diff --git a/cuda_core/cuda/core/_memoryview.pyx b/cuda_core/cuda/core/_memoryview.pyx
index 260980c1daf..d6dd9bb2454 100644
--- a/cuda_core/cuda/core/_memoryview.pyx
+++ b/cuda_core/cuda/core/_memoryview.pyx
@@ -346,13 +346,15 @@ cdef class StridedMemoryView:
             data = cpython.PyCapsule_GetPointer(
                 self.metadata, DLPACK_VERSIONED_TENSOR_USED_NAME)
             dlm_tensor_ver = <DLManagedTensorVersioned*>data
-            dlm_tensor_ver.deleter(dlm_tensor_ver)
+            if dlm_tensor_ver.deleter != NULL:
+                dlm_tensor_ver.deleter(dlm_tensor_ver)
         elif cpython.PyCapsule_IsValid(
                 self.metadata, DLPACK_TENSOR_USED_NAME):
             data = cpython.PyCapsule_GetPointer(
                 self.metadata, DLPACK_TENSOR_USED_NAME)
             dlm_tensor = <DLManagedTensor*>data
-            dlm_tensor.deleter(dlm_tensor)
+            if dlm_tensor.deleter != NULL:
+                dlm_tensor.deleter(dlm_tensor)
 
     def view(
         self, layout : _StridedLayout | None = None, dtype : numpy.dtype | None = None
@@ -1176,7 +1178,7 @@ cdef object dtype_dlpack_to_numpy(DLDataType* dtype):
 
 cpdef StridedMemoryView view_as_cai(obj, stream_ptr, view=None):
     cdef dict cai_data = obj.__cuda_array_interface__
-    if cai_data["version"] < 3:
+    if cai_data.get("version", 0) < 3:
         raise BufferError("only CUDA Array Interface v3 or above is supported")
     if cai_data.get("mask") is not None:
         raise BufferError("mask is not supported")
@@ -1232,7 +1234,7 @@ cpdef StridedMemoryView view_as_cai(obj, stream_ptr, view=None):
 
 cpdef StridedMemoryView view_as_array_interface(obj, view=None):
     cdef dict data = obj.__array_interface__
-    if data["version"] < 3:
+    if data.get("version", 0) < 3:
         raise BufferError("only NumPy Array Interface v3 or above is supported")
     if data.get("mask") is not None:
         raise BufferError("mask is not supported")
diff --git a/cuda_core/tests/test_utils.py b/cuda_core/tests/test_utils.py
index ebee8d87b04..e637aac0a0f 100644
--- a/cuda_core/tests/test_utils.py
+++ b/cuda_core/tests/test_utils.py
@@ -3,6 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import ctypes
+import functools
 import math
 
 # TODO: replace optional imports with pytest.importorskip
@@ -676,20 +677,26 @@ def test_from_array_interface_unsupported_strides(init_cuda):
         StridedMemoryView.from_array_interface(b)
 
 
-def _make_cuda_array_interface_obj(*, shape, strides, typestr="<f8", data=(0, False), version=3):
-    return type(
-        "SyntheticCAI",
-        (),
-        {
-            "__cuda_array_interface__": {
-                "shape": shape,
-                "strides": strides,
-                "typestr": typestr,
-                "data": data,
-                "version": version,
-            }
-        },
-    )()
+def _make_interface_obj(attr_name, *, shape, strides, typestr="<f8", data=(0, False), version=3, **extra):
+    """Create an object exposing an array-interface dict as ``attr_name``.
+
+    ``version=None`` omits the key entirely, so tests can exercise interfaces
+    that don't declare a version. Extra kwargs are added to the interface dict.
+    """
+    iface = {
+        "shape": shape,
+        "strides": strides,
+        "typestr": typestr,
+        "data": data,
+    }
+    if version is not None:
+        iface["version"] = version
+    iface.update(extra)
+    return type(f"Synthetic{attr_name}", (), {attr_name: iface})()
+
+
+_make_cuda_array_interface_obj = functools.partial(_make_interface_obj, "__cuda_array_interface__")
+_make_array_interface_obj = functools.partial(_make_interface_obj, "__array_interface__")
 
 
 def test_from_cuda_array_interface_unsupported_strides(init_cuda):
@@ -902,54 +909,12 @@ def test_dlpack_export_unsupported_dtype_raises():
         bad_view.__dlpack__()
 
 
-class _FakeCAIv2:
-    """Object with CUDA Array Interface v2 (unsupported)."""
-
-    def __init__(self):
-        self.__cuda_array_interface__ = {
-            "version": 2,
-            "shape": (5,),
-            "typestr": "<f4",
-            "data": (0, False),
-        }
-
-
-class _FakeCAIWithMask:
-    """Object with CUDA Array Interface that has a mask."""
-
-    def __init__(self):
-        self.__cuda_array_interface__ = {
-            "version": 3,
-            "shape": (5,),
-            "typestr": "<f4",
-            "data": (0, False),
-            "mask": np.ones(5, dtype=bool),
-        }
-
-
-class _FakeArrayInterfacev2:
-    """Object with NumPy Array Interface v2 (unsupported)."""
-
-    def __init__(self, arr):
-        iface = dict(arr.__array_interface__)
-        iface["version"] = 2
-        self.__array_interface__ = iface
-
-
-class _FakeArrayInterfaceWithMask:
-    """Object with NumPy Array Interface that has a mask."""
-
-    def __init__(self, arr):
-        iface = dict(arr.__array_interface__)
-        iface["mask"] = np.ones(arr.shape, dtype=bool)
-        self.__array_interface__ = iface
-
-
-def test_cai_v2_rejected():
-    """CUDA Array Interface v2 raises BufferError."""
+@pytest.mark.parametrize("version", [2, None])
+def test_cai_version_rejected(version):
+    """CUDA Array Interface below v3 (or missing version) raises BufferError."""
     from cuda.core._memoryview import view_as_cai
 
-    obj = _FakeCAIv2()
+    obj = _make_cuda_array_interface_obj(shape=(5,), strides=None, version=version)
     with pytest.raises(BufferError, match="v3 or above"):
         view_as_cai(obj, stream_ptr=-1)
 
@@ -958,38 +923,26 @@ def test_cai_mask_rejected():
     """CUDA Array Interface with mask raises BufferError."""
     from cuda.core._memoryview import view_as_cai
 
-    obj = _FakeCAIWithMask()
+    obj = _make_cuda_array_interface_obj(shape=(5,), strides=None, mask=np.ones(5, dtype=bool))
     with pytest.raises(BufferError, match="mask is not supported"):
         view_as_cai(obj, stream_ptr=-1)
 
 
-class _FakeCAIv3:
-    """Valid CUDA Array Interface v3 object (for stream=None test)."""
-
-    def __init__(self):
-        self.__cuda_array_interface__ = {
-            "version": 3,
-            "shape": (5,),
-            "typestr": "<f4",
-            "data": (0, False),
-        }
-
-
 def test_cai_stream_none_rejected():
     """CUDA Array Interface with stream=None raises BufferError."""
     from cuda.core._memoryview import view_as_cai
 
-    obj = _FakeCAIv3()
+    obj = _make_cuda_array_interface_obj(shape=(5,), strides=None)
     with pytest.raises(BufferError, match="stream=None is ambiguous"):
         view_as_cai(obj, stream_ptr=None)
 
 
-def test_array_interface_v2_rejected():
-    """NumPy Array Interface v2 raises BufferError."""
+@pytest.mark.parametrize("version", [2, None])
+def test_array_interface_version_rejected(version):
+    """NumPy Array Interface below v3 (or missing version) raises BufferError."""
     from cuda.core._memoryview import view_as_array_interface
 
-    arr = np.zeros(5, dtype=np.float32)
-    obj = _FakeArrayInterfacev2(arr)
+    obj = _make_array_interface_obj(shape=(5,), strides=None, version=version)
     with pytest.raises(BufferError, match="v3 or above"):
         view_as_array_interface(obj)
 
@@ -998,8 +951,7 @@ def test_array_interface_mask_rejected():
     """NumPy Array Interface with mask raises BufferError."""
     from cuda.core._memoryview import view_as_array_interface
 
-    arr = np.zeros(5, dtype=np.float32)
-    obj = _FakeArrayInterfaceWithMask(arr)
+    obj = _make_array_interface_obj(shape=(5,), strides=None, mask=np.ones(5, dtype=bool))
     with pytest.raises(BufferError, match="mask is not supported"):
         view_as_array_interface(obj)
 
diff --git a/cuda_core/tests/test_utils_dlpack.py b/cuda_core/tests/test_utils_dlpack.py
index 8033d8250b2..9c2ff64a52f 100644
--- a/cuda_core/tests/test_utils_dlpack.py
+++ b/cuda_core/tests/test_utils_dlpack.py
@@ -201,6 +201,56 @@ class _DLTensor(ctypes.Structure):
     ]
 
 
+class _DLManagedTensor(ctypes.Structure):
+    _fields_ = [
+        ("dl_tensor", _DLTensor),
+        ("manager_ctx", ctypes.c_void_p),
+        ("deleter", ctypes.c_void_p),
+    ]
+
+
+class _DLManagedTensorVersioned(ctypes.Structure):
+    _fields_ = [
+        ("version", _DLPackVersion),
+        ("manager_ctx", ctypes.c_void_p),
+        ("deleter", ctypes.c_void_p),
+        ("flags", ctypes.c_uint64),
+        ("dl_tensor", _DLTensor),
+    ]
+
+
+@pytest.mark.agent_authored(model="cursor-grok-4.5")
+@pytest.mark.parametrize(
+    "max_version, capsule_name, managed_cls",
+    [
+        pytest.param(None, b"dltensor", _DLManagedTensor, id="unversioned"),
+        pytest.param((1, 0), b"dltensor_versioned", _DLManagedTensorVersioned, id="versioned"),
+    ],
+)
+def test_from_dlpack_null_deleter_dealloc(max_version, capsule_name, managed_cls):
+    """``__dealloc__`` must tolerate a capsule whose deleter is NULL."""
+    src = np.arange(6, dtype=np.int32)
+    base = StridedMemoryView.from_any_interface(src, stream_ptr=-1)
+    capsule = base.__dlpack__(max_version=max_version)
+    dlm = ctypes.cast(_PyCapsule_GetPointer(capsule, capsule_name), ctypes.POINTER(managed_cls))
+    # Steal the producer deleter so __dealloc__ sees NULL, then invoke it ourselves.
+    producer_deleter = ctypes.CFUNCTYPE(None, ctypes.POINTER(managed_cls))(dlm.contents.deleter)
+    dlm.contents.deleter = None
+
+    class _Export:
+        def __dlpack_device__(self):
+            return base.__dlpack_device__()
+
+        def __dlpack__(self, stream=None, max_version=None, **kwargs):
+            if capsule_name == b"dltensor" and max_version is not None:
+                raise TypeError("force unversioned")
+            return capsule
+
+    view = StridedMemoryView.from_dlpack(_Export(), stream_ptr=-1)
+    del view  # __dealloc__ must not call the NULL deleter
+    producer_deleter(dlm)
+
+
 _FN_FROM_PY = ctypes.PYFUNCTYPE(ctypes.c_int, ctypes.c_void_p, ctypes.POINTER(ctypes.c_void_p))
 _FN_TO_PY = ctypes.PYFUNCTYPE(ctypes.c_int, ctypes.c_void_p, ctypes.POINTER(ctypes.c_void_p))
 _FN_DLTENSOR_FROM_PY = ctypes.PYFUNCTYPE(ctypes.c_int, ctypes.c_void_p, ctypes.c_void_p)