Ready reckoner
Across the lifecycle
| Phase | Threat modelling activity | Output |
|---|---|---|
| Requirements | Abuse cases; regulatory needs | Security requirements |
| Design | Full model: DFD, STRIDE, rating | Design decisions, mitigations |
| Build | Mitigations as stories / criteria | Tickets linked to threat IDs |
| Test | Test per mitigation; pen-test focus | Automated tests; findings fed back |
| Release | Review open threats | Signed residual-risk acceptance |
| Operate | Detection; incidents vs model | Updates; "did we anticipate this?" |
Traceability: threat → requirement → ticket → test → monitoring. Put the threat ID in tickets and test names.
Frameworks
- NIST SSDF SP 800-218 PW.1.1 — use risk modelling (threat modelling, attack modelling, attack surface mapping). SP 800-218A adds generative-AI practices.
- OWASP SAMM — Design → Threat Assessment: application risk profile + threat modelling streams; 3 maturity levels.
- Microsoft SDL — threat modelling core practice; free Threat Modeling Tool.
- Also: BSIMM (Architecture Analysis), IEC 62443-4-1, ISO/SAE 21434 (TARA), PCI Secure Software Standard, EU CRA.
Traditional vs continuous
| Traditional | Continuous |
|---|---|
| Once, expert-led, long document | Baseline + increments, team-owned |
| Thorough snapshot | Stays current |
| Goes stale | Can be shallow without a baseline |
Trigger rule: if the Level-1 diagram would change, update the model. Triggers: new interface · new boundary · new sensitive data · authN/authZ change · new AI component · incident/pen-test gap · major dependency · regulation.
Agile cadences
- Epic: 60–120 min session, refresh DFD, STRIDE on new crossings.
- Story (definition of ready): entry point? data store/sensitivity? who can do what? dependency? AI capability? → if yes, security acceptance criteria.
- Continuous: model in repo, PR review, threats in backlog, metrics.
As code and tools
pytm (Python) · Threagile (YAML) · Threat Dragon (OWASP, visual) · Microsoft TMT (STRIDE templates) · commercial platforms. Automation: versioning, diffs, baseline threats, pipeline gates. Cannot: business logic, emergent design flaws, the conversation.
Programme
- Risk tiers (data sensitivity, exposure, money movement, regulation, AI) set depth.
- Security champions (10–20% time, community, access to experts).
- Template: scope, DFD, assets, assumptions, register, accepted risks, review date.
- Metrics: coverage · freshness · flow (time to session, time to fix) · effectiveness (findings anticipated) · register health. Avoid "number of threats".
- Maturity: ad hoc → repeatable → defined → measured → optimised.