Join our Newsletter — 33% off our NHI Course
Home› Guides› SBOM operations: Running an SBOM and VEX Pipeline…
Guide Software & AI Supply Chain

SBOM operations: Running an SBOM and VEX Pipeline That Finds Real Exposure

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 28 October 2026 14 min read
On this page

Most organisations that collect SBOMs end up with a folder of JSON files that nobody queries until the next big open source incident, and then discover the files are stale, incomplete or impossible to match to what is actually running. SBOM operations is the work that turns SBOMs into an answer to "where am I affected?" within hours: validating what you receive, storing it against the right product and version, matching components to vulnerability data, applying supplier VEX, and routing the findings that remain to an owner. This guide assumes you already know what an SBOM and a VEX document are and focuses on running the pipeline day to day. It is written for product security teams, vulnerability management leads, platform engineers and procurement teams that receive supplier SBOMs.

Key takeaways

  • Treat every SBOM as untrusted input. Validate the schema, check the author, tool and generation context, and record what the document does not tell you before it reaches your inventory.
  • One SBOM per release, tied to a deployed version. An SBOM only answers exposure questions if you can join it to the artefact digest or version that is running today, and retire it when that version is gone.
  • Matching quality decides everything. Package URLs (purl) and hashes give reliable matches; name and version strings and CPE guesses generate both false positives and misses.
  • VEX suppresses, it never deletes. Apply supplier and internal VEX as an auditable analysis state, keep the source document, and reopen the finding when the statement changes.
  • Measure time to answer. The real test of an SBOM programme is how long it takes to list every affected product and owner after a new critical vulnerability is published.

What changes when you move from SBOM collection to SBOM operations

Generating SBOMs is covered in the SBOM Guide, and the VEX status model and formats in the VEX Guide. Operations adds four obligations that collection alone does not meet:

  • Freshness: the CISA draft 2025 minimum elements expect each software release to have its own SBOM, and a revised SBOM when new component details emerge or errors are found. Your store must replace or version documents, not just append them.
  • Continuous analysis: vulnerability data changes daily even when your SBOMs do not. OWASP Dependency-Track, for example, analyses components when a BOM is ingested and also re-analyses every component automatically on a daily basis.
  • Join to reality: an SBOM for a release you no longer run is noise. The operational inventory must know which product versions are deployed, where, and who owns them.
  • Decision history: every "not affected" or "false positive" decision needs an author, a timestamp and a reason that survives audit and can be reversed.

CISA's draft 2025 minimum elements (a pre-decisional public comment draft published in August 2025, comment period closed 3 October 2025; check CISA for the final version) also add four data fields that matter directly to operations: component hash, license, tool name and generation context (before build, during build or after build). Those fields are what let an operations team judge how far to trust a document.

Reference architecture for an SBOM pipeline

StageWhat happensTypical toolingControl that matters
ProduceBuild generates CycloneDX or SPDX SBOM per artefactBuild plugins, Syft, CycloneDX generatorsGenerate during build from the lock file or artefact, not from a developer laptop
AttestSBOM signed and bound to the artefact digestin-toto attestations via Sigstore cosign (cosign attest, cosign verify-attestation)Verification fails closed if the attestation is missing
ReceiveSupplier SBOMs arrive with software, by URL, API or portalProcurement intake, supplier portal, object storageSchema validation, author check, quarantine on failure
StoreVersioned per product and release, immutable originalsSBOM platform (for example Dependency-Track projects with parent and child versions), object storeWrite access only for pipeline identities; retain originals
AnalyseMatch components to vulnerability sources, licences and policyDependency-Track, Grype (grype sbom:./sbom.json), TrivyRe-analyse daily; record data source and match method
Apply VEXSupplier and internal statements set finding statesCycloneDX VEX upload, OpenVEX filters in scannersSuppression is logged, attributable and reversible
Route and reportRemaining findings go to the owning team; portfolio queries answer incidentsTicketing, notifications, GUAC graph queriesEvery product version has a named owner

For large estates that hold SBOMs, SLSA provenance, VEX and scorecards side by side, OpenSSF GUAC (an incubating project under the Supply Chain Integrity working group) aggregates that metadata into a graph so you can query relationships across artefacts. It is an aggregation layer, not a replacement for the analysis and triage platform.

Ingesting and validating SBOMs

Run every inbound document, internal or supplier, through the same intake gate before it is analysed:

  1. Format and version: accept the widely used machine-readable formats (SPDX and CycloneDX). The CISA draft advises against accepting new SBOMs in deprecated format versions and drops SWID as a format. Validate against the official JSON schema for the declared spec version and reject documents that fail.
  2. Integrity and origin: for internal SBOMs, verify the signed attestation against the artefact digest. For supplier SBOMs, record how it was delivered (shipped with the software, version-specific URL, API or repository) and check any signature the supplier provides.
  3. Minimum fields: check that the SBOM author, software producer, component name, version, at least one software identifier (purl or CPE), dependency relationships and timestamp are present. Record whether component hashes, licences, tool name and generation context are supplied.
  4. Coverage and known unknowns: the draft expects all components including transitive dependencies, and asks authors to flag incomplete dependency lists and distinguish unknown data from deliberately redacted data. Record those flags as data quality attributes rather than assuming completeness.
  5. Bind to a product version: attach the document to a product, version and, where you have it, the deployed artefact digest. Dependency-Track supports uploading by project UUID or by project name and version, and can create child projects under a parent, which suits one parent per product with a child per release.
  6. Quarantine failures: documents that fail validation go to a quarantine queue with an alert. In Dependency-Track the BOM_VALIDATION_FAILED and BOM_PROCESSING_FAILED notifications exist for exactly this purpose; subscribe to them.

Generation context matters for trust: a build-time SBOM produced from the resolved dependency graph is usually more complete than a post-build scan of a binary, which may miss statically linked or vendored code. Score documents accordingly.

Matching components to vulnerabilities

IdentifierMatch qualityOperational notes
Package URL (purl)High for package ecosystemsEcosystem, namespace, name and version are explicit; preferred for npm, PyPI, Maven, Go and OS packages
Component hashHigh for exact artefactsIdentifies the exact file; Dependency-Track policy can match MD5, SHA-1, SHA-2, SHA-3, BLAKE2b and BLAKE3; useful for counterfeit or known malicious components
CPEMedium, prone to noiseNeeded for NVD matching of commercial and OS software; vendor and product strings often drift
Name and version onlyLowAmbiguous across ecosystems; treat as a data quality defect and push suppliers for purl

Use more than one vulnerability source. Dependency-Track mirrors sources such as the NVD, GitHub Advisories and OSS Index, and Grype adds EPSS and KEV context to results. Record which source and which identifier produced each match, because "false positive" analysis usually comes down to a bad CPE or a misidentified component. For prioritisation, combine severity with exploitation evidence: see the CISA KEV Catalogue Guide and the EPSS Guide.

Applying VEX and recording triage decisions

CycloneDX 1.6 expresses impact analysis with six states (resolved, resolved_with_pedigree, exploitable, in_triage, false_positive, not_affected) and nine justifications, including code_not_present, code_not_reachable, requires_configuration, protected_at_runtime and protected_by_mitigating_control. Dependency-Track uses a similar set of analysis states on each finding (EXPLOITABLE, IN_TRIAGE, FALSE_POSITIVE, NOT_AFFECTED, NOT_SET) and keeps an audit trail of every change with the user and timestamp.

  • Supplier VEX: import it against the matching product version, set the finding state from the statement, and keep a link to the source document. Reject a "not affected" statement that carries no justification.
  • Internal VEX: when your own team decides a finding is not exploitable, record the justification in the platform. Dependency-Track can generate CycloneDX VEX from those audit decisions, which you can then publish to your own customers.
  • Disagreement: if your analysis contradicts a supplier statement, keep your own state, raise it with the supplier and record both. Dependency-Track's documentation describes consumers sharing their own VEX back with suppliers.
  • Reopening: a new supplier statement, a new SBOM version or a KEV listing should reopen the finding. Subscribe to PROJECT_AUDIT_CHANGE and GLOBAL_AUDIT_CHANGE notifications so changes are visible.

Suppressed findings do not raise security policy violations in Dependency-Track, which is exactly why suppression must be governed: an unreviewed suppression silently removes a finding from every gate that depends on it. The Reachability Analysis Guide covers the evidence that should sit behind a code_not_reachable claim.

Policy gates and routing

Dependency-Track evaluates policies when an SBOM is uploaded and groups violations into three types: licence, security (by vulnerability severity) and operational (allow or deny lists by coordinates, purl, CPE, SWID tag ID or hash). Use them at three points:

  • Build gate: fail the pipeline on known malicious components (hash or purl deny list), forbidden licences and new exploitable criticals with a fix available.
  • Procurement gate: require an SBOM that passes intake validation before a new supplier release is approved, as part of contract terms. See the Software Supply Chain Security Guide.
  • Runtime routing: send NEW_VULNERABILITY, NEW_VULNERABLE_DEPENDENCY and POLICY_VIOLATION notifications to the owning team's queue, not a shared inbox, and use the scheduled summary notifications for management reporting.

Where identity fits

An SBOM platform is a high-value target and is fed almost entirely by non-human identities. Dependency-Track API keys belong to teams, a team may hold several keys, and from version 4.13 keys are stored hashed so they are visible only at creation. Practical rules: one API key per pipeline or integration with only the permissions it needs (upload is not the same as VULNERABILITY_ANALYSIS, which is required to audit findings), keys held in a secrets manager, rotation on a schedule, and alerts on uploads from unexpected sources. A stolen upload key lets an attacker replace a product's SBOM with a clean one and hide a vulnerable component; a stolen analysis key lets them mark findings not affected. See the CI/CD Pipeline Identity Security Guide.

How to implement

  1. Weeks 1 to 4, pilot: pick five products with active owners. Generate build-time SBOMs, upload them to a single platform, and turn on daily re-analysis. Do not gate builds yet.
  2. Weeks 5 to 8, intake and identity: add schema validation, signed attestations for internal SBOMs, per-pipeline API keys and failure notifications. Map each project to an owner and a deployed version.
  3. Weeks 9 to 12, triage discipline: define who may set NOT_AFFECTED and FALSE_POSITIVE, require justifications, and import supplier VEX for the first critical suppliers.
  4. Quarter 2, gates in warn mode: enable policy evaluation in the build, report violations without failing, and tune false positives from bad identifiers.
  5. Quarter 3, enforce and scale: fail builds on the agreed policy, add supplier SBOM requirements to procurement, and run an incident drill: pick a recent critical CVE and time how long it takes to list every affected product and owner.
  6. Rollback: if a policy blocks releases unexpectedly, switch that policy back to report-only, keep the violation data, and fix the identifier or rule before re-enabling. Never roll back by deleting SBOMs or bulk suppressing findings.

Decision guidance

  • Platform or scanner only: a scanner such as Grype or Trivy run against SBOMs in CI is enough for a small team with a handful of products. Once you need portfolio queries, triage history and supplier SBOMs, you need a platform that stores documents and decisions.
  • Embedded or separate VEX: separate VEX documents are easier to update without reissuing the SBOM, which suits fast-changing advisories; embedded analysis is simpler for small internal products.
  • Supplier SBOMs: start by requiring them only for critical suppliers and internet-facing products, where the exposure question is most urgent.

Detection and measurement

  • Time to answer: hours from publication of a critical CVE to a complete list of affected products, versions and owners. When the XZ Utils backdoor affected only versions 5.6.0 and 5.6.1, teams with current SBOMs could answer this question quickly; teams without them searched hosts by hand.
  • Coverage: percentage of deployed product versions with a current SBOM no older than the deployed release.
  • Identifier quality: percentage of components with a purl or hash; track by supplier.
  • Intake failure rate: documents rejected or quarantined per month, by source.
  • Suppression hygiene: percentage of suppressed findings with a justification and a named author; count of suppressions older than 12 months without review.
  • Anomalies to alert on: an SBOM upload that removes a previously vulnerable component without a matching version change, uploads from an unfamiliar API key, and bulk state changes to NOT_AFFECTED.

Framework mapping

FrameworkReferenceHow SBOM operations supports it
NIST CSF 2.0ID.AM-02, ID.AM-08Software inventories maintained and managed through their life cycle
NIST CSF 2.0ID.RA-01, GV.SC-05Vulnerabilities identified and recorded; supply chain requirements built into supplier agreements
NIST SP 800-53 Rev. 5CM-8, RA-5System component inventory; vulnerability monitoring and scanning
NIST SP 800-53 Rev. 5SR-3, SR-4, SR-11Supply chain controls, provenance and component authenticity
MITRE ATT&CKT1195.001, T1195.002Faster scoping when software dependencies or the software supply chain are compromised

Common failure patterns

  • SBOM graveyard: documents stored but never re-analysed, so a new CVE in an old component is invisible until someone remembers to rescan.
  • No join to deployment: SBOMs exist for every build, but nobody knows which build is in production, so exposure lists are either empty or include every historic version.
  • Trusting supplier VEX blindly: importing "not affected" statements without a justification check, which removes findings from every gate.
  • Developer-generated SBOMs: SBOMs produced on a laptop or from a manifest rather than the resolved build, missing transitive and vendored code.
  • Shared admin key for uploads: one long-lived administrator API key used by every pipeline, so a single leak allows SBOM tampering across the portfolio. Leaked CI tokens in the tj-actions/changed-files compromise show how quickly pipeline secrets spread.

Practitioner checklist

  • Generate SBOMs at build time from the resolved dependency graph and sign them as attestations bound to the artefact digest.
  • Validate every inbound SBOM against the official schema and the minimum fields, and quarantine failures with an alert.
  • Store one SBOM per release against a product, version, deployed digest and named owner.
  • Re-analyse all components at least daily against more than one vulnerability source.
  • Prefer purl and hash matching; track CPE-only and name-only components as data quality debt.
  • Import supplier VEX as an auditable state, reject unjustified "not affected", and reopen on change.
  • Issue one scoped API key per pipeline, keep keys in a secrets manager and alert on unexpected uploaders.
  • Run policy gates in warn mode first, then enforce, with a documented rollback.
  • Drill the "where am I affected" question quarterly and record the time to answer.

Standards and references

Related NHI Mgmt Group resources: SBOM Guide · VEX Guide · Vulnerability Management Guide · SLSA and Build Provenance Guide · Software Supply Chain Maturity Model

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 28 October 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.