diff --git a/apps/ai-guard/base/deployment.yaml b/apps/ai-guard/base/deployment.yaml
index e86f27b..03bb296 100644
--- a/apps/ai-guard/base/deployment.yaml
+++ b/apps/ai-guard/base/deployment.yaml
@@ -1,3 +1,13 @@
+# Scaled to 0 (Phase 0 of the AI architecture doc — "Stabilize the Existing
+# Environment"): ai-guard has no Dockerfile in its source repo, so no image
+# has ever been successfully built by its CI. The pod has been in
+# ImagePullBackOff for 14+ days (90,000+ failed pulls) pulling an image that
+# doesn't exist. It also has no Service (unreachable even if healthy) and
+# depends on llm-guard/presidio, neither of which are deployed. Nothing
+# currently routes through it — both AI consumers (nxtgauge-ai-assistant,
+# nxtgauge-backend-rust) call LiteLLM directly. Scaling to 0 stops the
+# wasted kubelet pull-retry churn until Phase 3 (Rebuild ai-guard) is done
+# properly, per the architecture doc.
 apiVersion: apps/v1
 kind: Deployment
 metadata:
@@ -6,7 +16,7 @@ metadata:
   labels:
     app: ai-guard
 spec:
-  replicas: 1
+  replicas: 0
   selector:
     matchLabels:
       app: ai-guard