01
Scanners invent finding counts. Without graph + runtime + exposure, teams drown in tickets while the one internet-facing path stays open.
Case 05 · Flagship
BlastGuard
Supply Chain Blast-Radius Control Plane: don't count CVEs — predict where they can reach, decide ALLOW/WARN/QUARANTINE/BLOCK, and prove remediation.
01
How it works
From CVE to a blast-radius decision.
BlastGuard by Andrey Lesnikov — predict supply-chain blast radius, return explainable gate decisions, and prove remediation before the exploit window matters.
-
01
Discover
Attach SBOM, vulnerability advisories, VEX, and Sigstore provenance — build-time truth enters the twin.
-
02
Trace
Walk CVE → package → image → workload → service → ingress. Prune VEX, not-running, and unreachable branches.
-
03
Decide & Enforce
The decision engine returns ALLOW / WARN / QUARANTINE / BLOCK with an explicit reason; CLI exit codes gate CI or GitOps promotion. Admission enforcement is a reference integration, not in-tree. No fake risk percentage.
-
04
Verify
After patch, recompute radius. Zero affected running paths means REMEDIATION PROVED. Rehearse images and CVEs before production changes.
02
Evidence
From problem to proof.
How the system is framed, operated, and proven — not a slide deck summary.
02
Discover SBOM/CVE/Sigstore truth, Trace through package→image→workload→ingress, Decide with explainable policy, gate CI / GitOps promotion through CLI exit codes, Verify remediation until blast radius is zero. In-cluster admission (Kyverno / policy-controller) is a reference integration fed by these decisions, not shipped in-tree.
03
BlastGuard does not replace Syft/Grype/GUAC/Sigstore — it sits above them. The cost is twin hygiene; the payoff is decisions operators can defend.
04
In v0.2.0: OpenVEX ingest from file or OCI attestation — statements bind to an exact sha256 image digest (prefix or name mismatch is rejected), signed attestations are verified with cosign key or certificate identity and fail closed. A reachability triad (package_present / network_reachable / execution_observed, each TRUE / FALSE / UNKNOWN) feeds the policy, and missing evidence never yields ALLOW alone. Empty or incomplete twin inventory fail-closed (PR #26): no-blast-radius ALLOW requires inventory_complete=true; stub/empty twins yield UNKNOWN / incomplete-inventory (exit 1). blastguard remediate --image reports safer workloads, remaining CVEs and required changes (CI run 38132763675).
Constraints
SBOM truth, VEX, Sigstore provenance, Kubernetes runtime, ingress exposure, explainable gate decisions, and remediation proof.
03
Related
Nearby systems.
SBOM Integration
Software supply-chain data is often generated late, stored separately, and disconnected from deployment decisions
SBOM on the delivery path, not quarterly
Policy as Code Guardrails
Security and platform rules are often discovered only after deployment or during reviews
Standards enforced in CI, not slides
Agent Security Contract
Agent and MCP stacks expand the attack surface beyond prompt text: tool misuse, poisoned tool descriptions, argument injection, and unapproved high-impact actions run without a pre-execution gate
0 false-ALLOW · 0 false-BLOCK on this corpus
04
FAQ
Questions operators ask.
A supply-chain blast-radius control plane. It correlates SBOM, CVE, VEX, runtime presence, and exposure into ALLOW, WARN, QUARANTINE, or BLOCK — then proves remediation.
Scanners count findings. BlastGuard traces where a CVE can reach across package → image → workload → ingress, then issues an explainable decision.
blastguard rehearse asks what-if before production: for a CVE or proposed image, it projects blast radius and policy outcome without changing the cluster.
Available for meaningful infrastructure conversations