AWS Marketplace and AWS Foundational Technical Review, October 2026
- Available in AWS Marketplace. ChangeGuard AI has a public SaaS listing in AWS Marketplace. The 14-day free trial covers up to 10 EKS environments and does not become a paid subscription; execution stays off unless you set it up. View the listing.
- AWS Foundational Technical Review approved. AWS approved ChangeGuard AI’s FTR on October 14, 2026, valid until October 12, 2028. It requires a Well-Architected review with no high risks in security, reliability or operational excellence. It is an AWS review, not a certification. See Known limitations and roadmap.
ChangeGuard Permit, Executors and Connected Environments — October 2026
Governed execution: judgment → Permit → Executor → observed outcome. The decision layer now has a control layer. Authority for any ChangeGuard-governed change is bound to a ChangeGuard Permit — a short-lived (default 10 minutes, never more than 15), single-use, ES256 token signed by a key held in AWS KMS and bound to one action, one target, one environment, one approval or policy, and one time window. Nothing executes without a consumed Permit, and VERIFIED requires a fresh observation of the environment.New
- Cloud Executor — ChangeGuard applies approved fixes to reachable EKS environments with nothing installed, through a per-environment execution role in your account that carries no AWS permissions, an EKS access entry mapped to a per-environment Kubernetes group, and
get/list/patchon workloads in exactly the namespaces you choose. Reading and acting are separate identities, always. - Local Executor — the same Permit model from inside your cluster, generated per environment, for topologies or policies that require local presence. Executors that do not speak the Permit protocol are served nothing and read upgrade required.
- Verify execution access — proves, per environment, that the Executor holds exactly the required verbs and is denied delete, create, exec, Secrets, impersonation, RBAC writes, nodes, wildcards and cluster-admin, and that it verifies a signed probe Permit. Re-verified every 12 hours.
- Advise and Auto per environment — Advise (default) requires a signed-in operator’s approval for each change; Auto authorizes only inside an explicit, versioned execution policy and falls back to Advise outside it. A policy edit revokes Permits issued under the old version.
- AWS Account Center — which AWS accounts ChangeGuard is connected to and whether each connection is healthy, with verify, discover, detect-regions, repair and reauthorize actions, in customer words.
- Connected Environments — one place to add, authorize, verify, repair and retire environments, showing status, evidence source, judgment readiness, execution readiness, latest Permit and mode. Fleet and Edge reporting the same cluster is one environment.
- Change record — every record now carries its authority, its Permit, the execution ladder (Permit issued → Execution started → Action applied → Verifying → Verified) and the Permit’s own events; retained CI verdicts are retrievable by durable repo-scoped identity.
Assurance
The Permit safety matrix, the authority and failure-mode suites and the per-environment identity suite run in every build; the customer-visible behaviours are exercised through the real UI by a blocking customer-security gate; and the whole path is walked against production daily by the live acceptance journey (latest complete run 18 / 18). See Testing and assurance.Chart 5.3.6 / Operator v5.3.5 — August 2026
Configurable collector resources, with safe defaults. The collector now runs with default resource settings — requests 25m CPU / 64Mi memory, limits 500m CPU / 512Mi memory — grounded in measured usage with headroom, and configurable per-field viaagent.collector.resources (standard Kubernetes requests/limits shape). Previously the collector ran with no requests or limits and the Helm value was unwired. An unparseable value falls back to that field’s default, so a typo can never make the collector unschedulable. Existing installs pick up the defaults on their next operator upgrade.
v5.4.16 — August 2026
One verdict vocabulary: SHIP / HOLD / BLOCK — everywhere. The change-prediction subsystem historically used SHIP / CAUTION / HOLD while the pre-flight surface used SHIP / HOLD / BLOCK — two vocabularies with conflicting tier meanings. As of this release, every surface (pre-flight, change predictions, prediction insights, the approval gate, and the CI validate endpoint) speaks SHIP / HOLD / BLOCK: SHIP = safe to ship, HOLD = caution/review, BLOCK = expected to fail (and, at Approve/Auto autonomy, gated until a human approves).Changes
- Prediction verdicts unified — new predictions emit SHIP / HOLD / BLOCK; historical rows were migrated in place exactly once (legacy worst-tier
HOLD→BLOCK, legacyCAUTION→HOLD), so insight metrics and history remain correct. /api/validategate decision fixed — a change gated by an unapproved failure prediction now returnsdecision: "BLOCK". Previously this path returned an undocumentedHOLD, which the documented “fail the job onBLOCK” CI pattern did not catch. If your pipeline matched onHOLDfor this case, update it toBLOCK.- CLI score bands — the mid-range score label is now
REVIEW(matching the documented “Review Recommended” range); score bands are not verdicts.
v3.9.19 — June 2026
Cross-cluster GitOps, a signed supply chain, and a single-artifact enterprise install.New Features
- Cross-cluster GitOps discovery — the in-cluster agent reads ArgoCD
Applicationresources and FluxKustomization,HelmRelease, andSourceresources directly over the Kubernetes API. No ArgoCD or Flux API token is required, so discovery works even on clusters the SaaS control plane can’t reach. GitOps state persists in Postgres and survives restarts and replicas. - Signed supply chain — the Helm chart, the operator and agent images, and a CycloneDX SBOM for each are signed with cosign (backed by AWS KMS). Verify any artifact with
cosign verify --key https://charts.changeguard.ai/cosign.pub. - Single Helm chart install — the chart is now the one install artifact (the standalone agent is removed; the one-liner is a thin Helm bootstrap). It is published to both the chart repo and as an OCI artifact, with a
global.imageRegistryoverride for air-gapped clusters and ready-made ArgoCD/Flux install examples. - Automatic SBOM collection — Syft SBOMs are harvested and posted automatically on every install, like the other scanners.
Fixes
- Customer compliance scoring now excludes ChangeGuard AI’s own namespaces and cluster RBAC, so the score reflects only your workloads.
- Cross-cluster ArgoCD fleet view and Flux GitOps state now render reliably across backend replicas.
v3.9.3 — May 21, 2026
Operator as default install method. The ChangeGuard AI operator now manages the full lifecycle — data collector, security scanners, runtime detection, and AI analysis — through a single ChangeGuardAgent CRD.New Features
- Security Scanning by default — KubeBench, Grype, Falco, Pluto, and Syft enabled on every install
- Identity & Access view — RBAC analysis with 55+ finding types
- Attack Path Analysis — privilege escalation graph
- Compliance frameworks — SOC 2, PCI DSS, HIPAA, FedRAMP, EO 14028 mapping
- Updated design system — dark navy UI with engineering-tool density
Fixes
- Agent RBAC permissions now include rbac.authorization.k8s.io by default
- Casbin middleware ordering fix
- Sidebar navigation routing fix for all 31 views
v3.9.2 — May 2026
- Score history persistence
- Slack and Microsoft Teams webhooks
- CSV and JSON export for compliance
- Self-service signup with 14-day free trial
v3.8.0 — April 2026
- NVIDIA NIM AI integration
- Flux CD deep integration
- Air-gapped install support
- Helm chart for operator
v3.7.0 — March 2026
- ArgoCD deep integration via CRD watching
- GitOps drift detection and sync history
- Cross-cluster image vulnerability correlation
v3.6.0 — February 2026
- Initial release: CSC scoring, agent collection, ArgoCD fleet management, multi-tenant SaaS, REST API