diff --git a/docs/gateway/configuration-reference.md b/docs/gateway/configuration-reference.md
index c3f8cb72fab..e3a46d67795 100644
--- a/docs/gateway/configuration-reference.md
+++ b/docs/gateway/configuration-reference.md
@@ -1342,6 +1342,7 @@ Batches rapid text-only messages from the same sender into a single agent turn.
 
 - `auto` controls auto-TTS. `/tts off|always|inbound|tagged` overrides per session.
 - `summaryModel` overrides `agents.defaults.model.primary` for auto-summary.
+- `modelOverrides` is enabled by default; `modelOverrides.allowProvider` defaults to `false` (opt-in).
 - API keys fall back to `ELEVENLABS_API_KEY`/`XI_API_KEY` and `OPENAI_API_KEY`.
 
 ---
diff --git a/docs/tts.md b/docs/tts.md
index c52a1546cbd..24ca527e13a 100644
--- a/docs/tts.md
+++ b/docs/tts.md
@@ -210,6 +210,7 @@ Then run:
 - `summaryModel`: optional cheap model for auto-summary; defaults to `agents.defaults.model.primary`.
   - Accepts `provider/model` or a configured model alias.
 - `modelOverrides`: allow the model to emit TTS directives (on by default).
+  - `allowProvider` defaults to `false` (provider switching is opt-in).
 - `maxTextLength`: hard cap for TTS input (chars). `/tts audio` fails if exceeded.
 - `timeoutMs`: request timeout (ms).
 - `prefsPath`: override the local prefs JSON path (provider/limit/summary).
@@ -242,18 +243,20 @@ for a single reply, plus an optional `[[tts:text]]...[[/tts:text]]` block to
 provide expressive tags (laughter, singing cues, etc) that should only appear in
 the audio.
 
+`provider=...` directives are ignored unless `modelOverrides.allowProvider: true`.
+
 Example reply payload:
 
 ```
 Here you go.
 
-[[tts:provider=elevenlabs voiceId=pMsXgVXv3BLzUgSXRplE model=eleven_v3 speed=1.1]]
+[[tts:voiceId=pMsXgVXv3BLzUgSXRplE model=eleven_v3 speed=1.1]]
 [[tts:text]](laughs) Read the song once more.[[/tts:text]]
 ```
 
 Available directive keys (when enabled):
 
-- `provider` (`openai` | `elevenlabs` | `edge`)
+- `provider` (`openai` | `elevenlabs` | `edge`, requires `allowProvider: true`)
 - `voice` (OpenAI voice) or `voiceId` (ElevenLabs)
 - `model` (OpenAI TTS model or ElevenLabs model id)
 - `stability`, `similarityBoost`, `style`, `speed`, `useSpeakerBoost`
@@ -275,7 +278,7 @@ Disable all model overrides:
 }
 ```
 
-Optional allowlist (disable specific overrides while keeping tags enabled):
+Optional allowlist (enable provider switching while keeping other knobs configurable):
 
 ```json5
 {
@@ -283,7 +286,7 @@ Optional allowlist (disable specific overrides while keeping tags enabled):
     tts: {
       modelOverrides: {
         enabled: true,
-        allowProvider: false,
+        allowProvider: true,
         allowSeed: false,
       },
     },
diff --git a/src/config/types.tts.ts b/src/config/types.tts.ts
index 4eb4989b98c..82875d55e4a 100644
--- a/src/config/types.tts.ts
+++ b/src/config/types.tts.ts
@@ -9,7 +9,7 @@ export type TtsModelOverrideConfig = {
   enabled?: boolean;
   /** Allow model-provided TTS text blocks. */
   allowText?: boolean;
-  /** Allow model-provided provider override. */
+  /** Allow model-provided provider override (default: false). */
   allowProvider?: boolean;
   /** Allow model-provided voice/voiceId override. */
   allowVoice?: boolean;
diff --git a/src/tts/tts.test.ts b/src/tts/tts.test.ts
index 1c5ecfb558d..09dc90e642c 100644
--- a/src/tts/tts.test.ts
+++ b/src/tts/tts.test.ts
@@ -215,7 +215,7 @@ describe("tts", () => {
 
   describe("parseTtsDirectives", () => {
     it("extracts overrides and strips directives when enabled", () => {
-      const policy = resolveModelOverridePolicy({ enabled: true });
+      const policy = resolveModelOverridePolicy({ enabled: true, allowProvider: true });
       const input =
         "Hello [[tts:provider=elevenlabs voiceId=pMsXgVXv3BLzUgSXRplE stability=0.4 speed=1.1]] world\n\n" +
         "[[tts:text]](laughs) Read the song once more.[[/tts:text]]";
@@ -230,13 +230,22 @@ describe("tts", () => {
     });
 
     it("accepts edge as provider override", () => {
-      const policy = resolveModelOverridePolicy({ enabled: true });
+      const policy = resolveModelOverridePolicy({ enabled: true, allowProvider: true });
       const input = "Hello [[tts:provider=edge]] world";
       const result = parseTtsDirectives(input, policy);
 
       expect(result.overrides.provider).toBe("edge");
     });
 
+    it("rejects provider override by default while keeping voice overrides enabled", () => {
+      const policy = resolveModelOverridePolicy({ enabled: true });
+      const input = "Hello [[tts:provider=edge voice=alloy]] world";
+      const result = parseTtsDirectives(input, policy);
+
+      expect(result.overrides.provider).toBeUndefined();
+      expect(result.overrides.openai?.voice).toBe("alloy");
+    });
+
     it("keeps text intact when overrides are disabled", () => {
       const policy = resolveModelOverridePolicy({ enabled: false });
       const input = "Hello [[tts:voice=alloy]] world";
diff --git a/src/tts/tts.ts b/src/tts/tts.ts
index fb27eddd2d6..3130cf396b8 100644
--- a/src/tts/tts.ts
+++ b/src/tts/tts.ts
@@ -238,11 +238,12 @@ function resolveModelOverridePolicy(
       allowSeed: false,
     };
   }
-  const allow = (value?: boolean) => value ?? true;
+  const allow = (value: boolean | undefined, defaultValue = true) => value ?? defaultValue;
   return {
     enabled: true,
     allowText: allow(overrides?.allowText),
-    allowProvider: allow(overrides?.allowProvider),
+    // Provider switching is higher-impact than voice/style tweaks; keep opt-in.
+    allowProvider: allow(overrides?.allowProvider, false),
     allowVoice: allow(overrides?.allowVoice),
     allowModelId: allow(overrides?.allowModelId),
     allowVoiceSettings: allow(overrides?.allowVoiceSettings),