diff --git a/src/runtime/shared-app-server-state.ts b/src/runtime/shared-app-server-state.ts
index 27de5948..239fecbc 100644
--- a/src/runtime/shared-app-server-state.ts
+++ b/src/runtime/shared-app-server-state.ts
@@ -32,34 +32,62 @@ export function createSharedAppServerState(): SharedAppServerState {
 export function applySharedThreadList(state: SharedAppServerState, threads: readonly Thread[]): SharedAppServerState {
   return {
     ...state,
-    threads: { kind: "loaded", data: threads },
+    threads: { kind: "loaded", data: cloneThreads(threads) },
   };
 }
 
 export function applySharedAppServerMetadata(state: SharedAppServerState, metadata: SharedAppServerMetadata): SharedAppServerState {
+  const clonedMetadata = cloneSharedAppServerMetadata(metadata);
   return {
     ...state,
-    appServerMetadata: { kind: "loaded", data: metadata },
-    availableModels: metadata.availableModels,
+    appServerMetadata: { kind: "loaded", data: clonedMetadata },
+    availableModels: cloneModels(clonedMetadata.availableModels),
   };
 }
 
 export function applySharedModels(state: SharedAppServerState, models: readonly Model[]): SharedAppServerState {
+  const clonedModels = cloneModels(models);
   return {
     ...state,
     appServerMetadata:
       state.appServerMetadata.kind === "loaded"
-        ? { kind: "loaded", data: { ...state.appServerMetadata.data, availableModels: models } }
+        ? { kind: "loaded", data: { ...state.appServerMetadata.data, availableModels: cloneModels(clonedModels) } }
         : state.appServerMetadata,
-    availableModels: models,
+    availableModels: clonedModels,
   };
 }
 
 export function cachedSharedThreadList(state: SharedAppServerState): readonly Thread[] | null {
-  return state.threads.kind === "loaded" ? state.threads.data : null;
+  return state.threads.kind === "loaded" ? cloneThreads(state.threads.data) : null;
 }
 
 export function cachedSharedAppServerMetadata(state: SharedAppServerState): SharedAppServerMetadata | null {
-  if (state.appServerMetadata.kind === "loaded") return state.appServerMetadata.data;
+  if (state.appServerMetadata.kind === "loaded") return cloneSharedAppServerMetadata(state.appServerMetadata.data);
   return null;
 }
+
+function cloneSharedAppServerMetadata(metadata: SharedAppServerMetadata): SharedAppServerMetadata {
+  return {
+    ...metadata,
+    availableModels: cloneModels(metadata.availableModels),
+    availableSkills: metadata.availableSkills.map((skill) => ({ ...skill })),
+    appServerDiagnostics: {
+      probes: { ...metadata.appServerDiagnostics.probes },
+      mcpServers: metadata.appServerDiagnostics.mcpServers.map((server) => ({ ...server })),
+    },
+  };
+}
+
+function cloneThreads(threads: readonly Thread[]): Thread[] {
+  return threads.map((thread) => ({ ...thread, turns: [...thread.turns] }));
+}
+
+function cloneModels(models: readonly Model[]): Model[] {
+  return models.map((model) => ({
+    ...model,
+    supportedReasoningEfforts: [...model.supportedReasoningEfforts],
+    inputModalities: [...model.inputModalities],
+    additionalSpeedTiers: [...model.additionalSpeedTiers],
+    serviceTiers: [...model.serviceTiers],
+  }));
+}
diff --git a/tests/runtime/runtime-settings.test.ts b/tests/runtime/runtime-settings.test.ts
index 3f8db678..173d923e 100644
--- a/tests/runtime/runtime-settings.test.ts
+++ b/tests/runtime/runtime-settings.test.ts
@@ -1,6 +1,9 @@
 import { describe, expect, it } from "vitest";
 
 import type { ConfigReadResponse } from "../../src/generated/app-server/v2/ConfigReadResponse";
+import type { Model } from "../../src/generated/app-server/v2/Model";
+import type { Thread } from "../../src/generated/app-server/v2/Thread";
+import { createAppServerDiagnostics } from "../../src/app-server/compatibility";
 import {
   compactContextLabel,
   compactModelLabel,
@@ -27,6 +30,14 @@ import {
 } from "../../src/runtime/state";
 import { readRuntimeConfig } from "../../src/runtime/config";
 import { contextSummary, effectiveConfigSections, rateLimitSummary } from "../../src/runtime/view";
+import {
+  applySharedAppServerMetadata,
+  applySharedModels,
+  applySharedThreadList,
+  cachedSharedAppServerMetadata,
+  cachedSharedThreadList,
+  createSharedAppServerState,
+} from "../../src/runtime/shared-app-server-state";
 
 describe("runtime settings", () => {
   it("parses model overrides", () => {
@@ -35,6 +46,38 @@ describe("runtime settings", () => {
     expect(parseModelOverride("")).toBeUndefined();
   });
 
+  it("keeps shared app-server cache snapshots detached from caller-owned arrays", () => {
+    const sourceThreads = [threadFixture("thread-1")];
+    const threadState = applySharedThreadList(createSharedAppServerState(), sourceThreads);
+    sourceThreads.push(threadFixture("thread-2"));
+
+    const cachedThreads = cachedSharedThreadList(threadState);
+    expect(cachedThreads?.map((thread) => thread.id)).toEqual(["thread-1"]);
+
+    const mutableCachedThreads = cachedThreads as Thread[];
+    mutableCachedThreads.push(threadFixture("thread-3"));
+    expect(cachedSharedThreadList(threadState)?.map((thread) => thread.id)).toEqual(["thread-1"]);
+
+    const sourceModels = [modelFixture("gpt-5.5")];
+    const modelState = applySharedModels(createSharedAppServerState(), sourceModels);
+    sourceModels.push(modelFixture("gpt-5.6"));
+    expect(modelState.availableModels.map((model) => model.model)).toEqual(["gpt-5.5"]);
+
+    const metadataState = applySharedAppServerMetadata(createSharedAppServerState(), {
+      effectiveConfig: null,
+      availableModels: sourceModels,
+      availableSkills: [{ name: "skill", description: "", path: "/tmp/skill", scope: "repo", enabled: true }],
+      rateLimit: null,
+      appServerDiagnostics: {
+        ...createAppServerDiagnostics(),
+        mcpServers: [{ name: "server", startupStatus: "ready", authStatus: null, toolCount: 1, message: null }],
+      },
+    });
+    sourceModels.push(modelFixture("gpt-5.7"));
+    const cachedMetadata = cachedSharedAppServerMetadata(metadataState);
+    expect(cachedMetadata?.availableModels.map((model) => model.model)).toEqual(["gpt-5.5", "gpt-5.6"]);
+  });
+
   it("parses reasoning effort overrides", () => {
     expect(parseReasoningEffortOverride("high")).toBe("high");
     expect(parseReasoningEffortOverride("default")).toBeNull();
@@ -591,3 +634,48 @@ function configLayer(config: Record<string, unknown>, profile: string | null): N
     disabledReason: null,
   };
 }
+
+function threadFixture(id: string): Thread {
+  return {
+    id,
+    sessionId: "session",
+    forkedFromId: null,
+    preview: "",
+    ephemeral: false,
+    modelProvider: "openai",
+    createdAt: 1,
+    updatedAt: 1,
+    status: { type: "idle" },
+    path: null,
+    cwd: "/vault",
+    cliVersion: "0.0.0",
+    source: "appServer",
+    threadSource: null,
+    agentNickname: null,
+    agentRole: null,
+    gitInfo: null,
+    name: null,
+    turns: [],
+  };
+}
+
+function modelFixture(model: string): Model {
+  return {
+    id: model,
+    model,
+    upgrade: null,
+    upgradeInfo: null,
+    availabilityNux: null,
+    displayName: model,
+    description: "",
+    hidden: false,
+    supportedReasoningEfforts: [],
+    defaultReasoningEffort: "medium",
+    inputModalities: [],
+    supportsPersonality: false,
+    additionalSpeedTiers: [],
+    serviceTiers: [],
+    defaultServiceTier: null,
+    isDefault: false,
+  };
+}