diff --git a/python/packages/core/agent_framework/_evaluation.py b/python/packages/core/agent_framework/_evaluation.py
index 327258b846b..e7ffb79088d 100644
--- a/python/packages/core/agent_framework/_evaluation.py
+++ b/python/packages/core/agent_framework/_evaluation.py
@@ -1221,7 +1221,7 @@ def tool_call_args_match(item: EvalItem) -> CheckResult:
         for _, actual_args in matching:
             if actual_args is None:
                 continue
-            if all(actual_args.get(k) == v for k, v in exp.arguments.items()):
+            if all(k in actual_args and actual_args[k] == v for k, v in exp.arguments.items()):
                 found = True
                 break
 
diff --git a/python/packages/core/tests/core/test_local_eval.py b/python/packages/core/tests/core/test_local_eval.py
index f3713506c22..e00e0d6a951 100644
--- a/python/packages/core/tests/core/test_local_eval.py
+++ b/python/packages/core/tests/core/test_local_eval.py
@@ -5,6 +5,7 @@
 from __future__ import annotations
 
 import inspect
+import json
 
 import pytest
 
@@ -612,6 +613,37 @@ def test_args_mismatch(self):
         assert result.passed is False
         assert "args mismatch" in result.reason
 
+    @pytest.mark.parametrize("as_json", [False, True], ids=["dict", "json"])
+    @pytest.mark.parametrize(
+        "arguments, expected_passed",
+        [
+            pytest.param({"location": "NYC"}, False, id="missing"),
+            pytest.param({"location": "NYC", "units": None}, True, id="explicit-null"),
+            pytest.param({"location": "NYC", "units": "fahrenheit"}, False, id="non-null"),
+            pytest.param({"location": "NYC", "units": None, "days": 1}, True, id="extra-argument"),
+        ],
+    )
+    def test_null_argument_requires_key(self, arguments, expected_passed, as_json):
+        item = EvalItem(
+            conversation=[
+                Message(
+                    "assistant",
+                    [
+                        Content.from_function_call(
+                            "call_weather",
+                            "get_weather",
+                            arguments=json.dumps(arguments) if as_json else arguments,
+                        )
+                    ],
+                )
+            ],
+            expected_tool_calls=[ExpectedToolCall("get_weather", {"location": "NYC", "units": None})],
+        )
+
+        result = tool_call_args_match(item)
+
+        assert result.passed is expected_passed
+
     def test_tool_not_called(self):
         item = _make_tool_call_item(
             calls=[("get_news", None)],