenance groups and plans, patch cycles (monthly, Patch Tuesday), standard time frames, metrics and policy
- L7: Purpose: preventive maintenance so the organisation's technology reliably serves its mission; a cost of doing business
vulnerability management as practitioners and researchers describe it: the organisation's assets and their flaws, up to the mission the maintenance keeps (reader's view)
- L1: The estate: the organisation's assets (computers, servers, applications, libraries, firmware) and the flaws in them
- L2: Sensing: scanners, inventories and SBOM tools, advisory feeds, exploitation data, fuzzing
- L3: Findings: known vulnerabilities matched to assets and versions; filtered and validated
- L4: Assessment and prioritisation: CVSS, EPSS, KEV, SSVC, asset context; the ranked queue
- L5: Decision and coordination: risk response, testing, scheduling, approvals, deployment
- L6: The programme: patch cycles, maintenance groups, time frames, policy and metrics
- L7: Purpose: preventive maintenance for the mission
way in: Advisories and patch releases from vendors and the shared record
- L1: Vulnerabilities disclosed and patches released by vendors (Patch Tuesday bulletins, the Security Update Guide), recorded as CVEs
- L2: Security advisories (78%), vendor notifications (71%), mailing lists, forums, news, blogs, social media
- L3: Filtering and validating against what the organisation runs: no central platform, hours spent monitoring sources
way in: Business context: asset criticality, downtime limits, approvals
- L1: Business units, customers (hospitals) and end users whose services depend on the assets
- L2: Their constraints arrive: approvals for patch windows, change freezes, minimum-interruption policies
- L3: Asset criticality and operational constraints as contextual metrics
way in: Exploitation evidence: who is attacking which flaws
- L1: Exploitation in the wild by attackers, and exploit code published (Exploit DB, Metasploit modules)
- L2: Sensors of security vendors (Proofpoint, Fortinet, AlienVault, GreyNoise) record exploitation attempts
- L3: Turned into scores and lists: EPSS probability of exploitation in 30 days, the CISA KEV catalogue
way in: Incidents and breaches through unpatched flaws
- L1: Attacks exploiting known, already patchable flaws (Equifax; WannaCry stopped 45 NHS hospital groups)
- L2: Alerts and investigations by the SOC and incident responders
- L3: Out-of-band and emergency patching scenarios triggered
way in: Inventory: what software runs where, down to packages and libraries
- L1: The organisation's applications, operating systems, firmware and their packages and libraries
- L2: Asset inventory systems and SCA/SBOM tools for third-party components
- L3: Versions matched to known vulnerabilities; re-bundled open-source code hides in other dependencies; inventories are often incomplete
way in: Looking for unknown flaws: fuzzing, static analysis, testing
- L1: The organisation's own and third-party programs, with flaws nobody has reported
- L2: Fuzzers (AFL, libFuzzer) feed random or semi-random inputs; static analysis and symbolic execution read code without running it
- L3: Crashes and warnings triaged into candidate vulnerabilities
way in: Mandates: standards, regulators, auditors
- L1: Payment-card standard and federal directives that set CVSS thresholds and deadlines
- L2: PCI-DSS requires flaws above CVSS 4.0 remediated; DHS BOD 19-02 requires high and critical flaws remediated
- L3: Legal and regulatory requirements as motivators to patch
way in: Notifications from outsiders who scanned you
- L1: Researchers scan the whole IPv4 space for vulnerable hosts
- L2: Operators contacted by email through abuse contacts
- L3: Notified operators patch more (a 47% increase)
way in: Own patch state after deployment
- L1: Assets after deployment: patched, failed, or rolled back
- L2: Confirming the patch is installed and has taken effect
- L3: An overview of the patch state; often missing and manual
- L4: Continuous monitoring that no one undoes the response
way in: Patch quality and vendor support
- L1: Vendors' patches, some faulty: 20 of 112 patches revised or pulled in 1999-2001
- L2: Announcements that a patch was bad; vendor support tickets
- L3: Patches tested in a test environment before production; manual, slow and hard to make identical to production
way in: Scanning the estate: hosts, ports, services and installed software
- L1: The organisation's computers, servers and network devices, with open ports, services and configurations
- L2: Vulnerability scanners probe them: authenticated scans log in over SSH or RDP for low-level detail; unauthenticated scans look from outside
- L3: Findings: known vulnerabilities, misconfigurations and missing security patches per asset; unauthenticated scans give many false positives
way in: The team's own state: workload, expertise, fatigue
- L1: Administrators, security engineers, change managers
- L2: Overload, leave, missing expertise hold tasks; 67% feel they lack time to mitigate all vulnerabilities
- L3: Coordination load: delays in coordinating patch work are the top cause of delay
- L4: Burnout across the industry
way out: Avoid: uninstall, decommission, replace
- L1: The vulnerable attack surface gone
- L2: Software uninstalled, asset decommissioned or replaced (e.g. end-of-life Windows XP)
- L5: Avoidance chosen
way out: Mitigate without patching: controls and reconfiguration
- L1: The attack surface around the asset reduced
- L3: Disable the vulnerable feature, change network policy, segment, firewall
- L5: Mitigation chosen when no patch is viable
way out: Patch: install the fix
- L1: The flaw removed from the asset without removing function
- L2: Distributed and installed, automatically or by hand, often with reboots
- L3: Acquired, authenticity validated, tested
- L4: Assigned to maintenance groups and patch windows (usually 4 hours)
- L5: Patch prioritised and scheduled through change management
way out: Reports and compliance evidence
- L1: Leadership and customers informed
- L6: Patch status and progress reported in patch meetings and to management; compliance kept up
way out: Transfer: insurance or SaaS
- L1: Cyber insurance bought, or SaaS adopted so the provider patches
- L5: Transfer chosen
way out: Vulnerability context handed to the SOC and incident responders
- L1: Responders know which assets are exposed
- L4: Asset and vulnerability lists shared (SIEM/SOAR integration)