Frameworks and standards
The methodologies, scoring systems and control frameworks offensive-security work is measured against: what each one is, and when you will cite it.
On this page
Learning objectives
- Tell testing methodologies, scoring systems and control frameworks apart
- Choose the right framework to cite for a given engagement or audience
- Recognise the AI-specific frameworks, the OWASP LLM Top 10 and Agentic Top 10
A quick reference to the frameworks you will meet across engagements and reports. They fall into four groups: methodologies (how to test), scoring (how bad it is), control frameworks (what good looks like), and the newer AI-specific frameworks (how to test and secure LLM and agentic systems).
Testing methodologies
OWASP WSTG, the Web Security Testing Guide, is the de-facto methodology for web application testing: a structured catalogue of what to test (auth, session, access control, input handling, business logic) and how. Use it as the coverage checklist for a web engagement.
OWASP MASTG / MASVS, Mobile. The Mobile Application Security Testing Guide (method) and Verification Standard (requirements) are the reference for mobile testing: static and dynamic analysis, storage, crypto, and platform interaction.
OWASP Top 10 is not a methodology but a prioritised awareness list of the most impactful web risks (Broken Access Control is number one). It is good for framing executive risk; WSTG is the actual test plan. There is also an API Security Top 10 (BOLA, BFLA, mass-assignment).
NIST SP 800-115 is the US standard for technical security testing: a high-level, defensible structure for the whole engagement (planning, discovery, attack, reporting). It is often cited in the methodology section of a report.
OSSTMM, the Open Source Security Testing Methodology Manual, is a rigorous, metrics-driven testing methodology. Less common day to day, but cited where a measured, repeatable approach is required.
PTES, the Penetration Testing Execution Standard, is a practitioner-friendly end-to-end model that maps closely to the penetration-test lifecycle.
Adversary modelling and scoring
MITRE ATT&CK is the shared language of adversary technique: tactics, techniques and sub-techniques observed in the real world. Tag findings with ATT&CK IDs so blue teams can map them straight to detections.
Cyber Kill Chain is the linear, seven-stage model of an intrusion, best for telling the story of an attack to non-specialists (the report's executive walkthrough).
CVSS (v4.0), the Common Vulnerability Scoring System, is a standardised 0 to 10 severity score with a published vector. It is the yardstick for finding severity; tailor it with the Environmental metrics to reflect this client's risk. See the severity rubric.
AI and LLM security frameworks
As offensive work increasingly targets (and uses) LLM-backed systems, two OWASP projects have become the reference points.
OWASP Top 10 for LLM Applications catalogues the most impactful risks in LLM-backed systems: prompt injection (direct and indirect), insecure output handling, training-data poisoning, model denial of service, supply-chain risks, sensitive-information disclosure, insecure plugin and tool design, excessive agency, overreliance, and model theft. Use it to scope a test of any product that embeds an LLM.
OWASP Agentic Top 10 extends that to autonomous, tool-using agents: risks such as goal manipulation, tool misuse, excessive autonomy and privilege, memory poisoning, and unsafe orchestration across multiple agents. Cite it when the target is an agentic system that can plan and act rather than just answer.
NIST AI RMF (AI Risk Management Framework) is a programme-level framework for governing AI risk. It shows up where a client wants findings mapped into an existing AI governance effort rather than a purely technical write-up.
Control frameworks (what good looks like)
These are defensive frameworks. You map findings to them so clients can slot remediation into a programme they already run.
| Framework | What it is | Where it shows up |
|---|---|---|
| ISO 27001 (Annex A) | International ISMS standard and control set | Findings mapped to Annex A controls; common in AU, UK and EU reports |
| NIST CSF 2.0 | Cybersecurity Framework: Govern, Identify, Protect, Detect, Respond, Recover | Programme-level mapping |
| ASD Essential 8 | The Australian Signals Directorate's eight mitigation strategies and maturity levels | Australian engagements, especially government-adjacent |
| CIS Benchmarks / Controls | Prescriptive hardening baselines (including cloud) | Cloud reviews measure posture against these |
| PCI DSS | Payment-card security standard | Anything touching cardholder data or payment flows |
Putting them to work
- Findings carry framework tags (MITRE ATT&CK, OWASP categories, CWE) so the report speaks the defender's language.
- Engagements can be framed against a standard (for example ISO 27001 Annex A or Essential 8) so findings map directly into the client's existing compliance programme.
- The methodology section of a report cites the relevant testing standards (WSTG, NIST 800-115, and for AI systems the OWASP LLM Top 10) so the work is defensible and repeatable.
Related reading
- The tool index
- Severity and CVSS rubric
- Glossary
Knowledge check
Answer without AI. This is your own recall. 70% to pass.
1.Which framework is the shared language for tagging adversary techniques so defenders can map them to detections?
2.What is the number one risk in the OWASP LLM Top 10?
3.Which of these is a testing methodology rather than a scoring system or control framework?