diff --git a/python/runtime/cudaq/analysis/py_dem.cpp b/python/runtime/cudaq/analysis/py_dem.cpp
index d29d34220d7..65538da1b11 100644
--- a/python/runtime/cudaq/analysis/py_dem.cpp
+++ b/python/runtime/cudaq/analysis/py_dem.cpp
@@ -15,6 +15,7 @@
 #include <nanobind/stl/optional.h>
 #include <nanobind/stl/string.h>
 #include <optional>
+#include <stdexcept>
 #include <string>
 
 using namespace cudaq;
@@ -27,10 +28,23 @@ static std::string dem_from_kernel_impl(const std::string &kernelName,
   args = simplifiedValidateInputArguments(args);
 
   const cudaq::noise_model *noisePtr = noise ? &(*noise) : nullptr;
-  return cudaq::detail::runDemFromKernel(kernelName, platform, noisePtr, [&]() {
-    [[maybe_unused]] auto result =
-        cudaq::marshal_and_launch_module(kernelName, kernelMod, args);
-  });
+  // Re-throw as a module-local RuntimeError; plugin exceptions otherwise
+  // escape uncaught on macOS and abort instead of raising in Python.
+  try {
+    return cudaq::detail::runDemFromKernel(
+        kernelName, platform, noisePtr, [&]() {
+          [[maybe_unused]] auto result =
+              cudaq::marshal_and_launch_module(kernelName, kernelMod, args);
+        });
+  } catch (nanobind::python_error &) {
+    throw; // keep real Python exceptions (e.g. from argument marshalling)
+  } catch (const std::exception &e) {
+    throw std::runtime_error(e.what()); // dem_from_kernel reports via RuntimeError
+  } catch (...) {
+    throw std::runtime_error(
+        "cudaq::dem_from_kernel failed: the kernel uses an operation the Stim "
+        "analysis backend cannot simulate (only Clifford gates are supported).");
+  }
 }
 
 void cudaq::bindDemFromKernel(nanobind::module_ &mod) {