diff --git a/cuda_core/cuda/core/utils/_program_cache/_abc.py b/cuda_core/cuda/core/utils/_program_cache/_abc.py
index 82233540eb3..b9289ad0494 100644
--- a/cuda_core/cuda/core/utils/_program_cache/_abc.py
+++ b/cuda_core/cuda/core/utils/_program_cache/_abc.py
@@ -111,8 +111,21 @@ class ProgramCacheResource(abc.ABC):
         ``__contains__`` invites that pattern. ``get`` answers
         both questions in one filesystem-level operation, so a
         successful return always carries the bytes.
+
+        ``key in cache`` therefore raises ``TypeError``, and a cache
+        is not iterable.
     """
 
+    # Opt out of the legacy sequence-iteration protocol. Defining
+    # ``__getitem__`` without ``__iter__`` makes ``key in cache`` fall back to
+    # ``cache[0], cache[1], ...`` compared against the *values* -- which
+    # defeats the "no ``__contains__``" design above: on these backends it
+    # raises a baffling "cache keys must be bytes or str, got int", and on a
+    # subclass whose ``__getitem__`` accepts integers it silently answers about
+    # values instead of keys. ``__iter__ = None`` restores the plain
+    # ``TypeError: argument of type '...' is not iterable``.
+    __iter__ = None
+
     @abc.abstractmethod
     def __getitem__(self, key: bytes | str) -> bytes:
         """Retrieve the cached binary bytes.
diff --git a/cuda_core/docs/source/release/1.2.0-notes.rst b/cuda_core/docs/source/release/1.2.0-notes.rst
index 120d2c2a253..ac993a9d636 100644
--- a/cuda_core/docs/source/release/1.2.0-notes.rst
+++ b/cuda_core/docs/source/release/1.2.0-notes.rst
@@ -73,6 +73,14 @@ Fixes and enhancements
   Windows, both ``ctypes.CFUNCTYPE`` and ``ctypes.WINFUNCTYPE`` are accepted.
   (`#2439 <https://github.com/NVIDIA/cuda-python/issues/2439>`__)
 
+- ``key in cache`` on a :class:`~cuda.core.utils.ProgramCacheResource` now
+  raises ``TypeError`` instead of walking the cache as a legacy sequence.
+  The class deliberately provides no ``__contains__``, but defining
+  ``__getitem__`` without ``__iter__`` made ``in`` fall back to
+  ``cache[0], cache[1], ...`` compared against the values. Use
+  :meth:`~cuda.core.utils.ProgramCacheResource.get`, which answers presence and
+  retrieval in one operation.
+
 Deprecation Notices
 -------------------
 
diff --git a/cuda_core/tests/test_program_cache.py b/cuda_core/tests/test_program_cache.py
index a8d3fc85f7e..6ae16fbbe1a 100644
--- a/cuda_core/tests/test_program_cache.py
+++ b/cuda_core/tests/test_program_cache.py
@@ -38,6 +38,37 @@ def test_program_cache_resource_requires_core_methods():
     assert "__contains__" not in ProgramCacheResource.__abstractmethods__
 
 
+@pytest.mark.agent_authored(model="claude-opus-5")
+@pytest.mark.parametrize("factory", ["inmemory", "filestream"])
+def test_program_cache_is_not_iterable_and_rejects_in(tmp_path, factory):
+    """The ABC documents having no ``__contains__``, but defining
+    ``__getitem__`` without ``__iter__`` handed ``key in cache`` to the legacy
+    sequence protocol: it walked ``cache[0], cache[1], ...`` and compared
+    against the *values*. On these backends that surfaced as a baffling
+    "cache keys must be bytes or str, got int"; on a subclass whose
+    ``__getitem__`` accepts integers it silently answered about values instead
+    of keys.
+    """
+    from cuda.core.utils import FileStreamProgramCache, InMemoryProgramCache
+
+    cache = InMemoryProgramCache() if factory == "inmemory" else FileStreamProgramCache(tmp_path / "fc")
+    with cache:
+        cache[b"k"] = b"v"
+
+        with pytest.raises(TypeError, match="not iterable"):
+            b"k" in cache  # noqa: B015
+        with pytest.raises(TypeError, match="not iterable"):
+            iter(cache)
+        with pytest.raises(TypeError, match="not iterable"):
+            list(cache)
+
+        # The supported lookups are unaffected.
+        assert cache[b"k"] == b"v"
+        assert cache.get(b"k") == b"v"
+        assert cache.get(b"absent") is None
+        assert len(cache) == 1
+
+
 def _build_empty_subclass():
     from cuda.core.utils import ProgramCacheResource