diff --git a/onnxruntime/core/framework/kernel_def_builder.cc b/onnxruntime/core/framework/kernel_def_builder.cc
index 0d475eab46..6270e30d3e 100644
--- a/onnxruntime/core/framework/kernel_def_builder.cc
+++ b/onnxruntime/core/framework/kernel_def_builder.cc
@@ -28,10 +28,18 @@ inline bool AreVectorsOverlap(const std::vector<T>& v1, const std::vector<T>& v2
 bool KernelDef::IsConflict(const KernelDef& other) const {
   if (op_name_ != other.OpName() || provider_type_ != other.Provider())
     return false;
-  int start = 0;
-  int end = 0;
-  other.SinceVersion(&start, &end);
-  if (!AreIntervalsOverlap(op_since_version_start_, op_since_version_end_, start, end))
+  int other_since_version_start = 0;
+  int other_since_version_end = 0;
+  other.SinceVersion(&other_since_version_start, &other_since_version_end);
+
+  //When max version is INT_MAX, it means that it should be determined based on the
+  //SinceVersion of schema from a higher version.  Since this sometimes isn't known until
+  //all custom schema are available, make a conservative assumption here that the operator
+  //is valid for only one version.
+  int op_since_version_conservative_end = (op_since_version_end_ == INT_MAX) ? op_since_version_start_ : op_since_version_end_;
+  int other_conservative_since_version_end = (other_since_version_end == INT_MAX) ? other_since_version_start : other_since_version_end;
+
+  if (!AreIntervalsOverlap(op_since_version_start_, op_since_version_conservative_end, other_since_version_start, other_conservative_since_version_end))
     return false;
   //only one case they don't conflict:
   //There is a type_constraint, it exists in both hands, but they don't overlap
diff --git a/onnxruntime/core/providers/dml/DmlExecutionProvider/src/AbiCustomRegistry.cpp b/onnxruntime/core/providers/dml/DmlExecutionProvider/src/AbiCustomRegistry.cpp
index cdbfb759e6..5365d10c9f 100644
--- a/onnxruntime/core/providers/dml/DmlExecutionProvider/src/AbiCustomRegistry.cpp
+++ b/onnxruntime/core/providers/dml/DmlExecutionProvider/src/AbiCustomRegistry.cpp
@@ -499,11 +499,9 @@ HRESULT STDMETHODCALLTYPE AbiCustomRegistry::RegisterOperatorKernel(
         registration->requiresFloatFormatsExceptConstInputs = requiresFloatFormatsForGraph;
         registration->requiredConstantCpuInputs = constantCpuInputCapture;
 
-        // TODO: Propagate errors here once the presence of overlapping built-in DML kernels is addressed
-        if (m_kernelRegistry->RegisterCustomKernel(create_info).IsOK())
-        {
-            (*m_graphNodeFactoryMap)[create_info.kernel_def.get()] = registration;
-        }
+        onnxruntime::KernelDef* kernelDef = create_info.kernel_def.get();
+        THROW_IF_NOT_OK(m_kernelRegistry->RegisterCustomKernel(create_info));
+        (*m_graphNodeFactoryMap)[kernelDef] = registration;
     }
     else
     {
diff --git a/onnxruntime/test/framework/kernel_registry_test.cc b/onnxruntime/test/framework/kernel_registry_test.cc
index c186a0b035..d59de5cf6f 100644
--- a/onnxruntime/test/framework/kernel_registry_test.cc
+++ b/onnxruntime/test/framework/kernel_registry_test.cc
@@ -90,5 +90,15 @@ TEST(KernelRegistryTests, two_versions3) {
   function_table.emplace_back(KernelDefBuilder().MayInplace(0, 0).TypeConstraint("T", DataTypeImpl::GetTensorType<float>()).SetName("Elu").SetDomain("").SinceVersion(6).Provider(kCpuExecutionProvider).Build());
   function_table.emplace_back(KernelDefBuilder().MayInplace(0, 0).TypeConstraint("T", DataTypeImpl::GetTensorType<float>()).SetName("Elu").SetDomain("").SinceVersion(1).Provider(kCpuExecutionProvider).Build());
   Status st;
+  ASSERT_TRUE((st = RegKernels(r, function_table, CreateFakeKernel)).IsOK()) << st.ErrorMessage();
+}
+
+//One op two versions
+TEST(KernelRegistryTests, two_versions4) {
+  KernelRegistry r;
+  std::vector<std::unique_ptr<KernelDef> > function_table;
+  function_table.emplace_back(KernelDefBuilder().MayInplace(0, 0).TypeConstraint("T", DataTypeImpl::GetTensorType<float>()).SetName("Elu").SetDomain("").SinceVersion(5,6).Provider(kCpuExecutionProvider).Build());
+  function_table.emplace_back(KernelDefBuilder().MayInplace(0, 0).TypeConstraint("T", DataTypeImpl::GetTensorType<float>()).SetName("Elu").SetDomain("").SinceVersion(6,7).Provider(kCpuExecutionProvider).Build());
+  Status st;
   ASSERT_FALSE((st = RegKernels(r, function_table, CreateFakeKernel)).IsOK());
 }
\ No newline at end of file