Threat modeling fits in the design phase of the software development life cycle, before code is written, and it is repeated whenever the system changes in a way that adds assets, data flows, or trust boundaries. Done there, it is cheap and it shapes the build. Done after release, it is a penetration test with extra steps.
This article covers when to threat model, what the exercise actually produces, how the major secure-development frameworks place it, and how a threat model should feed an organization’s policies and maturity program rather than sitting in a folder.
What threat modeling is
The Threat Modeling Manifesto defines it as analyzing representations of a system to highlight concerns about security and privacy characteristics. At the highest level it answers four questions, which the OWASP Threat Modeling Cheat Sheet uses as the structure of the whole process:
- What are we working on? Model the system: assets, data flows, trust boundaries, actors.
- What can go wrong? Identify and rank threats against that model.
- What are we going to do about it? Choose mitigations and record accepted risks.
- Did we do a good enough job? Validate that the mitigations landed and the model still matches the system.
Microsoft’s STRIDE is the most common way to answer the second question. It sorts threats into spoofing, tampering, repudiation, information disclosure, denial of service, and elevation of privilege, so that for each element of the model you ask six specific questions instead of one vague one.
When to do it
| Phase | Trigger | What the model focuses on |
|---|---|---|
| Design | New system or major feature | Assets, data flows, trust boundaries, who can reach what |
| Design | Architecture change: new cloud service, new integration, new payment path | The new boundary and everything that now crosses it |
| Development | A component handles input, authentication, authorization, or sensitive data | Code-level threats: injection, broken access control, session handling |
| Testing | Before release | Validating that the mitigations from the model exist and work |
| Maintenance | Incident, new threat intelligence, or periodic review | Whether the model still matches the deployed system |
| Disposal | Decommissioning | Where the data goes and what access survives |
The common failure is treating threat modeling as a one-time gate. Systems change. A model from the original design says nothing about the payment integration added a year later.
How the frameworks place it
- NIST SP 800-218, the Secure Software Development Framework (SSDF), puts threat modeling under its Produce Well-Secured Software group, in the practice on designing software to meet security requirements and mitigate security risks. SSDF describes outcomes, so it says the modeling should happen and be used, not which method to use.
- Microsoft’s Security Development Lifecycle lists ten practices. The third is to perform security design review and threat modeling, placed after establishing standards and choosing proven frameworks, and before cryptography, supply chain, testing, and monitoring.
- The OWASP cheat sheet and the Threat Modeling Manifesto are method-level guidance: the four questions, system modeling with data flow diagrams, STRIDE or similar for threat identification, and a validation step.
All three agree on placement. Threat modeling belongs in design, early, and it recurs. They differ in how prescriptive they are about the method.
System life cycle versus software life cycle
Threat modeling means slightly different things at two scopes, and a good program does both.
At the system level, the model shows how interconnected systems interact: where sensitive data flows, where trust boundaries sit between a web tier, a database, a cloud account, and a remote-access path. Concerns surface before they become expensive, such as a database holding customer records reachable from the same network segment as employee workstations.
At the software level, the focus narrows to code and application logic: user input, authentication, authorization, session management, and data handling. For a commerce site built on a content management system with plugins, that means insecure or unpatched plugins, poor input validation, SQL injection, cross-site scripting, and session hijacking. The system model tells you the shopping cart is internet-facing and touches payment data. The software model tells you which classes of bug in the cart would matter most.
What the threat model should change
A threat model that only produces a diagram has failed. Its findings should move three things.
- Design decisions. A trust boundary with no control on it becomes a requirement: segment the network, add authentication, encrypt the flow.
- Policy. If the model shows excessive employee access to server data, that should shape the access management policy. If it shows the cart exposed to internet threats, that should shape the patch management policy with faster timelines for internet-facing systems. If it finds missing logs, that should strengthen audit requirements. Policies written from a threat model read as answers to specific risks instead of generic control lists.
- Maturity. A documented, repeatable threat modeling practice is itself a maturity step: it moves an organization from reacting to incidents to anticipating them, and it produces evidence that can be reviewed as systems change.
A minimum viable threat model
Teams that have never done it can start small. One data flow diagram of the system with trust boundaries drawn. One STRIDE pass over each boundary crossing. One table with three columns: threat, mitigation or accepted risk, owner. One date for the next review, tied to the next architecture change. That is enough to change the design and to satisfy an auditor that the practice exists. Depth can come later; placement in the life cycle is what matters first.
See it applied
- Security Policy Examples: SDLC, Patching, Threat ModelingMy case study: three policies for a retailer, with an analysis of where threat modeling fits the system life cycle, the software life cycle, maturity, and policy.
- NIST SSDF Secure SDLC PlanA secure development plan built on SP 800-218, with threat modeling in the design practices.
- NIST SSDF vs Microsoft SDL vs BSIMMHow the three frameworks differ and when to use each.
Frequently asked questions
- When should threat modeling be done?
- In the design phase, before code is written, and again whenever the system gains new assets, data flows, integrations, or trust boundaries. It should also be revisited after incidents and on a periodic schedule.
- What are the four questions of threat modeling?
- What are we working on? What can go wrong? What are we going to do about it? Did we do a good enough job? They come from the Threat Modeling Manifesto and structure the OWASP threat modeling process.
- What is STRIDE?
- A Microsoft model that categorizes threats as spoofing, tampering, repudiation, information disclosure, denial of service, and elevation of privilege. It is used to ask specific questions about each element of a system model.
- Does NIST require threat modeling?
- NIST SP 800-218, the SSDF, includes designing software to meet security requirements and mitigate risks, which covers threat modeling, as one of its practices. SSDF is a set of recommended outcomes rather than a regulation, though it is referenced by federal procurement requirements.
Sources
- Threat Modeling ManifestoDefinition and the four key questions.
- OWASP Threat Modeling Cheat SheetThe four-phase process built on the four questions.
- Microsoft: Threats in the Threat Modeling Tool (STRIDE)The six STRIDE categories.
- Microsoft Security Development Lifecycle: PracticesPractice 3: perform security design review and threat modeling.
- NIST SP 800-218: Secure Software Development Framework (SSDF)Threat modeling within the Produce Well-Secured Software practices.