diff --git a/CMakeLists.txt b/CMakeLists.txt
index fefd839f..9208f28d 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -53,6 +53,7 @@ option(ENABLE_IMPLICIT_CONVERSIONS "Enable implicit conversions to-from raw poin
 option(DISABLE_RM "Make ManagedArray a thin wrapper" Off)
 mark_as_advanced(DISABLE_RM)
 option(ENABLE_UM "Use CUDA unified (managed) memory" Off)
+option(ENABLE_BOUNDS_CHECK "Enable bounds checking for chai::ManagedArray<T>::operator[]" Off)
 
 set(ENABLE_TESTS On CACHE BOOL "")
 set(ENABLE_EXAMPLES On CACHE BOOL "")
diff --git a/src/chai/CMakeLists.txt b/src/chai/CMakeLists.txt
index 7821fa3f..64577fd9 100644
--- a/src/chai/CMakeLists.txt
+++ b/src/chai/CMakeLists.txt
@@ -47,6 +47,7 @@ set(CHAI_ENABLE_HIP ${ENABLE_HIP})
 set(CHAI_ENABLE_IMPLICIT_CONVERSIONS ${ENABLE_IMPLICIT_CONVERSIONS})
 set(CHAI_DISABLE_RM ${DISABLE_RM})
 set(CHAI_ENABLE_UM ${ENABLE_UM})
+set(CHAI_ENABLE_BOUNDS_CHECK ${ENABLE_BOUNDS_CHECK})
 
 configure_file(
   ${PROJECT_SOURCE_DIR}/src/chai/config.hpp.in
diff --git a/src/chai/ManagedArray.inl b/src/chai/ManagedArray.inl
index 18f4db7a..ecfe52b5 100644
--- a/src/chai/ManagedArray.inl
+++ b/src/chai/ManagedArray.inl
@@ -46,6 +46,28 @@
 #include "ManagedArray.hpp"
 #include "ArrayManager.hpp"
 
+#if defined(CHAI_ENABLE_BOUNDS_CHECK)
+#if defined(NDEBUG)
+
+#define HOST_ALWAYS_ASSERT(EXP) if (!(EXP)) { \
+   printf("Assert triggered at %s:%d\n", __FILE__, __LINE__); \
+   abort(); \
+}
+
+#define DEVICE_ALWAYS_ASSERT(EXP) if (!(EXP)) { \
+   asm("trap;") \
+}
+
+#else // defined(NDEBUG)
+
+#include <cassert>
+
+#define HOST_ALWAYS_ASSERT(EXP) assert(EXP)
+#define DEVICE_ALWAYS_ASSERT(EXP) assert(EXP)
+
+#endif // defined(NDEBUG)
+#endif // defined(CHAI_ENABLE_BOUNDS_CHECK)
+
 namespace chai {
 
 template<typename T>
@@ -389,6 +411,20 @@ template<typename T>
 template<typename Idx>
 CHAI_INLINE
 CHAI_HOST_DEVICE T& ManagedArray<T>::operator[](const Idx i) const {
+#if defined(CHAI_ENABLE_BOUNDS_CHECK)
+#if defined(__CUDA_ARCH__)
+  DEVICE_ALWAYS_ASSERT(i >= 0 && static_cast<size_t>(i) < m_elems);
+#else
+  //std::cout << "m_elems = " << m_elems << std::endl;
+  //std::cout << "i = " << i << std::endl;
+  //std::cout << "(size_t) i = " << static_cast<size_t>(i) << std::endl;
+
+  //printf("m_elems = %d\n", m_elems);
+  //printf("i = %d\n", i);
+  HOST_ALWAYS_ASSERT(i >= 0 && static_cast<size_t>(i) < m_elems);
+#endif
+#endif // defined(CHAI_ENABLE_BOUNDS_CHECK)
+
   return m_active_pointer[i];
 }
 
diff --git a/src/chai/config.hpp.in b/src/chai/config.hpp.in
index a7a443fe..310eeaa7 100644
--- a/src/chai/config.hpp.in
+++ b/src/chai/config.hpp.in
@@ -49,5 +49,6 @@
 #cmakedefine CHAI_ENABLE_IMPLICIT_CONVERSIONS
 #cmakedefine CHAI_DISABLE_RM
 #cmakedefine CHAI_ENABLE_UM
+#cmakedefine CHAI_ENABLE_BOUNDS_CHECK
 
 #endif // CHAI_config_HPP
diff --git a/src/util/forall.hpp b/src/util/forall.hpp
index 4cb537f0..d18b62a5 100644
--- a/src/util/forall.hpp
+++ b/src/util/forall.hpp
@@ -62,7 +62,7 @@ template <typename LOOP_BODY>
 void forall_kernel_cpu(int begin, int end, LOOP_BODY body)
 {
   for (int i = 0; i < (end - begin); ++i) {
-    body(i);
+    body(begin+i);
   }
 }
 
diff --git a/tests/integration/managed_array_tests.cpp b/tests/integration/managed_array_tests.cpp
index 842293e7..e271668c 100644
--- a/tests/integration/managed_array_tests.cpp
+++ b/tests/integration/managed_array_tests.cpp
@@ -1394,4 +1394,59 @@ GPU_TEST(ManagedArray, CopyZero)
 
   array.free();
 }
-#endif
\ No newline at end of file
+#endif
+
+#if defined(CHAI_ENABLE_BOUNDS_CHECK)
+
+TEST(ManagedArray, UpperOutOfRangeAccess)
+{
+   chai::ManagedArray<float> array(20);
+   array[19] = 0.0; // Should be fine
+   ASSERT_DEATH_IF_SUPPORTED(array[20] = 0.0, ".*");
+}
+
+TEST(ManagedArray, LowerOutOfRangeAccess)
+{
+   chai::ManagedArray<float> array(20);
+   array[0] = 0.0; // Should be fine
+   ASSERT_DEATH_IF_SUPPORTED(array[-1] = 0.0, ".*");
+}
+
+#if defined(CHAI_ENABLE_CUDA)
+
+GPU_TEST(ManagedArray, UpperOutOfRangeAccessGPU)
+{
+   ASSERT_EQ(cudaSuccess, cudaDeviceReset());
+   ASSERT_EQ(cudaSuccess, cudaDeviceSynchronize());
+
+   chai::ManagedArray<float> array(20);
+
+   ASSERT_DEATH_IF_SUPPORTED(forall(gpu(), 0, 1, [=] __device__ (int) {
+      array[20] = 0.0;
+   }), ".*");
+
+   //cudaError_t errorCode = cudaGetLastError();
+   cudaError_t errorCode = cudaDeviceSynchronize();
+   ASSERT_EQ(cudaErrorAssert, errorCode);
+   ASSERT_EQ(cudaDeviceReset(), cudaSuccess);
+   ASSERT_EQ(cudaSuccess, cudaDeviceSynchronize());
+}
+
+GPU_TEST(ManagedArray, LowerOutOfRangeAccessGPU)
+{
+   chai::ManagedArray<float> array(20);
+
+#if 0
+   forall(gpu(), 0, 1, [=] __device__ (int) {
+      array[-1] = 0.0;
+   });
+
+   //cudaError_t errorCode = cudaGetLastError();
+   cudaError_t errorCode = cudaDeviceSynchronize();
+   ASSERT_EQ(cudaErrorAssert, errorCode);
+   ASSERT_EQ(cudaDeviceReset(), cudaSuccess);
+#endif
+}
+
+#endif // defined(CHAI_ENABLE_CUDA)
+#endif // defined(CHAI_ENABLE_BOUNDS_CHECK)