diff --git a/src/utils/advisor.py b/src/utils/advisor.py
index 0f7df0b2..116e00f4 100644
--- a/src/utils/advisor.py
+++ b/src/utils/advisor.py
@@ -393,7 +393,7 @@ def decide_advisor_mode(
     *,
     force_client_mode: bool = False,
     advisor_provider: str | None = None,
-    advisor_enabled: bool = True,
+    advisor_enabled: bool = False,
 ) -> str:
     """Pick activation mode for the upcoming turn.
 
@@ -422,9 +422,9 @@ def decide_advisor_mode(
 
     ``advisor_enabled`` is the master switch (settings ``advisor_enabled``,
     default False in production): when False the advisor is INACTIVE regardless
-    of model/provider. The parameter defaults True so direct callers (the
-    activation truth-table tests) keep their behavior; production call sites pass
-    ``get_settings().advisor_enabled``.
+    of model/provider. The parameter deliberately defaults False so callers must
+    explicitly opt in, just like users do through ``advisor_enabled`` in config
+    or the ``/advisor`` command.
     """
     if not advisor_enabled:
         return ADVISOR_MODE_INACTIVE
diff --git a/tests/test_advisor_client_side.py b/tests/test_advisor_client_side.py
index f50a101c..6160ffcd 100644
--- a/tests/test_advisor_client_side.py
+++ b/tests/test_advisor_client_side.py
@@ -68,6 +68,16 @@ def test_inactive_when_env_disabled(self) -> None:
             )
             self.assertEqual(mode, ADVISOR_MODE_INACTIVE)
 
+    def test_inactive_by_default_even_when_fully_configured(self) -> None:
+        """The decision helper itself must be opt-in, not only its callers."""
+        mode = decide_advisor_mode(
+            self._first_party_provider(),
+            "claude-opus-4-6",
+            "claude-opus-4-6",
+            advisor_provider="anthropic",
+        )
+        self.assertEqual(mode, ADVISOR_MODE_INACTIVE)
+
     def test_server_side_for_1p_with_valid_models(self) -> None:
         # Server-side now requires advisor_provider == "anthropic" too.
         with patch(
@@ -79,6 +89,7 @@ def test_server_side_for_1p_with_valid_models(self) -> None:
                 "claude-opus-4-6",
                 "claude-opus-4-6",
                 advisor_provider="anthropic",
+                advisor_enabled=True,
             )
             self.assertEqual(mode, ADVISOR_MODE_SERVER_SIDE)
 
@@ -93,6 +104,7 @@ def test_client_side_when_1p_with_force_client(self) -> None:
                 "claude-opus-4-6",
                 force_client_mode=True,
                 advisor_provider="anthropic",
+                advisor_enabled=True,
             )
             self.assertEqual(mode, ADVISOR_MODE_CLIENT_SIDE)
 
@@ -108,6 +120,7 @@ def test_client_side_for_1p_with_unsupported_base_model(self) -> None:
                 "claude-opus-4-5",
                 "claude-opus-4-6",
                 advisor_provider="anthropic",
+                advisor_enabled=True,
             )
             self.assertEqual(mode, ADVISOR_MODE_CLIENT_SIDE)
 
@@ -123,6 +136,7 @@ def test_client_side_for_3p_provider(self) -> None:
                 "gpt-5.4",
                 "claude-opus-4-6",
                 advisor_provider="anthropic",
+                advisor_enabled=True,
             )
             self.assertEqual(mode, ADVISOR_MODE_CLIENT_SIDE)
 
@@ -139,6 +153,7 @@ def test_client_side_cross_provider(self) -> None:
                 "claude-opus-4-6",
                 "gemini-2.5-pro",
                 advisor_provider="gemini",
+                advisor_enabled=True,
             )
             self.assertEqual(mode, ADVISOR_MODE_CLIENT_SIDE)