diff --git a/onnxruntime/core/providers/cuda/math/gemm.cc b/onnxruntime/core/providers/cuda/math/gemm.cc
index a595153e32..57e1b37c0c 100644
--- a/onnxruntime/core/providers/cuda/math/gemm.cc
+++ b/onnxruntime/core/providers/cuda/math/gemm.cc
@@ -118,7 +118,7 @@ Status Gemm<T>::ComputeInternal(OpKernelContext* ctx) const {
           out_data, N, device_prop));
     } else {
       // B is (M, N), no broadcast needed.
-      CUDA_RETURN_IF_ERROR(cudaMemcpyAsync(out_data, b_data, M * N * sizeof(T), cudaMemcpyDeviceToDevice, Stream(ctx)));
+      CUDA_RETURN_IF_ERROR(cudaMemcpyAsync(out_data, b_data, static_cast<size_t>(M) * N * sizeof(T), cudaMemcpyDeviceToDevice, Stream(ctx)));
     }
   }
 
diff --git a/onnxruntime/core/providers/cuda/tensor/pad.cc b/onnxruntime/core/providers/cuda/tensor/pad.cc
index 3955f454d3..4584e5fd82 100644
--- a/onnxruntime/core/providers/cuda/tensor/pad.cc
+++ b/onnxruntime/core/providers/cuda/tensor/pad.cc
@@ -135,14 +135,14 @@ Status Pad<T>::ComputeInternal(OpKernelContext* ctx) const {
   TArray<int64_t> input_strides(input_pitches);
 
   auto output_dims(input_shape.AsShapeVector());
-  ORT_ENFORCE(static_cast<size_t>(dimension_count * 2) == p_pads->size(), "'pads' attribute has wrong number of values");
+  ORT_ENFORCE(static_cast<size_t>(dimension_count) * 2 == p_pads->size(), "'pads' attribute has wrong number of values");
 
   // Calculate output dimensions, and handle any negative padding
   TArray<int64_t> lower_pads(dimension_count);
   TArray<int64_t> upper_pads(dimension_count);
   for (auto i = 0; i < dimension_count; i++) {
     lower_pads[i] = (*p_pads)[i] + (*p_slices)[i];
-    upper_pads[i] = (*p_pads)[i + dimension_count] + (*p_slices)[i + dimension_count];
+    upper_pads[i] = (*p_pads)[static_cast<int64_t>(i) + dimension_count] + (*p_slices)[static_cast<int64_t>(i) + dimension_count];
     output_dims[i] += lower_pads[i] + upper_pads[i];
   }
 
diff --git a/onnxruntime/core/session/provider_bridge_ort.cc b/onnxruntime/core/session/provider_bridge_ort.cc
index 71ad046646..2d35f1da94 100644
--- a/onnxruntime/core/session/provider_bridge_ort.cc
+++ b/onnxruntime/core/session/provider_bridge_ort.cc
@@ -1043,8 +1043,8 @@ struct ProviderSharedLibrary {
     if (handle_)
       return;
 
-    auto full_path = Env::Default().GetRuntimePath() + 
-      PathString(LIBRARY_PREFIX ORT_TSTR("onnxruntime_providers_shared") LIBRARY_EXTENSION);
+    auto full_path = Env::Default().GetRuntimePath() +
+                     PathString(LIBRARY_PREFIX ORT_TSTR("onnxruntime_providers_shared") LIBRARY_EXTENSION);
     ORT_THROW_IF_ERROR(Env::Default().LoadDynamicLibrary(full_path, true /*shared_globals on unix*/, &handle_));
 
     void (*PProvider_SetHost)(void*);
@@ -1681,7 +1681,16 @@ ORT_API_STATUS_IMPL(OrtApis::SessionOptionsAppendExecutionProvider_CUDA_V2, _In_
 ORT_API_STATUS_IMPL(OrtApis::CreateCUDAProviderOptions, _Outptr_ OrtCUDAProviderOptionsV2** out) {
   API_IMPL_BEGIN
 #ifdef USE_CUDA
+
+// Need to use 'new' here, so disable C26409
+#ifdef _WIN32
+#pragma warning(push)
+#pragma warning(disable : 26409)
+#endif
   *out = new OrtCUDAProviderOptionsV2();
+#ifdef _WIN32
+#pragma warning(pop)
+#endif
   (*out)->device_id = 0;
   (*out)->cudnn_conv_algo_search = OrtCudnnConvAlgoSearch::OrtCudnnConvAlgoSearchExhaustive;
   (*out)->gpu_mem_limit = std::numeric_limits<size_t>::max();
@@ -1766,7 +1775,19 @@ ORT_API_STATUS_IMPL(OrtApis::GetCUDAProviderOptionsAsString, _In_ const OrtCUDAP
 
 ORT_API(void, OrtApis::ReleaseCUDAProviderOptions, _Frees_ptr_opt_ OrtCUDAProviderOptionsV2* ptr) {
 #ifdef USE_CUDA
+
+// Need to use 'delete' here, so disable C26409
+#ifdef _WIN32
+#pragma warning(push)
+#pragma warning(disable : 26409)
+#endif
+
   delete ptr;
+
+#ifdef _WIN32
+#pragma warning(pop)
+#endif
+
 #else
   ORT_UNUSED_PARAMETER(ptr);
 #endif