A threat modeling process is a repeatable way to find what could go wrong in a system before an attacker does. My threat modeling process has six steps: scope the system, diagram the data flows and trust boundaries, list threats for each element, rank the threats by exposure and impact, decide mitigations, then record the model and revisit the model on every significant change.
At a glance
- Definition: Analyzing a representation of a system to find security and privacy concerns (Threat Modeling Manifesto)
- Six steps: Scope, diagram, enumerate threats, rank, decide mitigations, record and revisit
- Starting point: The system: components plus the connections between them
- Threat checklist: STRIDE, applied per element as a second pass after a plain-language brainstorm
- Ranking: Exposure first (what the internet can reach), then impact
- Output: A dated record of threats, decisions, owners and review dates
- Revisit trigger: Any significant change to components, connections or access
What are the steps of the threat modeling process?
The threat modeling process in this guide runs in six steps, each with a concrete output:
- Scope the system. List the components, owners and people who know the system. Output: a component inventory and an invite list.
- Diagram data flows and trust boundaries. Draw what talks to what, where data moves and where trust changes. Output: a data flow diagram.
- Enumerate threats per element. Ask what could go wrong for each component and connection, then check coverage with STRIDE. Output: a threat list.
- Rank by exposure and impact. Put internet-facing systems first, then weigh what each system holds, what each system can reach and who notices if each system falls. Output: a ranked threat list.
- Decide mitigations. Choose to mitigate, accept, transfer or avoid each threat, and turn mitigations into security requirements. Output: decisions with owners.
- Record and revisit. Date the threat model and update the model when the system changes. Output: a versioned threat model.
The six threat modeling steps line up with the four questions of the Threat Modeling Manifesto:
| Threat Modeling Manifesto question | Threat modeling steps that answer the question |
| What are we working on? | Step 1 (scope) and step 2 (diagram) |
| What can go wrong? | Step 3 (enumerate threats) |
| What are we going to do about it? | Step 4 (rank) and step 5 (decide mitigations) |
| Did we do a good enough job? | Step 6 (record and revisit) |
Why should threat modeling start from the system and skip threat actor profiling?
Threat modeling works best when anchored on the system, because the ways into a system are knowable and the intentions of a specific adversary are mostly guesswork. Threat-actor-centric risk assessment asks who will attack and what that actor can do, and government agencies publish actor tier classifications to support the exercise. I was never convinced, because for most organizations the answer is speculation.
Two problems make threat actor profiling a weak base for a threat modeling process:
- The actor is unknowable for most organizations. Outside an intelligence agency, few teams can say which adversary is interested in them. Even with a named actor, nothing stops the same technique from being in someone else's hands.
- Capability is now a constant. Large language models give any motivated person access to tooling that used to belong to skilled groups, so ranking actors by capability models a world that no longer exists.
Key insight
A system-centric threat model defends the how, because the who is guesswork.
A system-centric threat model models the how. Attack vectors map directly to systems: exposed ports, missing authentication, vulnerable dependencies, leaked secrets and injection points. For each system, list the ways in and defend those. Defenders cannot defend a who. Defenders can defend a how.
Threat actor analysis still earns a place in a few narrow cases: nation-state targets, organizations with a dedicated threat intelligence team, and specific fraud schemes where knowing the actor's playbook changes the control. For how threat intelligence feeds a security program under ISO 27001, see the ISO 27001 A.5.7 threat intelligence guide.
Threat modeling step 1: how do you scope the system and pick the people?
Scoping a threat model means naming the system, listing the components inside the boundary, and inviting the people who know those components. The component list comes from the asset inventory: servers, services, databases, identity providers, SaaS tools, CI/CD pipelines and admin consoles.
The invite list matters as much as the component list. Examples of people to bring into the room:
- System administrators who run the hosts
- Developers who wrote the services
- Network engineers who own routing and firewall rules
- Managers and directors who know the business process the system supports
None of these people need cybersecurity experience. My approach is to open the session with three plain questions any stakeholder can answer:
- What do we have?
- What is it doing?
- What could go wrong?
I took the three questions from a five-minute health and safety course at a pharmaceutical plant. Everyone on the floor took the course, regardless of role. The hazard questions mapped cleanly onto threat modeling once the inventory and architecture were in hand.
Threat modeling step 2: how do you diagram data flows and trust boundaries?
A threat modeling diagram shows components, the connections between components, the data that moves along each connection, and the trust boundaries the data crosses. A list of components is too thin for threat modeling, because breaches travel through connections: the path from an internet-facing service to a database, or from a CI runner to production credentials.
A useful threat modeling diagram has these elements:
- Components: each service, data store, external system and user type.
- Connections: each flow, with the protocol and the direction.
- Data: what moves on each flow (credentials, customer records, logs).
- Trust boundaries: where control changes, for example internet to DMZ, application tier to data tier, company network to SaaS provider.
- Privileged paths: admin consoles, break-glass accounts and deployment pipelines.
For the drawing conventions and zones, see the guide to creating a security architecture diagram. The team cannot ask what could go wrong until the team can see the connections.
Threat modeling step 3: how do you enumerate threats with STRIDE?
Threat enumeration means asking what could go wrong for every component and every connection on the diagram. My approach runs in two passes. The first pass is a plain-language brainstorm with the whole group, captured without sorting. The second pass uses STRIDE as a coverage checklist to catch categories the brainstorm missed.
STRIDE threat modeling sorts threats into six categories, as used in the Microsoft Threat Modeling Tool:
| STRIDE category | Security property at risk | Example question for a connection or component |
| Spoofing | Authentication | Can a caller pretend to be another user or service? |
| Tampering | Integrity | Can data be changed in transit or at rest? |
| Repudiation | Non-repudiation | Can a user deny an action because no log records the action? |
| Information disclosure | Confidentiality | Can someone read data without permission? |
| Denial of service | Availability | Can someone make the service unavailable to valid users? |
| Elevation of privilege | Authorization | Can a low-privilege user gain admin rights? |
Watch out
Formal threat modeling methods help as a second pass and stall the room as a first pass.
STRIDE, PASTA, LINDDUN and attack trees all structure thinking. When a session opens with the methodology, I often see the group debate what the terms mean and whether the method is being followed correctly, and the non-security people go quiet for fear of saying something wrong. The people who hold the ground truth about the system stop talking. Capture everything in plain words first, then categorize.
For an applied example of per-element threat enumeration on an AI agent, see the AI agent threat model template.
Threat modeling step 4: how do you rank threats by exposure and impact?
Threat ranking starts with exposure, because the internet-facing systems are where an adversary starts. Every system is a little broken, and a team that tries to fix every threat at once fixes none. My ranking order:
- Internet-facing production systems. The external attack surface is the front door, with the highest exposure and the highest value.
- Internet-facing non-production systems. Exposure alone earns the place. A compromised test box is still a headline, and the box is a foothold an adversary can ride inward along the connections on the diagram.
- Administrative and privileged systems. Admin planes and privileged accounts may sit behind the perimeter, but a compromise there reaches everything. Blast radius makes these high impact at lower exposure.
- Everything else, in proportion. Internal, low-exposure, low-value systems get the controls they need and no more.
Exposure tells the team where to look. Impact tells the team what to fix. Teams tend to measure impact by what is on a system (sensitive data, money) and forget two more factors:
- What the system can reach. A non-production server with no sensitive data but credentials into production is a key to production.
- Who notices if the system falls. A public website scores near zero on confidentiality, yet a defaced homepage reaches the news, the CEO and customers.
Key insight
A public website with no sensitive data can still be a crown jewel.
In a crown-jewels session at a financial firm, the security architects named the transaction systems as the crown jewels. Nobody named the public marketing website, because the site held nothing sensitive and moved no money. I asked whether the CEO would care if the homepage were defaced or went down, and the room went quiet. Reputation does not follow data classification.
For ranking individual vulnerabilities on the same principles, see the risk-based vulnerability management guide.
Threat modeling step 5: how do you decide mitigations?
Each ranked threat gets one of four risk treatment decisions:
- Mitigate: add or change a control, for example adding authentication to an exposed admin endpoint.
- Accept: record the risk, the reason, the approver and a review date.
- Transfer: move the risk to another party, for example through a contract or insurance.
- Avoid: remove the feature, the flow or the exposure.
Mitigations cover protect, detect, respond and recover. A threat the team cannot fully prevent still needs a log that detects the attack and a runbook for the response.
Baseline controls such as MFA and encryption are questions the industry has answered, so the threat model confirms the baseline is in place and spends the hours elsewhere. The practitioner's guide to security architecture covers how threat models feed the three types of security requirements. Write each mitigation as a requirement the engineering team can build and test, with an owner.
Threat modeling step 6: how do you record and revisit a threat model?
A threat model is useful after the workshop only as a dated record. A threat model record has these fields:
- System name, scope and diagram version
- Each threat, the element the threat applies to, and the STRIDE category
- Exposure and impact ranking
- Decision (mitigate, accept, transfer, avoid), owner and due date
- Review date and approver
Revisit the threat model on change. Examples of changes that trigger a review:
- A new internet-facing endpoint or service
- A new data flow, integration or third-party vendor
- New privileged access or a new admin path
- A move to a new hosting environment or architecture
The review answers the fourth Threat Modeling Manifesto question: did we do a good enough job? For where threat modeling fits inside a full design review, see the security architecture review process.
What evidence does a threat modeling process produce for SOC 2 and ISO 27001?
A dated threat model with named threats, decisions and owners is risk assessment evidence as a byproduct of the security work:
| Framework | Requirement | Threat modeling evidence |
| SOC 2 CC3.2 | The entity identifies risks to the achievement of objectives and analyzes risks as a basis for determining how to manage them (AICPA Trust Services Criteria) | Threat list, ranking and treatment decisions |
| SOC 2 CC3.4 | The entity identifies and assesses changes that could significantly affect the system of internal control | Threat model reviews triggered by system changes |
| ISO/IEC 27001:2022 clause 6.1.2 | Information security risk assessment | Threat list, ranking, owners |
| ISO/IEC 27001:2022 A.8.25 | Secure development life cycle | Threat models for significant features at design |
| ISO/IEC 27001:2022 A.8.27 | Secure system architecture and engineering principles | Data flow diagrams with trust boundaries |
How should a team run a first threat modeling session?
A first threat modeling session needs a small scope and the right people more than a perfect method. Steps for the first session:
- Pick one system with an internet-facing component.
- Invite the people who built and run the system.
- Draw the components and connections on a whiteboard together.
- Ask what could go wrong for each connection, in plain words.
- Run the STRIDE checklist over the diagram to catch gaps.
- Rank by exposure, then impact, and assign owners for the top threats.
- Save the diagram and list with a date and a review trigger.
The method serves the thinking. Start from the system and plain questions, and bring in the formal tools once the insights are on the table.
Truvo Cyber runs threat modeling as part of security architecture engagements and cybersecurity risk assessments.
Threat model your critical systems
Turn threat models into decisions and owners inside an effective security program.
Frequently asked questions
What are the steps of the threat modeling process?
A practical threat modeling process has six steps: scope the system and invite the people who know the system, diagram data flows and trust boundaries, enumerate threats for each component and connection, rank threats by exposure and impact, decide a mitigation or treatment for each threat, and record the model with a date and revisit the model when the system changes.
What is STRIDE threat modeling?
STRIDE is a threat classification used in the Microsoft Threat Modeling Tool. The six categories are Spoofing, Tampering, Repudiation, Information disclosure, Denial of service and Elevation of privilege. Teams apply STRIDE to each component and data flow on a diagram to check that a brainstorm covered every type of threat. STRIDE works well as a second-pass checklist.
Who should take part in threat modeling?
Threat modeling works best with the people who built and run the system: system administrators, developers, network engineers, and the managers who own the business process. No cybersecurity experience is required. A security practitioner facilitates with plain questions: what do we have, what is it doing, and what could go wrong.
How often should a threat model be updated?
Update a threat model whenever the system changes in a way that adds exposure or access, for example a new internet-facing endpoint, a new integration or vendor, new privileged access or a new hosting environment. Many teams also set a fixed review date, such as annually, so a model for a stable system still gets a dated review.
What is the difference between threat modeling and a risk assessment?
Threat modeling analyzes one system: components, connections and the ways each could be attacked. A risk assessment covers risks to the organization's objectives across all systems and processes. A threat model feeds the risk assessment with system-level threats, rankings and decisions, and both appear as evidence for SOC 2 CC3.2 and ISO 27001 clause 6.1.2.
Ready to Start Your Compliance Journey?
Get a clear, actionable roadmap with our readiness assessment.
Contact Us
Sample SOC 2 Control List
Input your email to download the list.
About the Author
Former security architect for Bank of Canada and Payments Canada. 20+ years building compliance programs for critical infrastructure.