Engineer working on a laptop connected to a device test rig, surrounded by wiring harnesses and test hardware

Firmware vulnerabilities & fixes

NIST SP 800-207A: zero trust for cloud-native, multi-cloud applications

Last updated:

NIST SP 800-207A ("A Zero Trust Architecture Model for Access Control in Cloud-Native Applications in Multi-Cloud Environments") is a NIST Special Publication, final as of September 2023 and authored by Ramaswamy Chandramouli (NIST) and Zack Butcher (Tetrate), that takes the zero trust principles of NIST SP 800-207 and works out what they mean specifically for cloud-native applications spread across multiple cloud providers and on-premises environments. Its own abstract frames the problem directly: "One of the basic tenets of zero trust is to remove the implicit trust in users, services, and devices based only on their network location, affiliation, and ownership." SP 800-207A's objective, per that same abstract, is "to provide guidance for realizing an architecture that can enforce granular application-level policies while meeting the runtime requirements of ZTA for multi-cloud and hybrid environments."

Key takeaways

  • The base NIST SP 800-207 model already rejects the idea that being inside a network perimeter should imply trust.
  • The full text turns "identity-based segmentation" into five runtime policy checks.
  • SP 800-207A argues for both tiers, not identity alone.
  • Enforcing identity-tier policy consistently across clouds requires specific platform components.

Why isn't network-location trust enough for cloud-native?

The base NIST SP 800-207 model already rejects the idea that being inside a network perimeter should imply trust. SP 800-207A pushes that further for cloud-native applications specifically: per its abstract, "a key paradigm shift in ZTAs is the change in focus from security controls based on segmentation and isolation using network parameters (e.g., Internet Protocol (IP) addresses, subnets, perimeter) to identities." For an application made of many microservices spread across several clouds, a firewall rule keyed to an IP range cannot reliably express "this specific service is allowed to call that specific service" — so the model shifts enforcement to service and application identity instead.

Network-tier policy vs. identity-tier policy

SP 800-207A defines two distinct policy tiers that a zero trust deployment has to reconcile:

Policy tier Definition (as stated in the publication)
Network-tier "Allowed communication between enterprise network elements (e.g., firewall rules, which are relatively static)."
Identity-tier "Access scope for services and resources based on service and user identities (e.g., dynamic application-to-application communication rules based on identities through a dedicated infrastructure layer)."

The document's point is that these two tiers are complementary, not competing: network-tier controls remain a coarse, relatively static baseline, while identity-tier controls carry the dynamic, fine-grained enforcement that cloud-native applications actually need as services scale up, move, and get redeployed across clouds.

What are the five identity-based segmentation requirements?

The full text turns "identity-based segmentation" into five runtime policy checks. NIST labels them ID-SEG-REC-1 to ID-SEG-REC-5. They are the closest thing in the document to a requirement list you can audit against:

ID Requirement (summarised from SP 800-207A) What to check in a real backend
ID-SEG-REC-1 Encrypted connection between service endpoints, wherever they run (subnet, zone, region, cloud, on-premises) No plaintext service-to-service hop, including the "internal" ones
ID-SEG-REC-2 Each service presents a short-lived, cryptographically verifiable identity, authenticated per connection and re-authenticated regularly mTLS between services; certificate lifetime and connection lifetime
ID-SEG-REC-3 Service-to-service authorization based on that runtime identity, with the option to call an external authorization service Allow rules written against service identities, not IP ranges
ID-SEG-REC-4 End-user authentication with phishing-resistant MFA, carried as a verifiable runtime token (e.g., a JWT) checked at each hop How the user's identity travels past the API gateway
ID-SEG-REC-5 End-user-to-resource authorization on every request, in the application, from token claims, or at an external policy decision point JWT libraries decode and verify signatures

Source: NIST SP 800-207A (September 2023), "Requirements for Identity-Based Segmentation Policies for ZTA".

NIST adds two practical notes. Authenticating every single request is too disruptive, so services authenticate per connection via mTLS. Connections are kept short, "usually as long as the time to live [TTL] of the service's identity certificate or as short as 15-30 minutes, depending on the configuration." For external users, NIST recommends short-lived credentials such as OAuth 2.0 tokens, exchanged for a locally verifiable token like a JWT that is checked at each hop.

Why can't identity-tier policies replace firewall rules entirely?

SP 800-207A argues for both tiers, not identity alone. It gives three reasons why purely identity-based policies "cannot currently be adopted":

  • Location still matters. Hybrid estates must locate resources by network parameters, and network location stays a useful risk signal for context-based authorization.
  • Service identities can be hard to keep consistent. Identity assignments are often tied to specific domains, which makes consistent policy across on-premises, cloud and different runtimes difficult. NIST's mitigation is a universal identity domain with consistent service names, as recommended in SP 800-204A.
  • Network-only policy has its own cost. Firewall rules need constant upkeep as containers and VMs move.

NIST also notes that network-tier policies cannot be removed entirely under current compliance requirements, naming PCI DSS as an example. The recommended pattern is to relax network rules (for example, allowing traffic only across a defined set of gateways) and add identity-tier policies with layer-7 controls. NIST says this gives "a better overall security posture than either a purely identity-tier or purely network-tier approach."

The infrastructure the model assumes

Enforcing identity-tier policy consistently across clouds requires specific platform components. Per the abstract, this "requires a platform that consists of Application Programming Interface (API) gateways, sidecar proxies, and application identity infrastructures (e.g., Secure Production Identity Framework for Everyone [SPIFFE]) that can enforce those policies irrespective of the location of the services or applications, whether on-premises or on multiple clouds." The publication's own keyword list names the specific building blocks this implies: ingress gateways and egress gateways (controlling traffic entering and leaving a cluster or cloud boundary), transit gateways, sidecar proxies attached to individual services, and the broader service mesh pattern that ties them together — with SPIFFE providing the cryptographic service-identity layer those components rely on.

Which component enforces which policy?

SP 800-207A assumes one service mesh per cluster. A global control plane and a management plane sit above the clusters and push one uniform policy set to every local control plane. Inside each mesh, the control plane acts as the policy administration point, the policy tools act as the policy decision point, and the proxies are the policy enforcement points (PEPs). NIST calls the mesh "the modern cloud-native security kernel." The document gives each element a distinct job:

Element Role in SP 800-207A NIST's comparison
Sidecar proxy Sits beside each application instance and intercepts all of its traffic. Handles east-west calls between services The primary service-mesh use case
Ingress gateway Controls how cluster services are exposed: names, certificates, ports, protocols, endpoints A mesh-managed reverse proxy, similar to NGINX or HAProxy
Egress gateway Controls outbound calls. Can apply identity-based policy and present credentials (e.g., an mTLS certificate for a partner API) on the application's behalf "a next-generation identity-aware Squid proxy"
Edge gateway Accepts external traffic before the ingress gateway. Load-balances across clusters or sites and applies policy per request A software local traffic manager
Transit gateway A paired egress and ingress proxy plus the firewall rules that connect them —

Source: NIST SP 800-207A, "Functional Roles of Application Infrastructure Elements in Enforcing Policies".

NIST gives one reason not to run a single control plane across every cluster: doing so "may make the multiple clusters a single failure domain." One control plane per cluster, coordinated globally, isolates failures. It also suits environments such as air-gapped systems, where every workload cannot reach one central control plane.

Worked example: a cloud service calling an on-premises service

NIST's reference scenario has Service 1 in a cloud cluster calling Service 2 in an on-premises cluster. Traffic leaves through an egress gateway and an outbound firewall, then enters through an inbound firewall and an ingress gateway. Three layers of policy apply:

  1. Coarse-grained network tier (north-south)Mostly static firewall rules that only connect the egress gateway to the ingress gateway. NIST's example is allow 10.100.2.3/30 15443 to 10.1.2.3/30:15443.
  2. Fine-grained network tier (microsegmentation)The allowed paths from Service 1's subnet to the egress gateway, and from the ingress gateway to Service 2's subnet, including east-west flows inside the perimeter.
  3. Identity tier (mesh level)A rule enforced by the sidecars: only Service 1's identity may call Service 2, only on port 443, only GET, only on /public. NIST's Figure 4 shows it as a service-mesh authorization policy:
selector:
  matchLabels:
    app: service-2
action: ALLOW
rules:
- from:
  - source:
      principals: ["cluster.local/ns/service-1/sa/service-1"]
  to:
  - operation:
      ports: ["443"]
      methods: ["GET"]
      paths: ["/public"]

The firewall rules stay broad and stable. The identity-tier rule carries the precise intent and keeps working if either service moves to another subnet, cloud or site. NIST's second scenario applies the same layering to a three-tier application reached from a mobile app or website through a DMZ, with edge, ingress and egress gateways plus a central coordination infrastructure.

What monitoring does SP 800-207A expect?

Policies only hold if you can see them working. The document adds monitoring requirements for cloud-native applications and two rules for using the data:

ID Requirement
MON-CNA-REQ-1 Monitor every category of resource: enterprise-owned, not enterprise-managed, and personally owned
MON-CNA-REQ-2 Cover application targets (containers), mesh control-plane elements and data-plane elements (sidecars)
MON-CNA-REQ-3 Cover every user request and the chain of service calls it triggers
MON-CNA-REQ-4 Cover changes to enterprise directories and tie each change to a valid request
MON-DATA-USE-1 Base access decisions on assigned permissions plus context, including behavioural data from traffic flows
MON-DATA-USE-2 Use telemetry to grant, revoke or restrict access, and to trigger step-up authentication such as phishing-resistant MFA

Source: NIST SP 800-207A, "Support for Multi-tier Policies Through a Monitoring Framework".

How this extends SP 800-207

SP 800-207A does not replace the base zero trust architecture document — it builds on it. Its abstract is explicit that "NIST Special Publication 800-207 has laid out a comprehensive set of zero trust principles and referenced zero trust architectures (ZTA) for turning those concepts into reality," and that SP 800-207A's role is to work out the application-level, multi-cloud specifics that the general-purpose SP 800-207 model does not itself prescribe. Organizations that have already mapped SP 800-207's principles against device or network access (see our SP 800-207 guide) can treat SP 800-207A as the next layer down, specifically for the application and service-to-service traffic running behind those access decisions.

Where this matters for connected devices

Most connected-device and firmware products now depend on cloud backends — fleet management APIs, telemetry ingestion, OTA update services — that are themselves built as cloud-native, multi-service applications. SP 800-207A's identity-tier model gives device and firmware teams a vendor-neutral way to reason about how those backend services should authenticate and authorize each other, independent of which cloud or clouds the backend actually runs on.

SoftWSP's device and firmware security work can help map a connected-device backend's service-to-service access against this model rather than relying on network-location trust alone: see device & firmware security services.

Implementation checklist for a device-fleet backend

The steps below apply the document's requirements to a typical connected-device backend: device API, telemetry ingestion, OTA update service and admin console. Each step cites the part of SP 800-207A it comes from.

  1. Inventory the services and their real call graphNIST recommends an "observe-and-lock-down" method: use the mesh's observability to see the "requests made today," then write fine-grained policies for the allowed requests. It also calls for an authenticated service registry to discover dependencies, including SaaS endpoints.
  2. Give every workload a cryptographic identityNIST's example is a SPIFFE ID carried in an SVID, most commonly an X.509 certificate. Use one identity domain across clouds and sites.
  3. Turn on mTLS everywhere and keep connections short(ID-SEG-REC-1 and -2).
  4. Write allow rules per identity, method and path(ID-SEG-REC-3), starting with the OTA and signing paths, where a wrong caller does the most damage.
  5. Carry the user's identity as a verified token at every hop(ID-SEG-REC-4 and -5). Admin users get phishing-resistant MFA.
  6. Reduce firewall rules to gateway-to-gateway connectivityand keep the ones your compliance regime requires.
  7. Manage policies as code in CI/CDNIST lists "policy as code" and automated policy testing as advantages of identity-tier policies.
  8. Wire in monitoringto MON-CNA-REQ-1 to -4, including directory changes.

Common pitfalls

  • Treating the mesh as optional. The reference platform is a service mesh with sidecar proxies. NIST leaves the sidecar-less variant out of scope because "the deployment of this technology is still in early stages." Designs without sidecars need their own analysis.
  • Stopping at CNI network policy. NIST calls container-network-interface policy (eBPF or BGP-based) "a strong upgrade" over traditional microsegmentation. It also notes that this policy typically cannot apply per-request rules in application context, which identity-based segmentation needs.
  • Decoding JWTs without verifying them. ID-SEG-REC-5 requires libraries that both decode and verify the signature.
  • Long-lived service credentials. They undo ID-SEG-REC-2.
  • Ignoring application-tier rules. NIST mentions WAF rules, payload validation and Open Policy Agent checks as a further tier, outside this document's scope. Service identity does not replace input validation.

This guide summarises a public NIST publication; it does not constitute legal or compliance advice.

Put this guide to work

Frequently asked questions

What is NIST SP 800-207A?

A Zero Trust Architecture Model for Access Control in Cloud-Native Applications in Multi-Cloud Environments — a NIST Special Publication, final as of September 2023, authored by Ramaswamy Chandramouli (NIST) and Zack Butcher (Tetrate). Its own abstract states its objective is "to provide guidance for realizing an architecture that can enforce granular application-level policies while meeting the runtime requirements of ZTA for multi-cloud and hybrid environments." It is published as a companion to, not a replacement for, the foundational NIST SP 800-207 zero trust architecture document.

How does it relate to NIST SP 800-207?

SP 800-207A's own abstract states that the earlier "NIST Special Publication 800-207 has laid out a comprehensive set of zero trust principles and referenced zero trust architectures (ZTA) for turning those concepts into reality," and that this publication "expands" that reference environment specifically to multi-cloud and hybrid environments with cloud-native, microservices-based applications. See our [NIST SP 800-207 zero trust guide](/guides/nist-sp-800-207-zero-trust-architecture/) for the base model this document builds on.

What is the difference between network-tier and identity-tier policies?

The document defines network-tier policies as "allowed communication between enterprise network elements (e.g., firewall rules, which are relatively static)" and identity-tier policies as the "access scope for services and resources based on service and user identities (e.g., dynamic application-to-application communication rules based on identities through a dedicated infrastructure layer)." Its core argument is that network-tier controls alone are insufficient for cloud-native applications, because, per the abstract, zero trust requires "authentication and authorization policies based on application and service identities in addition to the underlying network parameters and user identities."

What infrastructure does the model assume?

Per the abstract, enforcing identity-tier policy "requires a platform that consists of Application Programming Interface (API) gateways, sidecar proxies, and application identity infrastructures (e.g., Secure Production Identity Framework for Everyone [SPIFFE])" capable of enforcing policy "irrespective of the location of the services or applications, whether on-premises or on multiple clouds." The document's own keyword list also names ingress gateways, egress gateways, transit gateways, microservices and service mesh as components of this architecture.

Sources

  1. NIST CSRC: SP 800-207A — A Zero Trust Architecture Model for Access Control in Cloud-Native Applications in Multi-Cloud Environments (publication detail, abstract, keywords)
  2. NIST SP 800-207A — full text (identity-tier/network-tier policy definitions, gateway and sidecar architecture)
  3. NIST CSRC: SP 800-207 — Zero Trust Architecture (the foundational publication this document expands)
  4. NIST CSRC: SP 800-204A — Building Secure Microservices-based Applications Using Service-Mesh Architecture (companion document cited by SP 800-207A)
  5. SPIFFE: overview of SPIFFE IDs and SVIDs (the workload identity example named in SP 800-207A)
Was this guide helpful?

In this cluster