diff --git a/api/gateway_chat.py b/api/gateway_chat.py
index 548bd60e..f2bf9e8c 100644
--- a/api/gateway_chat.py
+++ b/api/gateway_chat.py
@@ -826,6 +826,12 @@ def _run_gateway_chat_streaming(
             s = get_session(session_id)
             if not _stream_writeback_is_current(s, stream_id):
                 return
+            # A late Stop can land after Gateway has yielded a full answer but
+            # before success writeback. Treat it as cancellation so any
+            # credential-exhausted process-wakeup pause stays in place.
+            if cancel_event.is_set():
+                put_gateway_event("cancel", {"message": "Cancelled by user"})
+                return
             now = time.time()
             # Preserve subsecond ordering for gateway-backed turns. Using an
             # integer seconds timestamp gives the user and assistant rows the
diff --git a/tests/test_issue3929_process_wakeup_pause.py b/tests/test_issue3929_process_wakeup_pause.py
index 3c2c8551..32bcc911 100644
--- a/tests/test_issue3929_process_wakeup_pause.py
+++ b/tests/test_issue3929_process_wakeup_pause.py
@@ -375,13 +375,16 @@ def test_success_path_clears_process_wakeup_pause_after_late_cancel_checks():
     assert state_sync_idx < final_cancel_idx < pause_clear_idx < done_payload_idx
 
 
-def test_gateway_success_path_clears_process_wakeup_pause_before_save():
+def test_gateway_success_path_checks_cancel_before_clearing_process_wakeup_pause():
     src = Path(__file__).parent.parent.joinpath("api", "gateway_chat.py").read_text(encoding="utf-8")
+    current_idx = src.index("if not _stream_writeback_is_current(s, stream_id):")
+    final_cancel_idx = src.index("if cancel_event.is_set():", current_idx)
     pending_clear_idx = src.index("s.pending_user_source = None")
     pause_clear_idx = src.index('clear_process_wakeup_pause(s, reason="run_completed")')
     save_idx = src.index("s.save()", pause_clear_idx)
 
-    assert pending_clear_idx < pause_clear_idx < save_idx
+    assert current_idx < final_cancel_idx < pause_clear_idx < save_idx
+    assert pending_clear_idx < pause_clear_idx
 
 
 def test_process_wakeup_pause_does_not_suppress_explicit_non_wakeup_turn(tmp_path, monkeypatch):