EVOPEDIA

← all systems · see it in The System

Software vulnerability

In computer security, a vulnerability is a flaw or weakness in a system's design, implementation, or management that can be exploited by a malicious actor to compromise its security (src:wp:vulnerability-computer-security). Walked as the flaw together with its life as the research literature describes it: introduced with code (or hardware), active while that code runs, discovered by a vendor, a third party or an attacker, disclosed or kept secret or sold, catalogued (CVE, CWE, CVSS), patched or exploited, and ended when patched or retired; with the markets (bounties, brokers) and rules (coordinated disclosure) around it.

Properties published scienceour proposal

Sourced fields, and our own readings.

born at1970 (the Ware report names inadvertent loopholes in protective barriers as the most serious risk in system software); 1972 the buffer overflow described (Computer Security Technology Planning Study); 1972-73 the Multics penetration exercise; earliest documented hostile exploitation 1988
born precisionyear
boundaryInside: the flaw itself (a weakness in design, implementation, configuration or hardware behaviour that an input can trigger to compromise confidentiality, integrity or availability), its states (carrier, dormant, active; undisclosed or disclosed; patched or not), the acts that find it (code review, static analysis, fuzzing, telemetry, reverse-engineering patches), the routes it travels once known (reports, coordinated or full disclosure, sale), the shared record and scores (CVE identifiers and records, CWE root causes, CVSS severity, NVD, exploitation probability), and the acts that end it (patch, mitigation, retirement) or use it (exploit, stockpile). Outside: the software, hardware and operating systems that carry the flaw; the people who write, find, buy, sell and fix; the malware an exploit delivers; computer security and information security, the practices that manage it; the markets and laws it sits in.
nested levelA part of software (and hardware) that exists only while the code runs or can run (rung 12); its catalogue and market are institutions carried by people and organisations. Managed inside computer security's vulnerability management; exploited by threat actors, often to deliver malware.
scaletechnological

Measures (40) published science

characteristic clock: average vulnerability lifetime in code~4 (Chromium ~2, OpenSSL ~7) years
vulnerabilities living more than three yearsabout one third fraction of CVEs
median life span of a vulnerability in its code base (682 projects)438 days
disclosed issues still unpatched 30 days after disclosure26 %
median lines changed: security commit vs general bug fix7 vs 16 lines of code
duration of zero-day attacks before disclosure (mean)312 (range 19 days to 30 months)
rise in attack volume after disclosureup to 10^5 factor
median lifetime of OpenBSD foundational vulnerabilitiesat least 2.6 years
vulnerability half-life, highest depletion estimate (Rescorla)~2.5 years
vulnerability density, new code against 5-year-old code3.4x to 7.4x higher in new code ratio
share of high-severity security bugs that are memory-safety bugs (Chromium)~70 %
Heartbleed: Alexa Top 1 Million HTTPS sites still vulnerable11 after two days; 3 almost two months after %
memory-safety share of Android vulnerabilities76 (2019) to 24 (2024) %
Heartbleed: share of Alexa Top 1 Million HTTPS sites initially vulnerable24-55 %
patching increase among notified operators (Heartbleed)47 %
per-hand-off yield after Heartbleed: sites vulnerable at 48 h that replaced certificates within a month AND changed the key0.10 x 0.86 = ~0.086 fraction (computed here from two sourced shares)
package releases affected per vulnerability (median)30 (npm), 59 (RubyGems) releases
exposed dependency vulnerabilities with a fix in the same major range33 (projects), 40 (packages) %
Chrome VRP cost and bounties (3 years)~$580,000; 501 bounties USD; count
Firefox VRP cost and bounties (3 years)~$570,000; 190 bounties USD; count
share of patched browser vulnerabilities from VRP reports28 (Chrome), 24 (Firefox) %
bounty discovery decay after programme launch~t^-0.40 power-law exponent
bounty discoveries per researcher per programme (tail)P(X>x) ~ 1/x^1.63 power-law exponent
time to remediate (median, about 300 companies)100 (75% within 392) days
share of EPSS predictions at 2% or less76 %
vulnerabilities recorded in the CVE database>327,000 (April 2026) records
CVE records in the CWE Top 25 dataset (2024)31,770 records
usable life of a zero-day exploit (RAND, mean)6.9 (1.4 if bought from a third party) years
rediscovery (collision) rate of a zero-day stockpile per year5.76 (RAND: a private set of 207 zero-days over 14 years, as reported by Herr et al.; 5.7 as quoted by the encyclopedia); 10.8-21.9 (Herr, Schneier and Morris 2017, revised: duplicate disclosures in vendor bug trackers, Chrome 2009-2017 10.8, Android 2016-2017 21.9, aggregate 12.7) %
bugs and security-critical bugs in a 35-million-line program (model)1,000,000 bugs; ~1% security-critical count; fraction
Unix utilities crashed by random input (1988-90)25-33 %
CVSS score range and bands0.0-10.0; Low 0.1-3.9, Medium 4.0-6.9, High 7.0-8.9, Critical 9.0-10.0 score
coordinated disclosure deadline (common)90 days
Heartbleed: time in code before disclosureabout 2 (OpenSSL 1.0.1 released 2012-03-14; disclosed 2014-04-07) years
honeymoon: release to first disclosed vulnerability (median, 700 releases)110 (open source 115, closed source 98); median honeymoon ratio 1.54 days; ratio
patch release: instantaneous probability with disclosure against withoutnearly 2.5 ratio (hazard)
CWE Top 25 (2024): top-ranked weaknesses1 CWE-79 cross-site scripting (56.92); 2 CWE-787 out-of-bounds write (45.20); 3 CWE-89 SQL injection (35.88); 8 CWE-416 use after free (10.19) rank (score)
share of Microsoft's CVEs that are memory-safety issues~70 % per year
CVE submissions growth 2020-2025263 % increase
vulnerabilities found and fixed through OSS-Fuzz (cumulative)over 10,000 (and 36,000 bugs) across 1,000 projects, as of August 2023 count

Open contentions (7) open contention

Where sources or readings disagree. Both stay on the map.

Cross-scale patterns seen here (21) our proposal

Shapes that recur when the same kind of problem is solved at different scales, and how each shows in Software vulnerability. Each links to every other system that shows it.

Valence: what it trades, and with whom (18) published science

Each exchange with a neighbour, read from both sides: who gains, who pays, and what carries it.

Partners and parts published science

Built from / builds (14)

Sits inside / holds (3)

expressed by (41)

floor filled by (25)

represents (1)

Kinds (13)

Notes published science

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.

software vulnerability as vulnerability researchers describe it: flawed code and the inputs that reach it, up to the fix or the attack (reader's view)

  1. L1: Code and the inputs that reach it: software and hardware as written, shipped and running, and the bytes that cross into them
  2. L2: The flaw: a weakness in design, implementation, configuration or hardware behaviour, of a kind (memory safety, input validation, race condition, access control...)
  3. L3: Reachable and triggerable: active while the code runs; an input that crosses a trust boundary can make it misbehave; an exploit
  4. L4: Discovery: known to the vendor, a researcher or an attacker
  5. L5: Disclosure and coordination: secret, sold, reported, published; a CVE identifier assigned
  6. L6: The shared record and scores: CVE, CWE, CVSS, NVD, EPSS
  7. L7: Ended or used: remediated, mitigated, accepted, or exploited; dead when patched or retired

the boardroom (shared floors)

  1. L4: Discovery: the flaw known to someone: the vendor, a third party or an attacker; which came first often cannot be told
  2. L5: Disclosure and coordination: kept secret, sold or used; or reported to the vendor and published fully or after a fix (coordinated disclosure); a CVE identifier assigned by a CVE Numbering Authority
  3. L6: The shared record and its scores: the CVE entry (more than 327,000 records by April 2026: a count of records issued, shaped by CNA policy, e.g. the Linux kernel assigning a CVE to any bug fix it identifies), its CWE root cause, CVSS severity (0.0-10.0; v4.0 since November 2023 adds Threat metrics for proof-of-concept code or active exploitation), the NVD (a backlog of unenriched CVEs from early 2024), exploitation probability (EPSS) and CISA's catalogue of known exploited vulnerabilities (KEV)
  4. L7: Action back on the world: remediate (patch), mitigate, accept residual risk, disable; or exploit, stockpile, sell; the vulnerability ends when patched or removed from use

way in: Automated and AI-assisted discovery: continuous fuzzing, agents

  1. L1: Open-source projects enrolled in continuous fuzzing (OSS-Fuzz: over 10,000 vulnerabilities and 36,000 bugs across 1,000 projects by August 2023)
  2. L2: Coverage-guided fuzzers with sanitizers, and since 2024 language-model agents reading code (Big Sleep)
  3. L3: A flaw found and reported, sometimes before release: SQLite's fixed the same day

way in: Dependencies: flaws inherited from packages

  1. L1: Reusable packages shared through language-specific distributions; common code is better known to attackers
  2. L2: Declared dependencies, direct and indirect (a chain of length 2 or longer)
  3. L3: The dependent exposed: a vulnerability in npm affects a median of 30 releases

way in: Fuzzing: random inputs, watched for crashes

  1. L1: Random or mutated inputs and command-line parameters generated automatically
  2. L2: The program run under a monitor
  3. L3: Crashes flagged and debugged (25 to 33 percent of Unix utilities crashed in 1988-90)

way in: Hardware: flaws under the software

  1. L1: Microprocessors from Intel, AMD and ARM used in billions of devices
  2. L2: Speculative execution: guessed paths executed early for performance, discarded if wrong
  3. L3: Software's security assumptions violated (process separation, containers, JIT, side-channel countermeasures)

way in: Inputs across a trust boundary

  1. L1: Input that crosses a trust boundary: a network message, an uploaded file
  2. L2: Parsed by the receiving code (Heartbleed: the HeartbeatRequest's payload length field read without a bounds check)
  3. L3: The flaw triggered: memory read or overwritten, unintended behaviour (up to ~64 KB returned per request)

way in: Law, liability and disclosure policy

  1. L1: Laws and guidelines: no law requiring disclosure; the Dutch NCSC guideline (2013); US guidelines (2017)
  2. L2: Prosecutors take the guideline as a decision framework
  3. L3: The finder's and vendor's exposure: cease-and-desist letters, prosecution when a discloser 'went too far', and vendors seldom liable

way in: Money: rewards and prices

  1. L1: Rewards and prices: Chrome bounties from $500 to $31,336; zero-day exploits may exceed $100,000; value crashes once known
  2. L2: Programmes and markets: vendor VRPs, platforms such as HackerOne, brokers, gray and black markets (estimated ten times the white market, 2015)
  3. L3: Researchers drawn in and switching: new programmes attract researchers at the expense of old ones
  4. L4: The finder's choice to report or to sell, weighed against uncertainty

way in: New code: flaws written in

  1. L1: Code written by developers under deadlines and in large, complex code bases; 'ship it on Tuesday and get it right by version 3'
  2. L2: A weakness introduced, of a CWE kind. Across the 31,770 CVE records of 2024 the top-ranked kinds are cross-site scripting (CWE-79), out-of-bounds write (CWE-787) and SQL injection (CWE-89); in large C/C++ code bases memory-safety mistakes dominate (about 70% of Chromium's high-severity bugs; ~70% of the vulnerabilities Microsoft assigns a CVE each year)
  3. L3: Shipped and running: carrier before configured, dormant, active while the code runs

way in: Patches read backwards

  1. L1: A patch released for a vulnerability
  2. L2: Reverse engineered by attackers
  3. L3: The underlying flaw located on still-unpatched machines (half-days, zombie vulnerabilities)

way in: Reading the code: review, static analysis, penetration tests

  1. L1: Source text and running systems put in front of reviewers and testers
  2. L2: Code review and static analysis tools; penetration tests that try to enter via exploits
  3. L3: A suspected flaw, or none found (which does not prove the system secure)

way in: Telemetry: exploitation seen in the field

  1. L1: Files and traffic on hosts: 32 billion reports on about 300 million files from 11 million hosts (WINE)
  2. L2: Antivirus signatures and reputation data collected; network traces
  3. L3: Matched back to vulnerabilities after the fact: 11 of 18 zero-days not previously known as such

way in: Third-party reports

  1. L1: A finder outside the vendor writes up the flaw
  2. L2: Received through a disclosure policy or rewards programme (28% of Chrome's and 24% of Firefox's patched vulnerabilities came through VRPs)
  3. L3: Rated by the vendor's severity guideline (a critical bug runs arbitrary native code)

way out: Change how new code is written

  1. L1: Fewer memory-safety flaws written in: 76% of Android vulnerabilities in 2019, 24% in 2024
  2. L5: New code written in memory-safe languages
  3. L7: Decision to change development practice (Safe Coding for new code)

way out: Exploit

  1. L1: The machine does what its makers did not intend: memory read, code injected, service denied
  2. L3: Delivered as crafted input; harmless vulnerabilities chained into access
  3. L5: Scripted and published: the exploit becomes a rote exercise that attackers without the skill can run; in CERT/CC data 1996-1999 incidents surged after the script appeared, not after disclosure
  4. L7: Decision to exploit, alone or chained

way out: Mitigate: make exploitation of a class harder

  1. L1: The running machine changed: memory segments placed at random, data made non-executable
  2. L3: Exploitation of the whole memory-corruption class made much harder, not ended: code reuse (ROP) and information leaks bypass the deployed protections
  3. L5: Protections built into compilers and operating systems (stack cookies, exception handler validation, Data Execution Prevention, ASLR); deployed only when their overhead is in reasonable bounds
  4. L7: Decision to mitigate: measures that do not close the flaw but make it harder to exploit

way out: Patch and deploy

  1. L1: The code on machines changed (or not): of sites still vulnerable at 48 h, 10% replaced certificates within a month and 14% of those kept the old key
  2. L3: Released and deployed: the Alexa Top 100 patched within 24 h for Heartbleed; patching plateaued after about two weeks
  3. L5: The patch written: security commits are small (median 7 lines against 16 for other fixes); some first patches are incomplete or regressive
  4. L7: Decision to remediate

way out: Publish and notify

  1. L1: Read by administrators and users (and by attackers)
  2. L5: Operators notified (Heartbleed: 150,000 hosts; a 47% increase in patching by notified operators)
  3. L6: The CVE record and advisory published

way out: Sell or stockpile

  1. L1: Money paid: remote zero-click exploits the dearest
  2. L4: Through brokers and markets that lack transparency
  3. L7: Kept, used, stockpiled or sold; governments may also notify the vendor

Sources (51) published science

Vulnerability (computer security) (Wikipedia, via Wikimedia Enterprise snapshot)
Bilge L, Dumitras T. Investigating Zero-Day Attacks. ;login: 38(4), 2013 (summary of 'Before We Knew It: An Empirical Study of Zero-Day Attacks in the Real World', ACM CCS 2012, doi:10.1145/2382196.2382284) (behind its links)
Durumeric et al., The Matter of Heartbleed, ACM Internet Measurement Conference 2014 (doi:10.1145/2663716.2663755) (behind its links)
Maillart T, Zhao M, Grossklags J, Chuang J. Given enough eyeballs, all bugs are shallow? Revisiting Eric Raymond with bug bounty programs. Journal of Cybersecurity 3(2), 2017; arXiv:1608.03445 (behind its links)
Vander Stoep J, Rebert A (Google). Eliminating Memory Safety Vulnerabilities at the Source. Google Security Blog, September 2024 (behind its links)
Zero-day vulnerability (Wikipedia, via Wikimedia Enterprise snapshot) (behind its links)
Clark S, Frei S, Blaze M, Smith J. Familiarity Breeds Contempt: The Honeymoon Effect and the Role of Legacy Code in Zero-Day Vulnerabilities. ACSAC 2010 (behind its links)
Szekeres L, Payer M, Wei T, Song D. SoK: Eternal War in Memory. IEEE Symposium on Security and Privacy 2013 (open copy; abstract and sections I-III read) (behind its links)
Kocher et al. 2018, Spectre Attacks: Exploiting Speculative Execution (arXiv:1801.01203) (behind its links)
Full disclosure (computer security) (Wikipedia, via Wikimedia Enterprise snapshot) (behind its links)
Alexopoulos N, Brack M, Wagner JP, Grube T, Mühlhäuser M. How Long Do Vulnerabilities Live in the Code? A Large-Scale Empirical Measurement Study on FOSS Vulnerability Lifetimes. USENIX Security 2022 (behind its links)
Li F, Paxson V. A Large-Scale Empirical Study of Security Patches. ;login: 43(1), 2018 (summary of the ACM CCS 2017 paper, doi:10.1145/3133956.3134072) (behind its links)
Ozment A, Schechter SE. Milk or Wine: Does Software Security Improve with Age? USENIX Security 2006 (behind its links)
Rescorla E. Is Finding Security Holes a Good Idea? IEEE Security & Privacy 3(1), 2005 (doi:10.1109/MSP.2005.17); open copy (behind its links)
The Chromium Projects. Memory safety (behind its links)
Zerouali A, Mens T, Decan A, De Roover C. On the Impact of Security Vulnerabilities in the npm and RubyGems Dependency Networks. Empirical Software Engineering 2022; arXiv:2106.06747 (behind its links)
Finifter M, Akhawe D, Wagner D. An Empirical Study of Vulnerability Rewards Programs. USENIX Security 2013 (behind its links)
Jacobs J, Romanosky S, Edwards B, Roytman M, Adjerid I. Exploit Prediction Scoring System (EPSS). arXiv:1908.04856 (behind its links)
MITRE. 2024 CWE Top 25 Most Dangerous Software Weaknesses (behind its links)
Anderson, Why Information Security is Hard: An Economic Perspective, ACSAC 2001 (doi:10.1109/ACSAC.2001.991552) (behind its links)
Fuzzing (Wikipedia, via Wikimedia Enterprise snapshot) (behind its links)
FIRST. Common Vulnerability Scoring System v3.1: Specification Document (behind its links)
Heartbleed (Wikipedia, via Wikimedia Enterprise snapshot) (behind its links)
Arora A, Krishnan R, Telang R, Yang Y. An Empirical Analysis of Software Vendors' Patch Release Behavior: Impact of Vulnerability Disclosure. Information Systems Research 21(1):115-132, 2010 (doi:10.1287/isre.1080.0226); abstract on RePEc (the open PDF link given by the specialist returned 'page not found') (behind its links)
MITRE. 2024 CWE Top 25 Most Dangerous Software Weaknesses: the ranked list (behind its links)
MSRC Team (Gavin Thomas). A proactive approach to more secure code. Microsoft Security Response Center blog, 16 July 2019 (read through the page-reading tool; the figure caption quoted as it returned it) (behind its links)
NIST. NIST Updates NVD Operations to Address Record CVE Growth. News release, 15 April 2026 (updated 17 April) (read through the page-reading tool; sentences quoted as it returned them) (behind its links)
Google. OSS-Fuzz documentation (front page, 'Trophies') (behind its links)
Exploit (computer security) (Wikipedia, via Wikimedia Enterprise snapshot) (behind its links)
Common Vulnerabilities and Exposures (Wikipedia, via Wikimedia Enterprise snapshot) (behind its links)
Weulen Kranenbarg M, Holt TJ, van der Ham J. Don't shoot the messenger! A criminological and computer science perspective on coordinated vulnerability disclosure. Crime Science 7:16, 2018 (doi:10.1186/s40163-018-0090-8) (behind its links)
Computer security (Wikipedia) (behind its links)
Morris worm (Wikipedia, via Wikimedia Enterprise snapshot) (behind its links)
Computer memory (Wikipedia, WME snapshot) (behind its links)
Malware (Wikipedia, via Wikimedia Enterprise snapshot) (behind its links)
OWASP Top 10:2021, A01 Broken Access Control (CC BY-SA 4.0) (behind its links)
Hutchins, Cloppert and Amin, Intelligence-Driven Computer Network Defense Informed by Analysis of Adversary Campaigns and Intrusion Kill Chains, ICIW 2011 (behind its links)
Souppaya M, Scarfone K, Dodson D. Secure Software Development Framework (SSDF) version 1.1. NIST SP 800-218, 2022, doi:10.6028/NIST.SP.800-218 (behind its links)
Axelrod & Iliev 2014, Timing of cyber conflict, PNAS 111(4):1298-1303, doi:10.1073/pnas.1322638111 (PMC3910638) (behind its links)
Zimmermann, Staicu, Tenny, Pradel 2019. Small World with High Risks: A Study of Security Threats in the npm Ecosystem (USENIX Security 2019). arXiv:1902.09217 (behind its links)
Vulnerability management (Wikipedia, via Wikimedia Enterprise snapshot 2026-11-01) (behind its links)
Scarfone K, Souppaya M, Cody A, Orebaugh A (2008). NIST SP 800-115: Technical Guide to Information Security Testing and Assessment. NIST (behind its links)
Google Project Zero (Big Sleep team). From Naptime to Big Sleep: Using Large Language Models To Catch Vulnerabilities In Real-World Code. Project Zero blog, October 2024 (behind its links)
Browne HK, Arbaugh WA, McHugh J, Fithen WL. A Trend Analysis of Exploitations. University of Maryland CS-TR-4200 / UMIACS-TR-2000-76, November 2000 (the open tech report at the URL the specialist gave; the specialist titled it 'Windows of Vulnerability: A Case Study Analysis', IEEE Computer 2000, a companion paper not opened here) (behind its links)
Buffer overflow (Wikipedia, rev 1374582723, 2026-09-12, via Wikimedia Enterprise) (behind its links)
Bug bounty program (Wikipedia, via Wikimedia Enterprise snapshot) (behind its links)
Buchinger TJ, Scott AM et al. A pheromone antagonist liberates female sea lamprey from a sensory trap to enable reliable communication. PNAS 117:7284, 2020 (doi:10.1073/pnas.1921394117; PMC7132252) (behind its links)
Fissette SD, Buchinger TJ, Tamrakar S, Scott AM, Li W. Sensory trap leads to reliable communication without a shift in nonsexual responses to the model cue. Behavioral Ecology 2024 (doi:10.1093/beheco/arae006; PMC10878365) (behind its links)
Ware WH (ed.) (1970) Security Controls for Computer Systems: Report of Defense Science Board Task Force on Computer Security (RAND R-609-1; NIST early papers copy) (behind its links)
Karger PA, Schell RR. Multics Security Evaluation: Vulnerability Analysis. ESD-TR-74-193 Vol. II, Electronic Systems Division (AFSC), June 1974 (NIST early computer security papers copy; scanned text, front matter and sections read) (behind its links)
MITRE. CWE VIEW: Hardware Design (CWE-1194) (behind its links)