diff --git a/cmake/external/onnxruntime_external_deps.cmake b/cmake/external/onnxruntime_external_deps.cmake
index 3095968795d1a..c9a5270492222 100644
--- a/cmake/external/onnxruntime_external_deps.cmake
+++ b/cmake/external/onnxruntime_external_deps.cmake
@@ -598,10 +598,6 @@ if(NOT (onnx_FOUND OR ONNX_FOUND)) # building ONNX from source
   endif()
 endif()
 
-if (onnxruntime_RUN_ONNX_TESTS)
-  add_definitions(-DORT_RUN_EXTERNAL_ONNX_TESTS)
-endif()
-
 if(onnxruntime_ENABLE_DLPACK)
   message(STATUS "dlpack is enabled.")
 
diff --git a/onnxruntime/test/framework/inference_session_test.cc b/onnxruntime/test/framework/inference_session_test.cc
index bc01135fbbf1e..6131eff92ac78 100644
--- a/onnxruntime/test/framework/inference_session_test.cc
+++ b/onnxruntime/test/framework/inference_session_test.cc
@@ -588,93 +588,6 @@ TEST(InferenceSessionTests, RequestLoadCancellation) {
   }
 }
 
-#ifdef ORT_RUN_EXTERNAL_ONNX_TESTS
-static bool Compare(const InputDefList& f_arg, const InputDefList& s_arg) {
-  if (f_arg.size() != s_arg.size()) {
-    std::cout << "Sizes differ: f_arg size: " << f_arg.size() << " s_arg size: " << s_arg.size() << std::endl;
-    return false;
-  }
-
-  for (size_t i = 0; i < f_arg.size(); ++i) {
-    const onnxruntime::NodeArg* x = f_arg[i];
-    const onnxruntime::NodeArg* y = s_arg[i];
-    if ((x->Shape() == nullptr) ^ (y->Shape() == nullptr)) {
-      return false;
-    }
-    if (!x->Shape()) {
-      continue;
-    }
-    auto x_shape = utils::GetTensorShapeFromTensorShapeProto(*x->Shape());
-    auto y_shape = utils::GetTensorShapeFromTensorShapeProto(*y->Shape());
-    if (x->Name() == y->Name() && x_shape == y_shape && *x->Type() == *y->Type()) {
-      continue;
-    }
-    return false;
-  }
-
-  return true;
-}
-
-TEST(InferenceSessionTests, ModelMetadata) {
-  SessionOptions so;
-
-  so.session_logid = "InferenceSessionTests.ModelMetadata";
-  InferenceSession session_object{so, GetEnvironment()};
-  auto model_uri = ORT_TSTR("../models/opset8/test_squeezenet/model.onnx");
-  ASSERT_STATUS_OK(session_object.Load(model_uri));
-
-  std::shared_ptr<onnxruntime::Model> p_model;
-  ASSERT_STATUS_OK(onnxruntime::Model::Load(model_uri, p_model, nullptr, DefaultLoggingManager().DefaultLogger()));
-  const onnxruntime::Graph& graph = p_model->MainGraph();
-
-  // 1. first test the model meta
-  {
-    auto retval = session_object.GetModelMetadata();
-    ASSERT_TRUE(retval.first.IsOK());
-    const ModelMetadata* m = retval.second;
-    ASSERT_TRUE(m->custom_metadata_map == p_model->MetaData() &&
-                m->description == p_model->DocString() &&
-                m->domain == p_model->Domain() &&
-                m->graph_name == graph.Name() &&
-                m->producer_name == p_model->ProducerName() &&
-                m->version == p_model->ModelVersion());
-  }
-
-  {
-    // 2. test inputs
-    auto& inputs = graph.GetInputs();
-    auto weights = graph.GetAllInitializedTensors();
-
-    // skip the weights
-    InputDefList inputs_no_weights;
-    for (auto& elem : inputs) {
-      if (weights.find(elem->Name()) != weights.end()) {
-        continue;
-      } else {
-        inputs_no_weights.push_back(elem);
-      }
-    }
-
-    auto retval = session_object.GetModelInputs();
-    std::cout << "weights size: " << weights.size()
-              << " inputs.size(): " << inputs.size()
-              << " from session: " << retval.second->size() << std::endl;
-    ASSERT_TRUE(retval.first.IsOK());
-    ASSERT_TRUE(Compare(inputs_no_weights, *retval.second));
-  }
-
-  // 3. test outputs
-  {
-    auto retval = session_object.GetModelOutputs();
-    ASSERT_TRUE(retval.first.IsOK());
-
-    auto& outputs = graph.GetOutputs();
-    retval = session_object.GetModelOutputs();
-    ASSERT_TRUE(retval.first.IsOK());
-    ASSERT_TRUE(Compare(outputs, *retval.second));
-  }
-}
-#endif
 TEST(InferenceSessionTests, CheckRunLogger) {
   if constexpr (!SessionOptions::DEFAULT_USE_PER_SESSION_THREADS) {
     GTEST_SKIP() << "Skipping the test";
diff --git a/onnxruntime/test/ir/onnx_model_test.cc b/onnxruntime/test/ir/onnx_model_test.cc
index 9327d86966981..55fc4f42bec64 100644
--- a/onnxruntime/test/ir/onnx_model_test.cc
+++ b/onnxruntime/test/ir/onnx_model_test.cc
@@ -26,44 +26,6 @@ class ONNXModelsTest : public ::testing::Test {
 
   std::unique_ptr<logging::Logger> logger_;
 };
-#ifdef ORT_RUN_EXTERNAL_ONNX_TESTS
-// Tests that Resolve() properly clears the state of topological sorted nodes,
-// inputs, outputs and valueInfo.
-// Assumes the graph passed in has been previously resolved.
-static void TestResolve(onnxruntime::Graph& graph) {
-  GraphViewer graph_viewer(graph);
-  auto& nodes_before = graph_viewer.GetNodesInTopologicalOrder();
-  auto& inputs_before = graph.GetInputs();
-  auto& outputs_before = graph.GetOutputs();
-  auto& value_info_before = graph.GetValueInfo();
-
-  // Touch the graph to force Resolve() to recompute.
-  graph.SetGraphResolveNeeded();
-  graph.SetGraphProtoSyncNeeded();
-  ASSERT_STATUS_OK(graph.Resolve());
-
-  GraphViewer graph_viewer_2(graph);
-  auto& nodes_after = graph_viewer_2.GetNodesInTopologicalOrder();
-  auto& inputs_after = graph.GetInputs();
-  auto& outputs_after = graph.GetOutputs();
-  auto& value_info_after = graph.GetValueInfo();
-
-  // Multiple calls to Resolve() should not alter the sorted nodes,
-  // inputs, outputs and valueInfo. The internal state should be
-  // cleared.
-  EXPECT_EQ(nodes_before, nodes_after);
-  EXPECT_EQ(inputs_before, inputs_after);
-  EXPECT_EQ(outputs_before, outputs_after);
-  EXPECT_EQ(value_info_before, value_info_after);
-}
-
-TEST_F(ONNXModelsTest, squeeze_net) {
-  // NOTE: this requires the current directory to be where onnxruntime_ir_UT.exe is located
-  std::shared_ptr<Model> model;
-  ASSERT_STATUS_OK(Model::Load(ORT_TSTR("../models/opset8/test_squeezenet/model.onnx"), model, nullptr, *logger_));
-  TestResolve(model->MainGraph());
-}
-#endif
 
 TEST_F(ONNXModelsTest, non_existing_model) {
   // NOTE: this requires the current directory to be where onnxruntime_ir_UT.exe is located
@@ -96,76 +58,6 @@ class ONNXModelsTest1 : public ::testing::TestWithParam<const ORTCHAR_T*> {
     return oss.str();
   }
 };
-#ifdef ORT_RUN_EXTERNAL_ONNX_TESTS
-TEST_F(ONNXModelsTest, bvlc_alexnet_1) {
-  using ::google::protobuf::io::CodedInputStream;
-  using ::google::protobuf::io::FileInputStream;
-  using ::google::protobuf::io::ZeroCopyInputStream;
-  int fd;
-  ASSERT_STATUS_OK(Env::Default().FileOpenRd(ORT_TSTR("../models/opset8/test_bvlc_alexnet/model.onnx"), fd));
-  ASSERT_TRUE(fd > 0);
-  std::unique_ptr<ZeroCopyInputStream> raw_input(new FileInputStream(fd));
-  std::unique_ptr<CodedInputStream> coded_input(new CodedInputStream(raw_input.get()));
-  // Allows protobuf library versions < 3.2.0 to parse messages greater than 64MB.
-  coded_input->SetTotalBytesLimit(INT_MAX);
-  ModelProto model_proto;
-  bool result = model_proto.ParseFromCodedStream(coded_input.get());
-  coded_input.reset();
-  raw_input.reset();
-  EXPECT_TRUE(result);
-  ASSERT_STATUS_OK(Env::Default().FileClose(fd));
-
-  std::shared_ptr<Model> model;
-  ASSERT_STATUS_OK(Model::Load(ORT_TSTR("../models/opset8/test_bvlc_alexnet/model.onnx"), model, nullptr,
-                               *logger_));
-
-  // Check the graph input/output/value_info should have the same size as specified in the model file.
-  EXPECT_EQ(static_cast<size_t>(model_proto.graph().value_info_size()), model->MainGraph().GetValueInfo().size());
-  EXPECT_EQ(static_cast<size_t>(model_proto.graph().input_size()), model->MainGraph().GetInputs().size() + model->MainGraph().GetAllInitializedTensors().size());
-  EXPECT_EQ(static_cast<size_t>(model_proto.graph().output_size()), model->MainGraph().GetOutputs().size());
-  TestResolve(model->MainGraph());
-}
-
-TEST_P(ONNXModelsTest1, LoadFromFile) {
-  std::shared_ptr<Model> model;
-  ASSERT_STATUS_OK(Model::Load(GetModelFileName(), model, nullptr,
-                               *logger_));
-  TestResolve(model->MainGraph());
-}
-
-TEST_P(ONNXModelsTest1, LoadFromProtobuf) {
-  using ::google::protobuf::io::CodedInputStream;
-  using ::google::protobuf::io::FileInputStream;
-  using ::google::protobuf::io::ZeroCopyInputStream;
-  int fd;
-  ASSERT_STATUS_OK(Env::Default().FileOpenRd(GetModelFileName(), fd));
-  ASSERT_TRUE(fd > 0);
-  std::unique_ptr<ZeroCopyInputStream> raw_input(new FileInputStream(fd));
-  std::unique_ptr<CodedInputStream> coded_input(new CodedInputStream(raw_input.get()));
-  coded_input->SetTotalBytesLimit(INT_MAX);
-  ModelProto model_proto;
-  bool result = model_proto.ParseFromCodedStream(coded_input.get());
-  coded_input.reset();
-  raw_input.reset();
-  ASSERT_TRUE(result);
-  ASSERT_STATUS_OK(Env::Default().FileClose(fd));
-  std::shared_ptr<Model> model;
-  ASSERT_STATUS_OK(Model::Load(std::move(model_proto), model, nullptr,
-                               *logger_));
-  TestResolve(model->MainGraph());
-}
-
-#ifndef DISABLE_CONTRIB_OPS
-INSTANTIATE_TEST_SUITE_P(ONNXModelsTests,
-                         ONNXModelsTest1,
-                         ::testing::Values(ORT_TSTR("bvlc_alexnet"), ORT_TSTR("bvlc_googlenet"), ORT_TSTR("bvlc_reference_caffenet"), ORT_TSTR("bvlc_reference_rcnn_ilsvrc13"), ORT_TSTR("densenet121"), ORT_TSTR("emotion_ferplus"), ORT_TSTR("inception_v1"), ORT_TSTR("inception_v2"), ORT_TSTR("mnist"), ORT_TSTR("resnet50"), ORT_TSTR("shufflenet"), ORT_TSTR("squeezenet"), ORT_TSTR("tiny_yolov2"), ORT_TSTR("vgg19"), ORT_TSTR("zfnet512")));
-#else
-INSTANTIATE_TEST_SUITE_P(ONNXModelsTests,
-                         ONNXModelsTest1,
-                         ::testing::Values(ORT_TSTR("bvlc_alexnet"), ORT_TSTR("bvlc_googlenet"), ORT_TSTR("bvlc_reference_caffenet"), ORT_TSTR("bvlc_reference_rcnn_ilsvrc13"), ORT_TSTR("densenet121"), ORT_TSTR("emotion_ferplus"), ORT_TSTR("inception_v1"), ORT_TSTR("inception_v2"), ORT_TSTR("mnist"), ORT_TSTR("resnet50"), ORT_TSTR("shufflenet"), ORT_TSTR("squeezenet"), ORT_TSTR("vgg19"), ORT_TSTR("zfnet512")));
-#endif
-
-#endif
 
 // test a model that conforms to ONNX IR v4 where there are initializers that are not graph inputs.
 // a NodeArg should be created for all initializers in this case.
diff --git a/onnxruntime/test/optimizer/resnet50_fusion_test.cc b/onnxruntime/test/optimizer/resnet50_fusion_test.cc
index 5cb0206156a84..7e6677c8e1ddf 100644
--- a/onnxruntime/test/optimizer/resnet50_fusion_test.cc
+++ b/onnxruntime/test/optimizer/resnet50_fusion_test.cc
@@ -16,7 +16,6 @@
 
 namespace onnxruntime {
 namespace test {
-// #define ORT_RUN_EXTERNAL_ONNX_TESTS
 // #define MLAS_F16VEC_INTRINSICS_SUPPORTED
 
 #define MODEL_FOLDER ORT_TSTR("testdata/transform/")
@@ -28,54 +27,7 @@ class ResNet50FusionTests : public ::testing::Test {
   }
   std::unique_ptr<logging::Logger> logger;
 };
-#if defined(ORT_RUN_EXTERNAL_ONNX_TESTS)
-TEST_F(ResNet50FusionTests, FuseConvIntegrationTest) {
-  std::basic_string<ORTCHAR_T> fp32_model_path = ORT_TSTR("../models/opset10/Resnet50_Fusion_Testing/resnet50.onnx");
-  std::shared_ptr<Model> fp32_model;
-  std::basic_string<ORTCHAR_T> fp16_model_path = ORT_TSTR("../models/opset10/Resnet50_Fusion_Testing_fp16/resnet50.fp16.onnx");
-  std::shared_ptr<Model> fp16_model;
-  if (Model::Load(fp32_model_path, fp32_model, nullptr, *logger) != Status::OK()) {
-    GTEST_SKIP() << "Failed to load model: " << fp32_model_path;
-  }
-  if (Model::Load(fp16_model_path, fp16_model, nullptr, *logger) != Status::OK()) {
-    GTEST_SKIP() << "Failed to load model: " << fp16_model_path;
-  }
-  //  ASSERT_STATUS_OK(Model::Load(fp32_model_path, fp32_model, nullptr, *logger));
-  Graph& fp32_graph = fp32_model->MainGraph();
-  for (auto& node : fp32_model->MainGraph().Nodes()) {
-    node.SetExecutionProviderType(kCpuExecutionProvider);
-  }
-  Graph& fp16_graph = fp16_model->MainGraph();
-  for (auto& node : fp16_model->MainGraph().Nodes()) {
-    node.SetExecutionProviderType(kCpuExecutionProvider);
-  }
-  //  std::cout << "-------Op Counts Before Fusion---------" << std::endl;
-  std::map<std::string, int> fp32_op_count = CountOpsInGraph(fp32_graph);
-  std::map<std::string, int> fp16_op_count = CountOpsInGraph(fp16_graph);
-  for (auto& op : fp32_op_count) {
-    //    std::cout << op.first << " " << op.second << std::endl;
-    ASSERT_EQ(op.second, fp16_op_count[op.first]);
-  }
 
-  onnxruntime::GraphTransformerManager graph_transformation_mgr_32{5};
-  ASSERT_STATUS_OK(graph_transformation_mgr_32.Register(std::make_unique<ConvActivationFusion>(), TransformerLevel::Level3));
-  ASSERT_STATUS_OK(graph_transformation_mgr_32.Register(std::make_unique<ConvAddActivationFusion>(), TransformerLevel::Level3));
-  ASSERT_STATUS_OK(graph_transformation_mgr_32.ApplyTransformers(fp32_graph, TransformerLevel::Level3, *logger));
-  ASSERT_STATUS_OK(Model::Save(*fp32_model, ORT_TSTR("resnet50_fused.onnx")));
-
-  onnxruntime::GraphTransformerManager graph_transformation_mgr_16{5};
-  ASSERT_STATUS_OK(graph_transformation_mgr_16.Register(std::make_unique<ConvActivationFusion>(), TransformerLevel::Level3));
-  ASSERT_STATUS_OK(graph_transformation_mgr_16.Register(std::make_unique<ConvAddActivationFusion>(), TransformerLevel::Level3));
-  ASSERT_STATUS_OK(graph_transformation_mgr_16.ApplyTransformers(fp16_graph, TransformerLevel::Level3, *logger));
-  ASSERT_STATUS_OK(Model::Save(*fp16_model, ORT_TSTR("resnet50_fp16_fused.onnx")));
-  //  std::cout << "-------Op Counts After Fusion---------" << std::endl;
-  fp32_op_count = CountOpsInGraph(fp32_graph);
-  fp16_op_count = CountOpsInGraph(fp16_graph);
-  //  for (auto& op : fp32_op_count) {
-  //    ASSERT_EQ(op.second, fp16_op_count[op.first]);
-  //  }
-}
-#endif  // defined(ORT_RUN_EXTERNAL_ONNX_TESTS)
 TEST_F(ResNet50FusionTests, FuseConvAddReluUnitTest) {
   constexpr const ORTCHAR_T* model_uri = MODEL_FOLDER "fusion/conv_add_relu_fp16.onnx";
   std::shared_ptr<Model> p_model;
diff --git a/onnxruntime/test/providers/cpu/model_tests.cc b/onnxruntime/test/providers/cpu/model_tests.cc
index a17982ecb5eab..cf49601e6c671 100644
--- a/onnxruntime/test/providers/cpu/model_tests.cc
+++ b/onnxruntime/test/providers/cpu/model_tests.cc
@@ -10,6 +10,7 @@
 #include <filesystem>
 #include <utility>
 #include <unordered_map>
+#include <unordered_set>
 #include <gtest/gtest.h>
 
 #include "core/session/onnxruntime_c_api.h"
@@ -64,6 +65,157 @@ using namespace onnxruntime::common;
 
 namespace onnxruntime {
 namespace test {
+
+// Models verified to exist in both VM and Zoo with identical checksums
+// These 20 unique models have been confirmed as public (33 instances across opsets)
+static const std::unordered_set<std::string> VERIFIED_PUBLIC_MODELS = {
+    "AlexNet",
+    "BERT-Squad",
+    "CaffeNet",
+    "DenseNet-121",
+    "Emotion FERPlus",
+    "Faster R-CNN R-50-FPN",
+    "GoogleNet",
+    "Inception-1",
+    "Inception-2",
+    "Mask R-CNN R-50-FPN",
+    "MNIST",
+    "MobileNet v2-7",
+    "R-CNN ILSVRC13",
+    "ShuffleNet-v1",
+    "SqueezeNet 1.0",
+    "SqueezeNet 1.1",
+    "SSD",
+    "VGG 19-caffe2",
+    "YOLOv3",
+    "ZFNet-512"};
+
+// All ONNX Model Zoo models (always safe as they're public)
+// Total: 158 models from https://github.com/onnx/models
+static const std::unordered_set<std::string> ONNX_ZOO_MODELS = {
+    // Verified models (20 unique)
+    "AlexNet",
+    "BERT-Squad",
+    "CaffeNet",
+    "DenseNet-121",
+    "Emotion FERPlus",
+    "Faster R-CNN R-50-FPN",
+    "GoogleNet",
+    "Inception-1",
+    "Inception-2",
+    "Mask R-CNN R-50-FPN",
+    "MNIST",
+    "MobileNet v2-7",
+    "R-CNN ILSVRC13",
+    "ShuffleNet-v1",
+    "SqueezeNet 1.0",
+    "SqueezeNet 1.1",
+    "SSD",
+    "VGG 19-caffe2",
+    "YOLOv3",
+    "ZFNet-512",
+    // Additional Zoo-only models (138)
+    "AlexNet-int8",
+    "BERT-Squad-int8",
+    "BiDAF",
+    "BiDAF-int8",
+    "CaffeNet-int8",
+    "CaffeNet-qdq",
+    "Candy",
+    "DenseNet-121-12",
+    "DenseNet-121-12-int8",
+    "EfficientNet-Lite4",
+    "EfficientNet-Lite4-int8",
+    "EfficientNet-Lite4-qdq",
+    "Emotion FERPlus int8",
+    "FCN ResNet-50",
+    "FCN ResNet-50-int8",
+    "FCN ResNet-50-qdq",
+    "FCN ResNet-101",
+    "Faster R-CNN R-50-FPN-fp32",
+    "Faster R-CNN R-50-FPN-int8",
+    "Faster R-CNN R-50-FPN-qdq",
+    "GoogleNet-int8",
+    "GoogleNet-qdq",
+    "GPT-2",
+    "GPT-2-LM-HEAD",
+    "Inception-1-int8",
+    "Inception-1-qdq",
+    "LResNet100E-IR",
+    "LResNet100E-IR-int8",
+    "Mask R-CNN R-50-FPN-fp32",
+    "Mask R-CNN R-50-FPN-int8",
+    "Mask R-CNN R-50-FPN-qdq",
+    "MNIST-12",
+    "MNIST-12-int8",
+    "MobileNet v2-1.0",
+    "MobileNet v2-1.0-fp32",
+    "MobileNet v2-1.0-int8",
+    "MobileNet v2-1.0-qdq",
+    "Mosaic",
+    "Pointilism",
+    "Rain Princess",
+    "ResNet18",
+    "ResNet18-v2",
+    "ResNet34",
+    "ResNet34-v2",
+    "ResNet50",
+    "ResNet50-caffe2",
+    "ResNet50-fp32",
+    "ResNet50-int8",
+    "ResNet50-qdq",
+    "ResNet50-v2",
+    "ResNet101",
+    "ResNet101-v2",
+    "ResNet101_DUC_HDC",
+    "ResNet101_DUC_HDC-12",
+    "ResNet101_DUC_HDC-12-int8",
+    "ResNet152",
+    "ResNet152-v2",
+    "ResNet-preproc",
+    "RetinaNet (ResNet101 backbone)",
+    "RoBERTa-BASE",
+    "RoBERTa-SequenceClassification",
+    "ShuffleNet-v2",
+    "ShuffleNet-v2-fp32",
+    "ShuffleNet-v2-int8",
+    "ShuffleNet-v2-qdq",
+    "SqueezeNet 1.0-int8",
+    "SqueezeNet 1.0-qdq",
+    "SSD-int8",
+    "SSD-qdq",
+    "SSD-MobilenetV1",
+    "SSD-MobilenetV1-12",
+    "SSD-MobilenetV1-12-int8",
+    "SSD-MobilenetV1-12-qdq",
+    "Super_Resolution",
+    "T5-decoder-with-lm-head",
+    "T5-encoder",
+    "Tiny YOLOv2",
+    "Tiny YOLOv3",
+    "Udnie",
+    "VGG 16",
+    "VGG 16-bn",
+    "VGG 16-fp32",
+    "VGG 16-int8",
+    "VGG 16-qdq",
+    "VGG 19",
+    "VGG 19-bn",
+    "version-RFB-320",
+    "version-RFB-320-int8",
+    "version-RFB-640",
+    "YOLOv2",
+    "YOLOv3-12",
+    "YOLOv3-12-int8",
+    "YOLOv4",
+    "ZFNet-512-int8",
+    "ZFNet-512-qdq"};
+
+// Helper function to check if a model is allowed
+inline bool IsModelAllowed(const std::string& model_name) {
+  return ONNX_ZOO_MODELS.count(model_name) > 0;
+}
+
 // parameter is provider_name + "_" + model_path
 class ModelTest : public testing::TestWithParam<std::basic_string<ORTCHAR_T>> {};
 
@@ -656,15 +808,12 @@ ::std::vector<::std::basic_string<ORTCHAR_T>> GetParameterStrings() {
     // Same as the above, except this one is for large models
 #if defined(NDEBUG) || defined(RUN_MODELTEST_IN_DEBUG_MODE)
 #ifdef _WIN32
-    ORT_STRING_VIEW model_test_root_path = ORT_TSTR("..\\models");
-    // thus, only the root path should be mounted.
     ORT_STRING_VIEW model_zoo_path = ORT_TSTR("..\\models\\zoo");
 #else
-    ORT_STRING_VIEW model_test_root_path = ORT_TSTR("../models");
     ORT_STRING_VIEW model_zoo_path = ORT_TSTR("../models/zoo");
 #endif
     for (auto p : kvp.second) {
-      paths.push_back(ConcatPathComponent(model_test_root_path, p));
+      // ONLY use Model Zoo path - guaranteed public models with public test data
       paths.push_back(ConcatPathComponent(model_zoo_path, p));
     }
 #endif
@@ -750,6 +899,13 @@ ::std::vector<::std::basic_string<ORTCHAR_T>> GetParameterStrings() {
         std::basic_string<PATH_CHAR_TYPE> test_case_name = path.parent_path().filename().native();
         if (test_case_name.compare(0, 5, ORT_TSTR("test_")) == 0)
           test_case_name = test_case_name.substr(5);
+
+        // Check if model is in the public whitelist
+        std::string model_name_str = ToUTF8String(test_case_name);
+        if (!IsModelAllowed(model_name_str)) {
+          continue;  // Skip models not in whitelist
+        }
+
         if (all_disabled_tests.find(test_case_name) != all_disabled_tests.end())
           continue;
 
diff --git a/onnxruntime/test/shared_lib/test_inference.cc b/onnxruntime/test/shared_lib/test_inference.cc
index 786c0ba713b85..b7a9da8e1b658 100644
--- a/onnxruntime/test/shared_lib/test_inference.cc
+++ b/onnxruntime/test/shared_lib/test_inference.cc
@@ -1901,14 +1901,6 @@ TEST(CApiTest, test_pyop_kwarg) {
 }
 #endif
 
-#ifdef ORT_RUN_EXTERNAL_ONNX_TESTS
-TEST(CApiTest, create_session_without_session_option) {
-  constexpr PATH_TYPE model_uri = TSTR("../models/opset8/test_squeezenet/model.onnx");
-  Ort::Session ret(*ort_env, model_uri, Ort::SessionOptions{nullptr});
-  ASSERT_NE(nullptr, ret);
-}
-#endif
-
 #ifdef REDUCED_OPS_BUILD
 TEST(ReducedOpsBuildTest, test_excluded_ops) {
   // In reduced ops build, test a model containing ops not included in required_ops.config cannot be loaded.