diff --git a/haystack/human_in_the_loop/dataclasses.py b/haystack/human_in_the_loop/dataclasses.py
index 31947365fca..12865825280 100644
--- a/haystack/human_in_the_loop/dataclasses.py
+++ b/haystack/human_in_the_loop/dataclasses.py
@@ -3,7 +3,9 @@
 # SPDX-License-Identifier: Apache-2.0
 
 from dataclasses import asdict, dataclass
-from typing import Any
+from typing import Any, Literal
+
+DecisionStatus = Literal["approved", "modified", "rejected"]
 
 
 @dataclass
@@ -20,11 +22,36 @@ class ConfirmationUIResult:
     :param new_tool_params:
         Optional set of new parameters for the tool. For example, if the user chooses to modify the tool parameters,
         they can provide a new set of parameters here.
+    :param status:
+        Optional explicit decision status. When omitted, built-in actions are normalized as:
+        `confirm` -> `approved`, `modify` -> `modified`, `reject` -> `rejected`.
     """
 
     action: str  # "confirm", "reject", "modify"
     feedback: str | None = None
     new_tool_params: dict[str, Any] | None = None
+    status: DecisionStatus | None = None
+
+    def resolved_status(self) -> DecisionStatus:
+        """
+        Resolve the explicit decision status for this UI result.
+
+        :raises ValueError:
+            If neither a supported built-in action nor an explicit status is provided.
+        """
+        if self.status is not None:
+            return self.status
+
+        if self.action == "confirm":
+            return "approved"
+        if self.action == "modify":
+            return "modified"
+        if self.action == "reject":
+            return "rejected"
+
+        raise ValueError(
+            "Unsupported confirmation action. Provide one of 'confirm', 'modify', 'reject' or set 'status' explicitly."
+        )
 
 
 @dataclass
@@ -45,6 +72,9 @@ class ToolExecutionDecision:
         modified, this can contain the modification details.
     :param final_tool_params:
         Optional final parameters for the tool if execution is confirmed or modified.
+    :param status:
+        Explicit decision status. When omitted, it is inferred from the legacy `execute`/`feedback` shape to remain
+        backward-compatible with older serialized payloads.
     """
 
     tool_name: str
@@ -52,6 +82,26 @@ class ToolExecutionDecision:
     tool_call_id: str | None = None
     feedback: str | None = None
     final_tool_params: dict[str, Any] | None = None
+    status: DecisionStatus | None = None
+
+    def __post_init__(self) -> None:
+        if self.status is None:
+            self.status = self._infer_status()
+        else:
+            self.execute = self.status != "rejected"
+
+    def _infer_status(self) -> DecisionStatus:
+        if not self.execute:
+            return "rejected"
+        if self.feedback is not None and self.final_tool_params is not None:
+            return "modified"
+        return "approved"
+
+    def resolved_status(self) -> DecisionStatus:
+        """
+        Return the normalized decision status for this execution decision.
+        """
+        return self.status
 
     def to_dict(self) -> dict[str, Any]:
         """
diff --git a/haystack/human_in_the_loop/policies.py b/haystack/human_in_the_loop/policies.py
index 46d2b002d39..dcd68c12c98 100644
--- a/haystack/human_in_the_loop/policies.py
+++ b/haystack/human_in_the_loop/policies.py
@@ -75,5 +75,5 @@ def update_after_confirmation(
         :param tool_params: The parameters that were passed to the tool.
         :param confirmation_result: The result from the confirmation UI.
         """
-        if confirmation_result.action == "confirm":
+        if confirmation_result.resolved_status() == "approved":
             self._asked_tools[tool_name] = tool_params
diff --git a/haystack/human_in_the_loop/strategies.py b/haystack/human_in_the_loop/strategies.py
index 2d0a8da05aa..b09d2795659 100644
--- a/haystack/human_in_the_loop/strategies.py
+++ b/haystack/human_in_the_loop/strategies.py
@@ -107,16 +107,21 @@ def run(
         )
 
         # Process the confirmation result
+        decision_status = confirmation_ui_result.resolved_status()
         final_args = {}
-        if confirmation_ui_result.action == "reject":
+        if decision_status == "rejected":
             explanation_text = self.reject_template.format(tool_name=tool_name)
             if confirmation_ui_result.feedback:
                 explanation_text += " "
                 explanation_text += self.user_feedback_template.format(feedback=confirmation_ui_result.feedback)
             return ToolExecutionDecision(
-                tool_name=tool_name, execute=False, tool_call_id=tool_call_id, feedback=explanation_text
+                tool_name=tool_name,
+                execute=False,
+                tool_call_id=tool_call_id,
+                feedback=explanation_text,
+                status="rejected",
             )
-        if confirmation_ui_result.action == "modify" and confirmation_ui_result.new_tool_params:
+        if decision_status == "modified" and confirmation_ui_result.new_tool_params:
             # Update the tool call params with the new params
             final_args.update(confirmation_ui_result.new_tool_params)
             explanation_text = self.modify_template.format(tool_name=tool_name, final_tool_params=final_args)
@@ -129,10 +134,15 @@ def run(
                 execute=True,
                 feedback=explanation_text,
                 final_tool_params=final_args,
+                status="modified",
             )
         # action == "confirm"
         return ToolExecutionDecision(
-            tool_name=tool_name, execute=True, tool_call_id=tool_call_id, final_tool_params=tool_params
+            tool_name=tool_name,
+            execute=True,
+            tool_call_id=tool_call_id,
+            final_tool_params=tool_params,
+            status="approved",
         )
 
     async def run_async(
@@ -542,7 +552,7 @@ def make_assistant_message(chat_message: ChatMessage, tool_calls: list[ToolCall]
                 # This shouldn't happen, if so something went wrong in _run_confirmation_strategies
                 continue
 
-            if not ted.execute:
+            if ted.resolved_status() == "rejected":
                 # rejected tool call
                 tool_result_text = ted.feedback or REJECTION_FEEDBACK_TEMPLATE.format(tool_name=tc.tool_name)
                 rejection_messages.extend(
@@ -555,7 +565,7 @@ def make_assistant_message(chat_message: ChatMessage, tool_calls: list[ToolCall]
 
             # Covers confirm and modify cases
             final_args = ted.final_tool_params or {}
-            if tc.arguments != final_args:
+            if ted.resolved_status() == "modified" or tc.arguments != final_args:
                 # In the modify case we add a user message explaining the modification otherwise the LLM won't know
                 # why the tool parameters changed and will likely just try and call the tool again with the
                 # original parameters.
diff --git a/releasenotes/notes/add-hitl-decision-status-semantics-phase3.yaml b/releasenotes/notes/add-hitl-decision-status-semantics-phase3.yaml
new file mode 100644
index 00000000000..b990c53c912
--- /dev/null
+++ b/releasenotes/notes/add-hitl-decision-status-semantics-phase3.yaml
@@ -0,0 +1,8 @@
+---
+enhancements:
+  - |
+    Adds an explicit `status` field to `ToolExecutionDecision` with the values
+    `approved`, `modified`, and `rejected`, while preserving the legacy
+    `execute` boolean for backward compatibility. `BlockingConfirmationStrategy`
+    now normalizes built-in confirmation UI actions into these explicit HITL
+    decision statuses.
diff --git a/test/human_in_the_loop/test_dataclasses.py b/test/human_in_the_loop/test_dataclasses.py
index daaabe1c032..679ae5ed21c 100644
--- a/test/human_in_the_loop/test_dataclasses.py
+++ b/test/human_in_the_loop/test_dataclasses.py
@@ -11,6 +11,11 @@ def test_init(self):
         assert original.action == "reject"
         assert original.feedback == "Changed my mind"
         assert original.new_tool_params is None
+        assert original.resolved_status() == "rejected"
+
+    def test_resolved_status_from_explicit_status(self):
+        original = ConfirmationUIResult(action="custom", status="modified")
+        assert original.resolved_status() == "modified"
 
 
 class TestToolExecutionDecision:
@@ -25,6 +30,7 @@ def test_init(self):
         assert decision.final_tool_params == {"param1": "new_value"}
         assert decision.tool_call_id == "test_tool_call_id"
         assert decision.tool_name == "test_tool"
+        assert decision.status == "approved"
 
     def test_to_dict(self):
         original = ToolExecutionDecision(
@@ -40,6 +46,7 @@ def test_to_dict(self):
             "tool_call_id": "test_tool_call_id",
             "feedback": None,
             "final_tool_params": {"param1": "new_value"},
+            "status": "approved",
         }
 
     def test_from_dict(self):
@@ -56,3 +63,16 @@ def test_from_dict(self):
         assert decision.tool_call_id == "another_tool_call_id"
         assert decision.feedback == "Not needed"
         assert decision.final_tool_params == {"paramA": 123}
+        assert decision.status == "rejected"
+
+    def test_from_dict_legacy_payload_without_status(self):
+        data = {
+            "execute": True,
+            "tool_name": "legacy_tool",
+            "tool_call_id": "legacy-tool-call-id",
+            "feedback": "The parameters for tool 'legacy_tool' were updated by the user to:\n{'x': 2}",
+            "final_tool_params": {"x": 2},
+        }
+        decision = ToolExecutionDecision.from_dict(data)
+        assert decision.execute is True
+        assert decision.status == "modified"