diff --git a/src/praisonai-agents/praisonaiagents/agent/agent.py b/src/praisonai-agents/praisonaiagents/agent/agent.py
index 0a68194e0..bdb096aaf 100644
--- a/src/praisonai-agents/praisonaiagents/agent/agent.py
+++ b/src/praisonai-agents/praisonaiagents/agent/agent.py
@@ -3751,6 +3751,11 @@ def last_stop_reason(self) -> str:
         # the agent (see ``_extract_llm_response_content``); prefer it so a
         # provider block/refusal isn't masked by a backend default of "completed".
         own = getattr(self, '_last_stop_reason', None)
+        # Prefer a specific agent-owned provider block/refusal/truncation before
+        # any backend fallback so a stale backend reason (e.g. ``max_steps``)
+        # cannot mask the OpenAI-native classification recorded for this turn.
+        if own and own != "completed":
+            return own
         # Read from the already-instantiated backends only. ``__openai_client``
         # is the raw (name-mangled) attribute, never the lazy ``_openai_client``
         # property, so this never triggers OpenAI client creation for
diff --git a/src/praisonai-agents/praisonaiagents/agent/chat_mixin.py b/src/praisonai-agents/praisonaiagents/agent/chat_mixin.py
index 27faa20e1..c83fabd28 100644
--- a/src/praisonai-agents/praisonaiagents/agent/chat_mixin.py
+++ b/src/praisonai-agents/praisonaiagents/agent/chat_mixin.py
@@ -5766,6 +5766,13 @@ def _model_fn(req):
 
     async def _achat_completion_with_retry_core(self, messages, temperature=None, tools=None, stream=None, reasoning_steps=False, task_name=None, task_description=None, task_id=None, response_format=None, stream_callback=None, emit_events=True):
         """Async retry/backoff core for chat completion (middleware-agnostic)."""
+        # Reset the agent-level finish-reason classification at the start of each
+        # async OpenAI-native turn, mirroring the sync ``_chat_completion`` reset,
+        # so a provider block/refusal recorded on a previous run (see
+        # ``_extract_llm_response_content``) never leaks into this one. The LiteLLM
+        # path resets its own backend flag independently.
+        self._last_stop_reason = "completed"
+
         retry_config = getattr(self, '_retry_config', None)
         if not retry_config:
             return await self._execute_unified_achat_completion(
diff --git a/src/praisonai-agents/praisonaiagents/agent/execution_mixin.py b/src/praisonai-agents/praisonaiagents/agent/execution_mixin.py
index 50c877088..cde0fd81a 100644
--- a/src/praisonai-agents/praisonaiagents/agent/execution_mixin.py
+++ b/src/praisonai-agents/praisonaiagents/agent/execution_mixin.py
@@ -312,9 +312,14 @@ def _outcome_for_result(self, result):
         """
         from .run_outcome import RunOutcome, PROVIDER_BLOCK_REASONS
         output = str(result) if result is not None else None
+        # Read the recorded classification without letting a lookup problem mask a
+        # provider block as a successful ``completed``. Only an expected
+        # missing-state lookup (``AttributeError``) is treated as "no reason
+        # recorded"; anything else propagates rather than silently succeeding.
+        reason = None
         try:
             reason = getattr(self, "last_stop_reason", None)
-        except Exception:
+        except AttributeError:
             reason = None
         if reason in PROVIDER_BLOCK_REASONS:
             return RunOutcome(reason=reason, output=output)
diff --git a/src/praisonai-agents/tests/test_run_outcome.py b/src/praisonai-agents/tests/test_run_outcome.py
index 7074bfc0c..6ba3373fe 100644
--- a/src/praisonai-agents/tests/test_run_outcome.py
+++ b/src/praisonai-agents/tests/test_run_outcome.py
@@ -205,3 +205,33 @@ def test_astart_outcome_surfaces_refusal():
     )
     assert o.reason == "refused"
     assert o.succeeded is False
+
+
+def test_agent_last_stop_reason_prefers_own_block_over_stale_backend():
+    # Regression: an agent-owned provider block (OpenAI-native classification)
+    # must win over a stale backend reason like "max_steps"; otherwise the
+    # actionable "refused" is masked and _outcome_for_result reports success.
+    from praisonaiagents.agent.agent import Agent
+
+    agent = Agent(instructions="test", llm="gpt-4o-mini")
+    agent._last_stop_reason = "refused"
+
+    class _StaleBackend:
+        _last_stop_reason = "max_steps"
+
+    agent.llm_instance = _StaleBackend()
+    assert agent.last_stop_reason == "refused"
+
+
+def test_agent_last_stop_reason_backend_used_when_own_completed():
+    # When the agent recorded nothing specific, the backend reason still wins.
+    from praisonaiagents.agent.agent import Agent
+
+    agent = Agent(instructions="test", llm="gpt-4o-mini")
+    agent._last_stop_reason = "completed"
+
+    class _Backend:
+        _last_stop_reason = "max_steps"
+
+    agent.llm_instance = _Backend()
+    assert agent.last_stop_reason == "max_steps"