diff --git a/sagemaker-serve/src/sagemaker/serve/predictor_async.py b/sagemaker-serve/src/sagemaker/serve/predictor_async.py
index 986d51c3d7..acaa87b824 100644
--- a/sagemaker-serve/src/sagemaker/serve/predictor_async.py
+++ b/sagemaker-serve/src/sagemaker/serve/predictor_async.py
@@ -335,7 +335,13 @@ def check_failure_file():
         output_thread.start()
         failure_thread.start()
 
-        while not output_file_found.is_set() and not failure_file_found.is_set():
+        # Stop once either file is found, or once both waiters have given up (WaiterError);
+        # otherwise a timeout on both would leave this loop spinning forever.
+        while (
+            not output_file_found.is_set()
+            and not failure_file_found.is_set()
+            and (output_thread.is_alive() or failure_thread.is_alive())
+        ):
             time.sleep(1)
 
         if output_file_found.is_set():
@@ -343,17 +349,15 @@ def check_failure_file():
             result = self.predictor._handle_response(response=s3_object)
             return result
 
-        failure_object = self.s3_client.get_object(Bucket=failure_bucket, Key=failure_key)
-        failure_response = self.predictor._handle_response(response=failure_object)
+        if failure_file_found.is_set():
+            failure_object = self.s3_client.get_object(Bucket=failure_bucket, Key=failure_key)
+            failure_response = self.predictor._handle_response(response=failure_object)
+            raise AsyncInferenceModelError(message=failure_response)
 
-        raise (
-            AsyncInferenceModelError(message=failure_response)
-            if failure_file_found.is_set()
-            else PollingTimeoutError(
-                message="Inference could still be running",
-                output_path=output_path,
-                seconds=waiter_config.delay * waiter_config.max_attempts,
-            )
+        raise PollingTimeoutError(
+            message="Inference could still be running",
+            output_path=output_path,
+            seconds=waiter_config.delay * waiter_config.max_attempts,
         )
 
     def update_endpoint(
diff --git a/sagemaker-serve/tests/unit/test_predictor_async.py b/sagemaker-serve/tests/unit/test_predictor_async.py
index aaf77a9c4a..a8ab24c8ab 100644
--- a/sagemaker-serve/tests/unit/test_predictor_async.py
+++ b/sagemaker-serve/tests/unit/test_predictor_async.py
@@ -1,9 +1,15 @@
 import io
+import threading
+import time
 import unittest
 from unittest.mock import Mock, patch
 
 from sagemaker.core.deserializers import JSONDeserializer
 from sagemaker.core.serializers import JSONSerializer
+from botocore.exceptions import WaiterError
+
+from sagemaker.core.exceptions import AsyncInferenceModelError, PollingTimeoutError
+from sagemaker.serve.async_inference import WaiterConfig
 from sagemaker.serve.predictor_async import AsyncPredictor
 
 
@@ -231,3 +237,59 @@ def test_reflects_serializers_set_on_wrapped_predictor(self):
 
 if __name__ == "__main__":
     unittest.main()
+
+
+class TestCheckOutputAndFailurePaths(unittest.TestCase):
+    """_check_output_and_failure_paths must return or raise once both waiters are done."""
+
+    def setUp(self):
+        self.mock_predictor = Mock()
+        self.mock_predictor.endpoint_name = "test-endpoint"
+        self.async_predictor = AsyncPredictor(self.mock_predictor)
+        self.async_predictor.s3_client = Mock()
+        self.waiter_config = WaiterConfig(max_attempts=1, delay=1)
+
+    def _run_with_timeout(self, timeout=10):
+        outcome = {}
+
+        def run():
+            try:
+                outcome["result"] = self.async_predictor._check_output_and_failure_paths(
+                    "s3://bucket/output", "s3://bucket/failure", self.waiter_config
+                )
+            except Exception as e:  # pylint: disable=broad-except
+                outcome["error"] = e
+
+        thread = threading.Thread(target=run, daemon=True)
+        thread.start()
+        thread.join(timeout=timeout)
+        self.assertFalse(thread.is_alive(), "_check_output_and_failure_paths did not return")
+        return outcome
+
+    def test_raises_polling_timeout_when_both_waiters_time_out(self):
+        waiter = Mock()
+        waiter.wait.side_effect = WaiterError("ObjectExists", "Max attempts exceeded", {})
+        self.async_predictor.s3_client.get_waiter.return_value = waiter
+
+        outcome = self._run_with_timeout()
+
+        self.assertIsInstance(outcome.get("error"), PollingTimeoutError)
+        self.async_predictor.s3_client.get_object.assert_not_called()
+
+    def test_raises_model_error_when_failure_file_appears_after_output_waiter_times_out(self):
+        def wait(Bucket, Key, WaiterConfig):
+            if Key == "output":
+                raise WaiterError("ObjectExists", "Max attempts exceeded", {})
+            time.sleep(0.2)
+
+        waiter = Mock()
+        waiter.wait.side_effect = wait
+        self.async_predictor.s3_client.get_waiter.return_value = waiter
+        self.mock_predictor._handle_response.return_value = "model failure"
+
+        outcome = self._run_with_timeout()
+
+        self.assertIsInstance(outcome.get("error"), AsyncInferenceModelError)
+        self.async_predictor.s3_client.get_object.assert_called_once_with(
+            Bucket="bucket", Key="failure"
+        )