Continuum Resources LLC — Applied AI Research Series
WP-CR-2025-05  ·  Unclassified  ·  Public Release Authorized

Zero Trust Architecture
for CI/CD Pipelines

Practical Implementation for Federal Programs — Identity-First Security, Artifact Integrity, Secret Management, and Compliance Alignment for DevSecOps in DoD and Civilian Agency Environments

Author
Kurt A. Richardson, PhD
Affiliation
Head of R&D, Continuum Resources LLC
Published
March 2025
Classification
Unclassified // Public
Policy Basis
EO 14028 · NIST SP 800-207 · CISA ZTA
Scroll to read
Section 00

Executive Summary

Software delivery pipelines are the most consequential yet consistently under-secured component of modern federal IT infrastructure. The CI/CD pipeline is where source code becomes deployed capability — the crucible through which all software passes before it reaches operational systems. It is also, for exactly that reason, the highest-value target in the software supply chain. A compromised pipeline does not merely expose one system; it compromises every system that pipeline touches, with a trust-propagation radius that can span an entire enterprise.

Executive Order 14028 on Improving the Nation's Cybersecurity made this explicit: federal agencies must adopt Zero Trust Architecture and secure their software supply chains. CISA's Zero Trust Maturity Model and NIST SP 800-207 provide the frameworks. The National Security Memorandum on Improving Cybersecurity for Critical Infrastructure extends these requirements into the defense industrial base. For federal contractors and agencies operating CI/CD pipelines in support of government programs, Zero Trust is no longer an architectural aspiration — it is a compliance mandate with a deadline.

45%
of all enterprise breaches in 2023 involved the software supply chain or CI/CD pipeline
EO 14028
Federal mandate establishing Zero Trust and software supply chain security requirements
SLSA 4
Highest supply-chain integrity level — required for DoD mission-critical software per emerging DISA guidance
⚡ Core Finding

Most federal CI/CD pipelines are built on an implicit trust model — tokens and credentials that grant broad access, service accounts with standing permissions, build environments that run arbitrary code with elevated privileges, and artifact stores with no integrity verification. Applying Zero Trust to CI/CD requires rebuilding the pipeline's trust architecture from first principles: every identity verified, every action authorized, every artifact attested, and every secret dynamically provisioned.

Section 01

Why CI/CD Pipelines Are High-Value Targets

The SolarWinds attack of 2020 was a watershed moment for software supply chain security. The attackers did not breach SolarWinds' customers directly — they compromised the build pipeline and inserted malicious code into a software update that customers then downloaded and trusted. Roughly 18,000 organizations installed the backdoored update. The attack demonstrated, at devastating scale, that a compromised pipeline is worth more to an adversary than a compromised endpoint — because it delivers persistence into every downstream system simultaneously.

CI/CD pipelines in federal environments are particularly high-value targets because they serve as the delivery mechanism for software that runs on systems ranging from unclassified administrative infrastructure to mission-critical operational platforms. A pipeline that deploys code to a Space Force ground system, an Army logistics application, or a financial regulatory reporting platform carries an operational consequence profile that dwarfs any single endpoint it touches.

"The pipeline is not the delivery vehicle for software — the pipeline is the software. Securing the output without securing the process is architectural theater."
— Kurt A. Richardson, PhD, Head of R&D, Continuum Resources LLC

The Trust Problem in CI/CD

Traditional CI/CD security relies on perimeter trust: if you are inside the network and have credentials, you are trusted. This model fails catastrophically when a single token, service account, or build worker is compromised — because the blast radius of that compromise is every system those credentials can reach. Zero Trust replaces perimeter trust with continuous verification: no identity is inherently trusted, every access decision is evaluated in context, and the principle of least privilege is enforced at every step.

Applied to CI/CD, Zero Trust means: every pipeline stage must prove its identity and be authorized for each action; every artifact must be cryptographically attested and verified before deployment; every secret must be dynamically issued with minimum scope and automatically rotated; every build environment must be ephemeral and isolated; and every action must be logged to an immutable audit trail.

Federal Program Context

Continuum Resources operates CI/CD pipelines in active DoD programs — including the first Space Force Operational Acceptance under the Software Acquisition Pathway (DoD 5000.87). The practices in this paper are drawn from that operational experience, combined with our DevSecOps expertise and the published standards that now govern federal software delivery. This is not theoretical guidance — it is what Zero Trust CI/CD looks like in practice for federal programs.

Section 02

Zero Trust Principles Applied to CI/CD

NIST SP 800-207 defines seven tenets of Zero Trust Architecture. Each tenet has direct, specific implications for CI/CD pipeline design. The following maps each tenet to its pipeline-specific implementation requirement — the translation from principle to practice that is the purpose of this paper.

TENET 01 · NIST 800-207 §2.1
All Data Sources Are Resources
Every component of the CI/CD ecosystem — source repositories, artifact registries, build caches, secret stores, deployment targets — is a protected resource. No resource is implicitly trusted because it is on an internal network. Access to each requires explicit authorization at every interaction.
TENET 02 · NIST 800-207 §2.2
All Communication Is Secured
Every connection in the pipeline — developer to SCM, runner to artifact registry, deployer to target system — must be authenticated and encrypted, regardless of network location. mTLS between pipeline components; no plaintext credentials or tokens in transit.
TENET 03 · NIST 800-207 §2.3
Access Per-Session, Not Per-Identity
Pipeline access is granted per-session and per-action — not as standing permission attached to an identity. A build runner authorized to pull from a source repository is not therefore authorized to push to a production registry. Each action requires its own authorization decision.
TENET 04 · NIST 800-207 §2.4
Access Determined by Dynamic Policy
Authorization decisions incorporate device health, pipeline context, behavioral baseline, time, and data sensitivity — not just identity. A build runner operating outside its expected execution window, from an unexpected IP range, or with anomalous resource access patterns triggers re-verification.
TENET 05 · NIST 800-207 §2.5
Integrity and Security Posture Monitored
All pipeline components — build workers, registry services, deployment agents — are continuously monitored for configuration drift, vulnerability state, and behavioral anomalies. Components that fail posture checks are quarantined before they can affect the pipeline.
TENET 06 · NIST 800-207 §2.6
Authentication and Authorization Dynamic
Authentication is continuous, not point-in-time. Long-lived tokens are replaced with short-lived, automatically rotated credentials issued by OIDC or workload identity systems. Elevated access requires step-up authentication validated against the policy engine in real time.
TENET 07 · NIST 800-207 §2.7
Data Collection to Improve Security Posture
All pipeline telemetry — access events, build logs, artifact operations, deployment actions — is collected, correlated, and fed to a SIEM for behavioral baseline modeling and anomaly detection. The audit trail is immutable; the data informs continuous improvement of the Zero Trust policy engine.
Section 03

Federal Policy Landscape

The Zero Trust mandate for federal CI/CD pipelines derives from an interlocking set of executive orders, agency guidance documents, and emerging contract requirements. Understanding the specific obligations each creates is essential for compliance planning.

PolicyIssuerKey CI/CD RequirementApplicability
EO 14028White HouseZero Trust adoption, software supply chain security, SBOM, pipeline security for federal softwareAll federal agencies + contractors delivering software
OMB M-22-09OMBFederal ZTA strategy — specific milestones for identity, devices, networks, applications, data; agencies must achieve defined maturity levelsFederal agencies; propagates to system contractors
NIST SP 800-207NISTSeven ZTA tenets, architecture models (agent-based, enclave, DAAS), implementation guidanceReference standard for all federal ZT implementations
CISA ZT Maturity ModelCISAFive pillars (Identity, Devices, Networks, Applications, Data) with Traditional → Optimal maturity progression; specific CI/CD controls at each levelFramework for measuring progress; referenced in agency plans
NIST SP 800-218 (SSDF)NISTSecure Software Development Framework — practices for secure pipeline design, build environment integrity, and supply chain risk managementFederal software developers and contractors
SLSA FrameworkGoogle / OpenSSFSupply-chain Levels for Software Artifacts — 4-level provenance attestation framework now referenced in CISA and DoD guidanceEmerging DoD requirement for mission-critical software
CMMC 2.0DoDLevel 2: 110 practices aligned to NIST SP 800-171 including access control, audit, configuration management; Level 3 adds NIST SP 800-172 practicesDefense contractors handling CUI
NSM-8White HouseNational Security Systems — ZTA requirements for systems processing classified information; most stringent controlsNSS operators; highest-classification DoD systems
📅 Compliance Timeline Pressure

OMB M-22-09 established fiscal year milestones for Zero Trust adoption. Agencies were required to have specific ZTA capabilities in place by end of FY2024, with full maturity targets in FY2026. Defense contractors whose pipelines deliver software to federal systems are increasingly finding ZTA requirements appearing in solicitations, task orders, and PWS language. Programs that have not begun ZTA implementation for their CI/CD pipelines are already behind the compliance curve.

Section 04

CI/CD Pipeline Threat Model

Securing a CI/CD pipeline requires understanding what attackers are actually trying to accomplish and through which specific attack paths. The following threat model is structured around the pipeline's stages — each stage has a distinct attack surface and corresponding threat actors.

Stage 1 — Source Control
Developer Workstation
SCM (Git/GitLab/GitHub)
Branch Protection Rules
Code Review Process
⚠ Malicious Commits
⚠ Credential Theft
Source triggers pipeline — commit signature verification required
Stage 2 — Build Environment
CI Runner (ephemeral)
Build Scripts
Dependency Resolution
External Package Registries
⚠ Dependency Confusion
⚠ Build Script Injection
Build output signed — SBOM generated — provenance attested
Stage 3 — Artifact Registry
Container Registry
Binary Repository
Signature Verification
Vulnerability Scanning
⚠ Tampered Artifacts
⚠ Registry Poisoning
Artifact integrity verified — policy gate enforced before deploy
Stage 4 — Deployment
Deployment Agent
Target Environment
Policy Admission Control
Config Management
⚠ Deployment Token Abuse
⚠ Config Drift
Runtime monitoring active — all deployment events logged
Cross-Cutting Security Controls
OIDC Workload Identity
Vault / Secrets Manager
Immutable Audit Log
SIEM / SOAR
Policy Engine (OPA)
ZT Policy Enforcement Point
Figure 1 — Zero Trust CI/CD Pipeline Architecture — Attack surfaces annotated in red; ZT controls in green/teal

Top Pipeline Attack Vectors

Attack VectorDescriptionExampleSeverity
Dependency ConfusionAttacker publishes malicious package to public registry with same name as private internal package; build system resolves public over privateSolarWinds, PyPI typosquatting attacksCritical
Build Script InjectionMalicious code in Makefile, Dockerfile, CI YAML, or dependency install scripts executes arbitrary commands during build with pipeline credentialsXZ Utils backdoor (2024)Critical
Secret ExfiltrationBuild script or compromised dependency reads environment variables or mounted secrets and exfiltrates them to attacker-controlled endpointTravis CI token exposure incidentsCritical
Artifact TamperingAttacker with registry write access modifies a signed artifact after signing validation but before deployment verificationCodecov bash uploader compromiseCritical
Privileged Token TheftLong-lived deployment tokens or service account credentials stolen from configuration files, environment variables, or CI runner cacheGitHub Actions token leakage via logsHigh
Pipeline Poisoning (PPE)Attacker with pull request access modifies CI workflow files to execute arbitrary code in the pipeline's privileged contextGitHub Actions PR workflow attacksHigh
Runner CompromisePersistent build runner (non-ephemeral) is compromised via a prior build job and used to poison subsequent builds or exfiltrate secretsPersistent runner credential cachingHigh
SCM Account TakeoverDeveloper credential compromise via phishing, password reuse, or MFA bypass gives attacker code commit capability with signing trustLinus Torvalds key infrastructure attacksHigh
Section 05

Identity & Access Management for Pipelines

Identity is the foundation of Zero Trust — CISA's ZT Maturity Model places Identity as the first pillar, and for CI/CD pipelines it is the most critical. Pipeline security fails when any of three identity categories are inadequately managed: human identities (developers, operators), machine identities (service accounts, runners), and workload identities (individual pipeline jobs, agents, containers). All three must be managed with the same rigor.

Human Identity Controls

  • Phishing-resistant MFA mandatory: FIDO2/WebAuthn hardware keys or platform authenticators for all accounts with SCM access. TOTP-based MFA is insufficient for federal programs; OMB M-22-09 requires phishing-resistant MFA for all privileged access by FY2024.
  • Role-based branch protection: Every SCM repository enforces branch protection requiring minimum two-person review, signed commits, and status check passage before merge to protected branches. No developer has direct push to main or release branches.
  • Commit signing enforced: All commits to protected repositories must be signed with developer GPG keys or SSH signing keys registered in the organization's key management system. Unsigned commits are rejected at the SCM layer.
  • Privileged access workstations: Developers with access to production pipeline configuration use designated hardened workstations with endpoint detection, full-disk encryption, and no local admin privileges — consistent with CISA ZT Devices pillar requirements.
  • Just-in-time elevation: Access to pipeline administration, secret management interfaces, and deployment authorization requires just-in-time elevation through a PAM system — no standing privileged access for human identities.

Machine Identity: Replacing Long-Lived Tokens with OIDC

The single highest-impact change a federal CI/CD program can make is replacing long-lived service account tokens with OIDC-based workload identity. This eliminates the class of attack where a static token — once stolen — grants persistent access indefinitely. Modern CI/CD platforms (GitHub Actions, GitLab CI, Tekton) support OIDC token exchange natively: the pipeline job authenticates to the identity provider, receives a short-lived token scoped to the specific job's permissions, and exchanges it for cloud provider or vault credentials that expire at job completion.

✓ OIDC Workload Identity Pattern
  • No stored secrets: The pipeline never stores cloud credentials or vault tokens in environment variables, files, or the CI system's secret store. Credentials are dynamically issued at runtime and expire at job completion.
  • Job-scoped permissions: Each pipeline job receives only the permissions required for that specific job — a build job can pull from a private registry but cannot push to production; a deploy job can push to a staging environment but not production without a separate authorization step.
  • Auditability: Every credential issuance is logged by the identity provider with full job context — which repository, which branch, which workflow, which commit triggered the token issuance.
  • Automatic rotation: Because tokens are issued fresh per job, there is no rotation requirement — tokens are self-expiring by design.

Least-Privilege Service Account Architecture

Pipeline StageRequired PermissionsProhibited PermissionsToken Lifetime
Source CheckoutRead repository (specific SHA/branch)Write, admin, other repositoriesJob duration only
Dependency PullRead internal registry (approved packages only)Write to registry, access to prod registriesJob duration only
Build & TestRead source, write build cache, read test fixturesNetwork egress to internet, write to registryJob duration only
Artifact PushWrite to designated staging registry pathWrite to prod registry, delete operationsJob duration only
Deploy to StagingDeploy to staging namespace/clusterAccess prod environment, modify cluster config15 minutes max
Deploy to ProductionDeploy to prod (post approval gate)Modify infrastructure, access secrets store directly10 minutes max
Section 06

Artifact Integrity & Supply Chain Security

Artifact integrity is the guarantee that the software deployed to a production system is exactly the software produced by the authorized build process — unmodified, from a known-good source, built by an authenticated identity, and verified at every stage from build to deployment. In a Zero Trust pipeline, artifact integrity is enforced cryptographically, not procedurally.

The SLSA Framework

Supply-chain Levels for Software Artifacts (SLSA, pronounced "salsa") provides a graduated framework for supply chain integrity that is increasingly referenced in federal guidance. SLSA defines four levels of provenance and build integrity:

SLSA LevelBuild RequirementsProvenance RequirementsFederal Context
SLSA 1Build process documentedProvenance available (unsigned)Minimum baseline — any program
SLSA 2Hosted build service; version controlledSigned provenance from build serviceEO 14028 SBOM baseline equivalent
SLSA 3Hardened build service; auditable buildNon-forgeable provenance; isolated buildCMMC L2 / IL4 programs
SLSA 4Hermetic build; two-person review requiredReproducible builds; comprehensive provenanceMission-critical DoD / NSS programs

Sigstore and Cosign for Federal Pipelines

Sigstore is the emerging open standard for artifact signing in cloud-native environments, now used by the Linux Foundation, Python, npm, and numerous federal-adjacent projects. Its Cosign tool enables keyless signing of container images using OIDC identity, with signatures stored in a public transparency log (Rekor) that provides non-repudiation without requiring key management infrastructure.

For federal programs requiring air-gapped or classified environments, Sigstore can be deployed as an internal instance — the same cryptographic guarantees, none of the external dependencies. Continuum has implemented internal Sigstore deployments for DoD programs where external transparency logs are not acceptable.

Software Bill of Materials (SBOM)

EO 14028 mandates SBOM generation for software sold to the federal government. An SBOM is a complete, machine-readable inventory of all components in a software artifact — direct dependencies, transitive dependencies, and their versions. For Zero Trust CI/CD, SBOM is a security control, not just a compliance deliverable:

  • Automated vulnerability correlation: SBOMs are continuously scanned against NVD, OSV, and CISA KEV feeds. When a new vulnerability is published, automated tooling immediately identifies every deployed artifact containing the vulnerable component.
  • Policy gate enforcement: SBOMs are evaluated against a policy engine (e.g., OPA) before deployment. Artifacts containing components with CVSS scores above a threshold, components on CISA KEV, or unlicensed components are blocked from deployment without exception approval.
  • Dependency provenance: SBOMs include download source and integrity hash for every component — enabling detection of dependency confusion attacks where a component was sourced from an unexpected registry.
Section 07

Secret Management in Zero Trust Pipelines

Secrets — API keys, database credentials, TLS certificates, signing keys, deployment tokens — are the most targeted data class in CI/CD pipelines. The 2022 Codecov breach, the 2023 CircleCI incident, and numerous GitHub Actions token exposure events all share a common root cause: static secrets stored in ways that were accessible to attackers. Zero Trust secret management eliminates static secrets from the pipeline entirely where possible, and strictly controls the few that remain necessary.

"The correct number of static secrets in a Zero Trust CI/CD pipeline is zero. Every credential that exists as a persistent, stored string is a vulnerability waiting to be exploited. Dynamic, short-lived, automatically rotated credentials are the only acceptable design."
— Kurt A. Richardson, PhD, Continuum Resources LLC

Vault Architecture for Federal Pipelines

A secrets management platform (HashiCorp Vault, AWS Secrets Manager, or equivalent FedRAMP-authorized service) provides dynamic secret generation, short-lived credential issuance, automatic rotation, and comprehensive access auditing. The architecture for federal CI/CD should implement three vault tiers:

VAULT TIER 1
Dynamic Credentials
Database passwords, cloud provider credentials, and API tokens generated dynamically on-demand per pipeline job. Credentials expire at job completion. No static credentials for these resource types exist anywhere in the pipeline. Supported by HashiCorp Vault dynamic secrets engines and cloud provider OIDC integrations.
VAULT TIER 2
Short-Lived Certificates
TLS certificates, signing certificates, and mTLS client certs issued with short TTLs (hours, not years) via Vault PKI or an internal CA. Certificates are auto-renewed by pipeline agents before expiry. Revocation is immediate via CRL/OCSP. No long-lived certificates in pipeline trust stores.
VAULT TIER 3
Managed Static Secrets
Unavoidable static secrets (third-party API keys without OIDC support, legacy system credentials) stored in Vault with strict access policies, automatic rotation where possible, and usage auditing. These secrets are never in environment variables, config files, or SCM — only retrieved at runtime from Vault via the pipeline's workload identity.
VAULT TIER 4
Signing Keys (HSM-Backed)
Artifact signing keys, code signing certificates, and pipeline attestation keys stored in HSM-backed Vault or a dedicated KMS. Keys never leave the HSM — signing operations are performed within the HSM, with the signed output returned to the pipeline. Eliminates the key exfiltration threat entirely.

Secret Scanning and Leak Prevention

Despite best practices, secrets occasionally appear in code. Pre-commit hooks using tools like Gitleaks or detect-secrets scan every commit for patterns matching known secret formats before they reach the remote repository. At the CI level, a mandatory secret scanning stage runs against every build. For federal programs, any detected secret triggers an automatic incident response workflow — the secret is immediately revoked, not merely removed from the repository.

Section 08

Network Segmentation & Micro-Perimeters

Zero Trust does not eliminate network security — it redefines its role. Rather than a perimeter that separates "inside" from "outside," the network in a ZT architecture enforces micro-perimeters around each pipeline component, ensuring that a compromise of one component cannot be laterally exploited to reach others. CISA's ZT Maturity Model identifies the Network pillar as requiring isolation of pipeline workloads, encrypted intra-pipeline communication, and network-level policy enforcement aligned to workload identity.

Pipeline Network Architecture

  • Isolated build namespaces: Each CI job executes in an isolated Kubernetes namespace (or equivalent) with its own network policy. A build job for Repository A cannot communicate with the network resources of a build job for Repository B — lateral movement between concurrent builds is blocked at the network layer.
  • Egress allowlisting: Build environments have no default internet egress. All external access — to package registries, artifact stores, external APIs — is allowlisted by destination and protocol. Unexpected egress (a build job attempting to contact an external IP not on the allowlist) triggers immediate alert and job termination.
  • Service mesh with mTLS: Internal pipeline service-to-service communication (runner to Vault, deployer to registry, agent to SIEM) uses a service mesh (Istio, Linkerd) with mutual TLS. No plaintext service communication. Certificate rotation is automated; expired certificates fail closed rather than open.
  • Air-gap capability for classified builds: Pipelines processing classified or sensitive workloads are physically segmented from pipelines handling unclassified work. No shared infrastructure, no shared credentials, no shared registries between classification levels. This is a physical requirement, not a logical one.

Software-Defined Perimeter for Remote Developers

Federal program developers accessing CI/CD systems remotely must do so through a Software-Defined Perimeter (SDP) or Zero Trust Network Access (ZTNA) solution — not through traditional VPN. ZTNA verifies device health, identity, and behavioral context before granting per-application access, and continuously re-evaluates the access decision throughout the session. This eliminates the "VPN as trusted network" assumption that has been exploited in numerous federal breaches.

🌐 ZTNA vs. VPN for Federal Programs

Traditional VPN grants broad network access to an authenticated user — effectively extending the "trusted inside" perimeter to remote locations. ZTNA grants access only to specific applications, evaluated per-session against device health, identity context, and behavioral baseline. For federal CI/CD, ZTNA means a developer can access the SCM and artifact registry they need without being granted network-level access to production systems, secret stores, or other developers' project namespaces.

Section 09

Runtime & Container Security

Zero Trust does not end at deployment — it extends into the runtime environment where deployed artifacts execute. Container security, Kubernetes admission control, and runtime behavioral monitoring are the runtime expression of Zero Trust principles: verify the workload's identity and integrity at execution time, enforce minimum privilege for running containers, and detect behavioral anomalies that indicate compromise or misuse.

Admission Control and Policy Gates

Kubernetes admission controllers (OPA/Gatekeeper, Kyverno) enforce policy at deployment time — the last line of defense before an artifact enters the runtime environment. Federal CI/CD programs should implement admission policies that enforce:

  • Images must come from approved internal registries only — no direct pulls from public Docker Hub or untrusted sources
  • Images must have a valid Cosign/Sigstore signature from the CI pipeline's signing key
  • Images must not run as root; non-root user and non-privileged security context required
  • Images must have a current SBOM with no CVSS 9.0+ vulnerabilities or CISA KEV components
  • Pod specifications must define resource limits — no unbounded CPU or memory access
  • Read-only root filesystems required for all production workloads
  • Network policies must be present — pods without explicit network policy are denied egress

Runtime Behavioral Monitoring

Even a perfectly signed, policy-compliant container can behave maliciously at runtime — through a zero-day vulnerability, a logic bomb, or an adversarial input that triggers unexpected execution paths. Runtime security tools (Falco, Sysdig, Aqua) monitor container system calls against a behavioral baseline, alerting on anomalies such as unexpected network connections, file system writes to read-only paths, new process spawning, or privilege escalation attempts. For federal programs, runtime alerts feed directly to the SIEM for correlation with pipeline-level telemetry.

Ephemeral Build Environments

Every CI build job must run in an ephemeral, clean environment that is created at job start and destroyed at job completion. Persistent build runners — runners that persist between jobs — are a Zero Trust violation: they accumulate state, cached credentials, and artifacts from previous builds that can be accessed by subsequent builds. Ephemeral runners eliminate this attack surface by design. Cloud-native CI platforms and Kubernetes-based runner controllers (GitHub Actions Runner Controller, GitLab's operator) provide ephemeral runner support natively.

Section 10

Observability, Audit & SIEM Integration

Zero Trust without observability is not Zero Trust — it is Zero Trust theater. The policy enforcement decisions, identity verifications, access grants, artifact operations, and runtime behaviors of a ZT pipeline must be continuously collected, correlated, and analyzed. The NIST SP 800-207 Tenet 7 requirement — "Collect as much information as possible about the current state of enterprise assets, network infrastructure, and communications" — mandates comprehensive observability as a ZT requirement, not an operational luxury.

Pipeline Telemetry Architecture

Telemetry SourceEvents CapturedRetentionSIEM Use Case
SCM (Git/GitHub/GitLab)Commit events, PR actions, branch operations, access grants, failed authentication1 year minimumInsider threat detection, unauthorized code changes, access pattern anomalies
CI OrchestratorJob start/completion, runner assignment, artifact pulls/pushes, secret accesses, environment variable access1 year minimumBuild anomaly detection, secret access anomalies, runner behavior baselines
Vault / Secrets ManagerEvery secret access — who, what secret path, from which workload, timestamp, credential issued TTL3 years (federal)Credential abuse detection, unauthorized secret access, lateral movement indicators
Artifact RegistryImage push/pull/delete, tag mutations, signature verifications, scan results3 years minimumArtifact tampering detection, unauthorized registry access, image provenance audit
Kubernetes API ServerAll API calls — deployments, config changes, RBAC changes, pod exec, admission decisions1 year minimumPrivilege escalation detection, unauthorized workload changes, admission policy violations
Runtime (Falco)Container system calls — file opens, network connections, process spawns, privilege operations90 days hot, 1 year coldBehavioral anomaly detection, malware detection, data exfiltration attempts
Identity Provider (IdP)Authentication events, MFA use, token issuance, failed auth, suspicious login patterns3 years (federal)Account compromise detection, credential stuffing, impossible travel alerts

Federal SIEM Requirements

Federal programs are subject to CISA's Continuous Diagnostics and Mitigation (CDM) program, which requires agencies and their contractors to feed security telemetry to CDM dashboards. For CI/CD pipelines, this means pipeline telemetry must be formatted and routed to meet CDM data requirements — not just stored locally. Additionally, CMMC Level 2 requires a SIEM capable of correlating events across the pipeline, generating alerts for defined suspicious patterns, and supporting incident response investigation.

Section 11

Interactive Compliance Mapper

The following compliance mapper lets you track implementation status of Zero Trust CI/CD controls against the major federal standards. Select your applicable frameworks using the tabs, then mark each control as Implemented, Partial, N/A, or leave it as a gap. The progress bar reflects your overall compliance posture across all marked items.

Zero Trust CI/CD Compliance Tracker
EO 14028 · NIST SP 800-207 · CISA ZTM · CMMC 2.0 · SSDF
Section 12

Zero Trust CI/CD Maturity Assessment

The CISA Zero Trust Maturity Model defines a progression from Traditional (implicit trust) through Advanced to Optimal (full ZT enforcement). The following assessment maps that progression to the five CI/CD-specific capability domains, allowing organizations to self-assess their current maturity, identify gaps, and prioritize investments. Rate each capability from 1 (Traditional) to 4 (Optimal) by clicking the dots.

Overall ZT Maturity
Rate capabilities above
Progress to Optimal
Section 13

Implementation Roadmap

Zero Trust CI/CD implementation is a phased journey, not a rip-and-replace event. The following 18-month roadmap reflects Continuum's delivery methodology, sequenced to maximize security improvement per implementation effort while maintaining pipeline availability and minimizing disruption to ongoing program delivery.

P1
Months 1–3 · Foundation
Inventory, Assessment & Quick Wins

Enumerate all pipelines, service accounts, stored secrets, and tokens. Conduct threat model assessment. Identify all static long-lived credentials. Enable MFA for all SCM accounts. Implement commit signing. Deploy secret scanning pre-commit hooks. These actions eliminate the most exploited vulnerabilities with minimal pipeline disruption.

Pipeline Inventory Threat Model Phishing-Resistant MFA Commit Signing Secret Scanning Branch Protection
P2
Months 4–7 · Identity & Secrets
OIDC Workload Identity & Vault Deployment

Deploy the secrets management platform. Migrate pipeline stages from static credentials to OIDC-based dynamic credential issuance. Implement per-stage least-privilege service accounts. Replace long-lived tokens with short-lived, automatically expiring OIDC tokens. This phase eliminates the credential theft attack class.

Vault / KMS Deploy OIDC Workload Identity Dynamic Credentials Per-Stage RBAC HSM-Backed Signing Keys
P3
Months 8–11 · Artifact & Network
Supply Chain Integrity & Network ZT

Implement artifact signing (Cosign/Sigstore or equivalent). Deploy SBOM generation and automated vulnerability scanning. Implement OPA admission control for artifact verification at deploy time. Migrate build environments to ephemeral runners. Implement network micro-segmentation and egress allowlisting. Achieve SLSA 3 for all priority pipelines.

Cosign / Sigstore SBOM Generation OPA Admission Control Ephemeral Runners Egress Allowlisting mTLS Service Mesh
P4
Months 12–15 · Observability & Runtime
SIEM Integration & Runtime Security

Integrate all pipeline telemetry into SIEM. Deploy runtime behavioral monitoring (Falco or equivalent). Implement anomaly detection for pipeline behavior baselines. Connect to CDM feeds for federal programs. Deploy runtime admission policies (read-only root FS, non-root containers). Achieve CMMC Level 2 audit trail requirements.

SIEM Integration Falco Runtime Monitor Behavioral Baselines CDM Feed Runtime Policies Incident Playbooks
P5
Months 16–18 · Optimization & SLSA 4
Continuous Verification & Optimal Maturity

Pursue SLSA 4 for mission-critical pipelines. Implement reproducible builds. Deploy continuous pipeline security testing (automated red-team exercises against the pipeline itself). Establish governance structure for ZT pipeline policy — quarterly review, annual red-team, documented exception process. Prepare ATO documentation package for the pipeline's security posture.

SLSA 4 (Priority) Reproducible Builds Pipeline Red-Team ZT Policy Governance ATO Documentation Continuous Improvement
Section 14

The Continuum Approach

Continuum Resources has operated CI/CD pipelines in active DoD programs for over eight years — including the first Space Force Operational Acceptance under the Software Acquisition Pathway. Our DevSecOps practice is not a service offering built around a vendor tool stack — it is a delivery methodology refined through real program experience, documented in the process of earning some of the most rigorous software delivery validations in the federal government.

Zero Trust for CI/CD is not a product Continuum sells. It is an architecture Continuum has built, operated, defended, and continuously improved in environments where the security requirements are defined by DoD directives, DISA STIGs, and program-specific ATOs. When we implement OIDC workload identity, ephemeral runners, Cosign artifact signing, or SIEM integration for a client program, we are applying patterns we have already proven operational.

✓ Continuum DevSecOps Services
  • ZT Pipeline Assessment: Gap assessment of current CI/CD pipelines against the ZT controls in this paper, mapped to EO 14028, NIST SP 800-207, CISA ZTM, and CMMC requirements. Deliverable: prioritized remediation roadmap with effort estimates and risk ratings.
  • Pipeline Design & Implementation: End-to-end design and delivery of ZT-native CI/CD pipelines — OIDC identity, Vault secret management, artifact signing, SBOM generation, admission control, ephemeral runners, and SIEM integration. Built to the program's specific classification level and compliance requirements.
  • ATO Documentation Support: Development of the pipeline security documentation package required for Authority to Operate — system security plan sections, control implementation statements, configuration baselines, and continuous monitoring procedures.
  • DevSecOps Maturity Assessment: Independent assessment of a program's DevSecOps maturity against CISA ZTM, DISA DevSecOps Fundamentals Guide, and DoD Software Acquisition Pathway requirements. Supports program management reporting and audit readiness.
  • Automated Testing Integration: Continuum's Automated Testing capability (see companion white papers) integrates directly with ZT CI/CD pipelines — security testing, functional testing, and compliance testing as native pipeline stages with policy-gated deployment decisions.
  • Continuous Security Program: Ongoing pipeline security management including quarterly configuration reviews, pipeline red-team exercises, vulnerability remediation SLA management, and ZT policy governance support.

Engagement Models

EngagementScopeDurationOutcome
ZT Pipeline AssessmentCurrent state gap analysis against ZT controls; threat model; risk-prioritized remediation roadmap3–4 weeksDocumented gaps, risk ratings, implementation roadmap, ATO gap analysis
Pipeline Hardening SprintImplementation of Phase 1–2 controls (identity, secrets, branch protection) on existing pipelines6–8 weeksOIDC workload identity, Vault deployment, MFA enforcement, secret scanning active
Full ZT Pipeline BuildEnd-to-end ZT-native pipeline delivery for a program — all five phases of the roadmap12–18 monthsFully compliant ZT pipeline with ATO documentation, SIEM integration, and governance
DevSecOps Continuous SupportOngoing pipeline security management, quarterly reviews, pipeline red-team exercisesOngoingSustained ZT posture with documented security history for ATO maintenance
Section 15

Conclusion

The CI/CD pipeline is the single point through which all software passes on its way to operational systems. A Zero Trust architecture for that pipeline — verified identities at every stage, cryptographically attested artifacts, dynamically issued secrets, isolated build environments, and comprehensive observability — is not a security enhancement. It is the minimum viable security posture for any federal program whose pipeline produces software that operates in a threat environment.

The regulatory mandate is clear: EO 14028, OMB M-22-09, and the CISA Zero Trust Maturity Model have established the requirement. The technical path is available: OIDC workload identity, Vault-based dynamic secrets, Sigstore artifact signing, and Kubernetes-native admission control are mature, well-documented, and deployable today. The SLSA framework provides a measurable progression from current-state to full supply chain integrity attestation. What remains is the organizational will to prioritize pipeline security with the same rigor applied to endpoint and network security — and to begin before the compliance deadline, not after it.

A Zero Trust pipeline does not slow software delivery. Done correctly, it accelerates it — because every stage has verified identity, attested artifacts, and dynamic credentials. The overhead of verification is lower than the overhead of investigating a compromised deployment.
— Kurt A. Richardson, PhD, Continuum Resources LLC, 2025
Start a Conversation

Ready to Build a Zero Trust Pipeline?

Contact Continuum Resources for a complimentary ZT Pipeline Assessment tailored to your program's classification level, compliance requirements, and current DevSecOps maturity.

Get in Touch →
References

References

  • [EO-14028] Executive Order 14028 — "Improving the Nation's Cybersecurity" — White House, May 12, 2021. The foundational mandate for Zero Trust adoption and software supply chain security in federal programs.
  • [OMB-M-22-09] OMB Memorandum M-22-09 — "Moving the U.S. Government Toward Zero Trust Cybersecurity Principles" — January 2022. Federal ZTA strategy with FY-specific milestones.
  • [NIST-800-207] National Institute of Standards and Technology — "Zero Trust Architecture" — SP 800-207, August 2020. The definitive technical reference for Zero Trust architecture implementation.
  • [CISA-ZTM] CISA — "Zero Trust Maturity Model v2.0" — April 2023. Five-pillar maturity progression framework for federal ZTA implementation.
  • [NIST-800-218] National Institute of Standards and Technology — "Secure Software Development Framework (SSDF) v1.1" — SP 800-218, February 2022. Practices for secure pipeline design referenced in EO 14028 implementation guidance.
  • [SLSA] OpenSSF / Google — "Supply-chain Levels for Software Artifacts (SLSA)" — slsa.dev, 2023. Four-level artifact provenance framework referenced in federal supply chain guidance.
  • [SIGSTORE] Sigstore Project — "Sigstore: A Non-Profit, Public Good Artifact Signing and Transparency Service" — sigstore.dev, 2023. The keyless signing standard for container artifacts.
  • [CMMC-2] Department of Defense — "Cybersecurity Maturity Model Certification (CMMC) 2.0" — 32 CFR Part 170, October 2024. Three-level certification framework for defense contractors handling CUI.
  • [DISA-DEVSECOPS] DISA — "DevSecOps Fundamentals Guide" — October 2021. DoD guidance for secure software delivery pipeline implementation.
  • [NIST-SSCS] National Institute of Standards and Technology — "Cybersecurity Supply Chain Risk Management Practices" — SP 800-161r1, May 2022.
  • [CR-01] Richardson, K.A. — "WP-CR-2025-01: Agentic AI in Mission-Critical Environments" — Continuum Resources, 2025. Agentic AI security architecture that relies on ZT CI/CD for pipeline delivery.
  • [CR-04] Richardson, K.A. — "WP-CR-2025-04: Prompt Injection & Adversarial Attacks on LLM Systems" — Continuum Resources, 2025. Adversarial threat model for AI systems delivered via ZT pipelines.