Most threat models stop at the application, while most serious intrusions are decided by what the network lets an attacker reach next. Network threat modelling applies the same structured questions used for software design to the network itself: where the trust boundaries are, which paths connect an entry point to a critical asset, where those paths can be narrowed, and whether the segmentation on paper is the segmentation in production. This guide is for network and security architects, assessors and defenders who already know the general method. It focuses on building the network model, enumerating attack paths, choosing choke points and validating segmentation, and links the general method pages rather than repeating them.
Key takeaways
- Model reachability, not just topology: a network diagram shows boxes and lines; a threat model shows which identity, from which zone, can reach which service on which port.
- Trust boundaries are where controls belong: every place traffic or credentials cross between zones of different trust is a candidate for enforcement, logging and testing.
- Think in paths: rank threats by complete routes from an entry point to a crown-jewel asset, and break the cheapest paths first.
- Choke points give leverage: a small number of enforcement points, such as management interfaces, jump hosts and identity infrastructure, sit on most paths.
- Validate continuously: segmentation drifts with every rule change, so the model needs a test that proves intended flows are allowed and everything else is not.
How network threat modelling differs from application threat modelling
The Threat Modeling Manifesto's four questions still apply: what are we working on, what can go wrong, what are we going to do about it, and did we do a good enough job. What changes is the unit of analysis. In an application model, the elements are processes, data stores and data flows. In a network model, they are zones, the services exposed in each zone, the paths between zones and the identities that can use those paths. The Threat Modelling Guide covers the general method; the table shows what shifts for networks.
| Aspect | Application threat model | Network threat model |
|---|---|---|
| Main diagram | Data flow diagram of one system | Zone and conduit diagram across many systems, with permitted flows |
| Trust boundary | Process or privilege boundary in code | Firewall, VLAN, VPC, VPN, cloud account or tenant boundary, and management plane boundary |
| Typical question | Can this input tamper with that store? | Can a host in this zone reach that service, and with which credential? |
| Main threats | Injection, broken authorisation, data exposure | Exposed services, lateral movement, boundary bridging, management plane abuse, unmonitored egress |
| Change driver | Code releases | Firewall and security group changes, new connectivity, mergers, cloud expansion |
| Validation | Code review, testing | Rule analysis, reachability testing, segmentation testing, flow log review |
Step 1: build the network model
A useful model is smaller than a full network diagram and more precise. Start from what matters and work outward.
- Name the crown jewels: systems whose compromise would be most damaging, such as identity infrastructure, payment or clinical systems, backup systems, source code and CI/CD, and OT controllers.
- Define zones by trust: group assets that share a trust level and exposure, for example internet edge, user access, server, management, backup, OT, partner and each cloud account. A VLAN is not automatically a zone; a zone is defined by policy.
- Draw the conduits: for each pair of zones, record the permitted flows as source, destination, protocol and port, and the enforcement point. Pull this from firewall and security group configuration, not memory.
- Add the identity layer: for each flow, note which accounts or credentials are used. Service accounts, API keys and certificates frequently cross zones that users never do.
- Mark entry points: internet-facing services, remote access, partner connections, wireless, user endpoints and third-party managed devices.
- Mark the management plane separately: SSH, RDP, WinRM, SNMP, hypervisor consoles, cloud control planes and network device admin interfaces often reach everything and deserve their own zone.
Keep each element traceable to a source: a configuration export, a cloud API inventory or a flow log sample. The Networking Fundamentals for Defenders Guide covers the underlying protocols, and the Threat Model Template gives a structure for recording the result.
Step 2: identify trust boundaries and what can go wrong at each
A trust boundary is wherever traffic or authority moves between zones of different trust. The OWASP Threat Modeling Cheat Sheet notes that trust boundaries often mark the likeliest attack points, and the same holds for networks. STRIDE still works at the boundary, with network-specific readings.
| STRIDE category (property) | Network reading at a boundary | Typical control |
|---|---|---|
| Spoofing (authentication) | A host or service impersonates another: address spoofing, rogue DHCP or DNS, stolen device certificate | 802.1X and NAC, mutual TLS, DHCP snooping, DNSSEC where relevant |
| Tampering (integrity) | Traffic altered in transit across an untrusted segment | TLS, IPsec or MACsec on links that cross trust levels |
| Repudiation (accountability) | No record of who crossed the boundary | Flow logs, firewall logs and authentication logs tied to identities |
| Information disclosure (confidentiality) | Sensitive data leaves a zone, or cleartext management protocols cross it | Egress filtering, encryption, removing cleartext protocols |
| Denial of service (availability) | A boundary device or shared link becomes a single point of failure | Redundancy, rate limiting, upstream DDoS protection |
| Elevation of privilege (authorisation) | Access in one zone is turned into access in a higher-trust zone | Deny by default between zones, privileged access paths, separate admin credentials |
Add two network-specific questions at every boundary: can this boundary be bridged by a dual-homed host, tunnel or misconfigured device, and is the boundary enforced in one place or many? MITRE ATT&CK lists T1599 Network Boundary Bridging and T1572 Protocol Tunneling as techniques that defeat a boundary drawn on paper.
Step 3: enumerate attack paths
An attack path is an ordered route from an entry point to a crown jewel, where each hop is a reachable service plus a usable credential or weakness. Paths are where network threat modelling earns its keep, because a single exposed service rarely matters on its own; what matters is where it leads. The table maps common hops to ATT&CK techniques so defenders can match them to detections.
| Path stage | Question to ask | ATT&CK technique |
|---|---|---|
| Initial access through the edge | Which internet-facing services exist, and what zone do they sit in? | T1190 Exploit Public-Facing Application |
| Remote access | Which VPN, VDI or remote services accept external logins, and with what authentication? | T1133 External Remote Services |
| Credential use | Which accounts work across zones, including service accounts? | T1078 Valid Accounts |
| Discovery | What can a foothold host scan or enumerate? | T1046 Network Service Discovery |
| Lateral movement | Which remote administration protocols are reachable from user and server zones? | T1021 Remote Services, T1210 Exploitation of Remote Services |
| Command and control and egress | Which zones can reach the internet directly or through proxies? | T1090 Proxy, T1572 Protocol Tunneling |
- Work backwards: start at each crown jewel and list every zone that can reach it, then every zone that can reach those. This finds paths faster than starting from the internet.
- Count credentials, not just ports: a reachable port with no usable credential is a weaker hop than an open management port where a widely deployed service account works.
- Rank paths: score by number of hops, effort per hop and detection coverage per hop. Short, cheap, unmonitored paths go first.
- Use tooling where it helps: attack path platforms and graph analysis of directory relationships can enumerate paths at scale. See Attack Path Analysis vs Attack Surface Management and the Identity Attack Paths Guide.
Assessments of attack paths should run under an authorised scope with change control, and the output belongs to defenders: a ranked list of paths and where to break them. The Lateral Movement Guide covers detection of the movement stage in more depth.
Step 4: choose choke points and responses
A choke point is an element that sits on many paths, so a control there breaks several at once. The OWASP cheat sheet's response options apply: mitigate, eliminate, transfer or accept, with each decision recorded. In networks, eliminating a flow is often cheaper than mitigating it.
| Choke point | Why it sits on many paths | Response options |
|---|---|---|
| Management plane | Admin protocols reach every server and device | Dedicated management zone, access only from privileged access workstations or jump hosts, separate admin credentials |
| Identity infrastructure | Domain controllers and identity providers authenticate everything | Tier the directory, restrict which zones reach it and on which ports, monitor authentication anomalies |
| Jump hosts and bastions | Designed as the single route into sensitive zones | Harden, require phishing-resistant MFA, record sessions; see If your jump host and bastion is compromised |
| Inter-zone firewalls | Enforce conduits between zones | Deny by default, allow by exception, rule ownership and expiry |
| Egress points | Most command and control and exfiltration crosses them | Default deny outbound from server zones, proxy with authentication, DNS filtering |
| Partner and third-party connections | Bridge into the network outside normal controls | Terminate in a dedicated zone, restrict to named services, review access regularly |
| Backup systems | Reach most data and are a ransomware target | Isolated zone, one-way flows where possible, separate credentials |
Choosing between enforcement styles is a design decision with trade-offs:
- Zone firewalls: fewer enforcement points and easier to audit, but coarse; they do not stop movement inside a zone. See the Network Segmentation Guide.
- Microsegmentation: host or workload level policy that limits movement within a zone; more precise but more rules to own. See Microsegmentation vs ZTNA.
- Identity-based access: zero trust approaches avoid implicit trust from network location; NIST SP 800-207 states that no implicit trust should be granted based solely on physical or network location. In practice most organisations combine all three. See Identity Segmentation vs Network Segmentation.
- Small organisations: start with three or four zones (edge, users, servers, management) and default-deny egress from servers. Large organisations: model per business unit or cloud account and federate the models through a shared zone taxonomy.
Step 5: validate segmentation against the model
The model makes a claim: these flows are allowed and nothing else is. Validation tests that claim. This is the fourth manifesto question, did we do a good enough job, applied to the network.
- Static rule analysis: compare firewall and security group rules with the model's conduits. Flag any-any rules, rules without owners and rules that permit flows the model does not list. The Firewall Rule Review Checklist helps here.
- Flow log comparison: compare observed flows (NetFlow, VPC flow logs, firewall logs) with the model. Observed flows that are not modelled are either missing from the model or should be blocked.
- Reachability testing: from a test host placed in each zone, under authorised scope, confirm that only the modelled services respond. Repeat after major changes. The Network Segmentation Testing Guide covers scoping and evidence.
- Path testing: for the top-ranked attack paths, have an authorised assessor or red team confirm whether the path is now broken, and record the evidence.
- Rollback plan: when tightening rules, start in log-only or monitor mode where the platform supports it, apply changes in rings (one zone pair at a time), and keep the previous rule set ready to restore if a business flow breaks.
Where identity fits
Network paths are only half of an attack path; the other half is the credential that makes the hop work. Service accounts used by monitoring, backup and deployment tools often have network reach into every zone and credentials that never expire. When modelling, list the non-human identities that cross each boundary, who owns them and whether they can authenticate from zones they should not. Restricting where a machine credential can be used is often a cheaper choke point than another firewall rule. See the Identity Attack Paths Guide.
Keeping the model alive
- Triggers for review: new internet-facing service, new partner connection, new cloud account or region, merger or acquisition, a major incident, and any firewall change that creates a new zone-to-zone flow.
- Link to change control: add a field to firewall change requests asking which modelled conduit the rule implements; if none, the model needs updating or the request needs challenge. See the Firewall Rule Request Template.
- Cadence: full review annually, partial review on each trigger, and continuous flow comparison where tooling allows.
Detection and measurement
Each modelled boundary should have logging that would show a crossing. Monitor denied inter-zone connections from unexpected sources, new flows not present in the model, admin protocol use outside the management zone, and outbound connections from server zones. Useful metrics:
- Model coverage: percentage of production zones and cloud accounts represented in the current model.
- Unmodelled flows: count of observed inter-zone flows not in the model, trended down over time.
- Path count to crown jewels: number of ranked attack paths of three hops or fewer to each crown jewel.
- Choke point coverage: percentage of top paths that cross at least one monitored, enforced choke point.
- Rule hygiene: percentage of inter-zone rules with an owner, a justification and a review date.
- Validation age: days since each zone pair was last reachability tested.
Framework mapping
| Activity | NIST SP 800-53 Rev. 5 | MITRE ATT&CK mitigations | Other |
|---|---|---|---|
| Modelling and risk assessment | RA-3 Risk Assessment, PM-7 Enterprise Architecture, SA-8, SA-17 | No direct mitigation ID | Threat Modeling Manifesto; NIST SP 800-154 (draft) |
| Boundaries and conduits | SC-7 Boundary Protection, SC-7(5) Deny by Default, AC-4 Information Flow Enforcement, CA-3 Information Exchange | M1030 Network Segmentation, M1037 Filter Network Traffic | CIS Controls v8.1 Control 12 Network Infrastructure Management |
| Restricting remote services | AC-4, SC-7 | M1035 Limit Access to Resource Over Network | NCSC zero trust principle: do not trust any network, including your own |
| Validation | CA-8 Penetration Testing, SA-11 | No direct mitigation ID | CIS Controls v8.1 Control 13 Network Monitoring and Defense |
For OT environments, the IEC 62443 zones and conduits model is the equivalent; see the OT Network Segmentation, Purdue Model and IEC 62443 Zones and Conduits Guide.
Common failure patterns
- Diagram from memory: the model is drawn in a workshop and never reconciled with firewall configuration, so it describes the intended network, not the real one.
- Perimeter only: the model covers internet ingress and ignores east-west flows, where lateral movement happens. See the East-West Traffic Inspection Guide.
- Management plane in the server zone: admin protocols reachable from anywhere in the data centre turn any foothold into a path to everything.
- Ignoring credentials: paths are ranked by ports alone, missing the service account that works on every host.
- Cloud left out: security groups, peering and private endpoints are modelled separately or not at all, so cross-account paths go unseen.
- Never validated: segmentation is assumed after a project closes and drifts with each emergency rule change.
Practitioner checklist
- List crown-jewel systems and define zones by trust level, including a separate management zone.
- Build conduits from firewall and security group exports, not from memory.
- Record which identities, including service accounts, use each inter-zone flow.
- Apply STRIDE and boundary bridging questions at every trust boundary.
- Enumerate attack paths backwards from each crown jewel and rank them.
- Pick choke points that break the most top-ranked paths and record the response for each.
- Default-deny egress from server zones and restrict admin protocols to the management zone.
- Compare observed flow logs with the model and resolve every unmodelled flow.
- Reachability test each zone pair under authorised scope and after major changes.
- Tie firewall change requests to modelled conduits and review the model on defined triggers.
Standards and references
- Threat Modeling Manifesto
- OWASP: Threat Modeling Cheat Sheet
- NIST: SP 800-207 Zero Trust Architecture
- NIST: SP 800-154 Guide to Data-Centric System Threat Modeling (initial public draft)
- NIST: SP 800-53 Rev. 5
- MITRE ATT&CK: M1030 Network Segmentation
- UK NCSC: Zero trust architecture design principles
- CIS: Control 12 Network Infrastructure Management
Related NHI Mgmt Group resources: Threat Modelling Guide · Networking Fundamentals for Defenders Guide · Threat Model Template · Network Segmentation Testing Guide · Lateral Movement Guide