diff --git a/sycl/include/sycl/ext/intel/esimd/memory.hpp b/sycl/include/sycl/ext/intel/esimd/memory.hpp
index eab1b92013ef7..e07d2386c0c89 100644
--- a/sycl/include/sycl/ext/intel/esimd/memory.hpp
+++ b/sycl/include/sycl/ext/intel/esimd/memory.hpp
@@ -529,12 +529,23 @@ gather_rgba(const Tx *p, simd<uint32_t, N> offsets, simd_mask<N> mask = 1) {
   return __esimd_svm_gather4_scaled<T, N, Mask>(addrs.data(), mask.data());
 }
 
+namespace detail {
+template <rgba_channel_mask M> static void validate_rgba_write_channel_mask() {
+  using CM = rgba_channel_mask;
+  static_assert(
+      (M == CM::ABGR || M == CM::BGR || M == CM::GR || M == CM::R) &&
+      "Only ABGR, BGR, GR, R channel masks are valid in write operations");
+}
+} // namespace detail
+
 /// @anchor usm_scatter_rgba
 /// Transpose and scatter pixels to given memory locations defined by the base
 /// pointer \c p and \c offsets. Up to 4 32-bit data elements may be accessed at
 /// each address depending on the channel mask \c Mask template parameter. Each
 /// pixel's address must be 4 byte aligned. This is basically an inverse
-/// operation for gather_rgba.
+/// operation for gather_rgba. Unlike \c gather_rgba, this function imposes
+/// restrictions on possible \c Mask template argument values. It can only be
+/// one of the following: \c ABGR, \c BGR, \c GR, \c R.
 ///
 /// @tparam Tx Element type of the returned vector. Must be 4 bytes in size.
 /// @tparam N Number of pixels to access (matches the size of the \c offsets
@@ -553,6 +564,7 @@ __ESIMD_API std::enable_if_t<(N == 8 || N == 16 || N == 32) && (sizeof(T) == 4)>
 scatter_rgba(Tx *p, simd<uint32_t, N> offsets,
              simd<Tx, N * get_num_channels_enabled(Mask)> vals,
              simd_mask<N> mask = 1) {
+  detail::validate_rgba_write_channel_mask<Mask>();
   simd<uint64_t, N> offsets_i = convert<uint64_t>(offsets);
   simd<uint64_t, N> addrs(reinterpret_cast<uint64_t>(p));
   addrs = addrs + offsets_i;
@@ -875,7 +887,7 @@ slm_gather_rgba(simd<uint32_t, N> offsets, simd_mask<N> mask = 1) {
 }
 
 /// Gather data from the Shared Local Memory at specified \c offsets and return
-/// it as simd vector. See @ref usm_gather_rgba for information about the
+/// it as simd vector. See @ref usm_scatter_rgba for information about the
 /// operation semantics and parameter restrictions/interdependencies.
 /// @tparam T The element type of the returned vector.
 /// @tparam N The number of elements to access.
@@ -889,6 +901,7 @@ __ESIMD_API std::enable_if_t<(N == 8 || N == 16 || N == 32) && (sizeof(T) == 4)>
 slm_scatter_rgba(simd<uint32_t, N> offsets,
                  simd<T, N * get_num_channels_enabled(Mask)> vals,
                  simd_mask<N> mask = 1) {
+  detail::validate_rgba_write_channel_mask<Mask>();
   const auto si = __ESIMD_GET_SURF_HANDLE(detail::LocalAccessorMarker());
   constexpr int16_t Scale = 0;
   constexpr int global_offset = 0;
diff --git a/sycl/test/esimd/gather_scatter_rgba.cpp b/sycl/test/esimd/gather_scatter_rgba.cpp
index 3572f509762f9..8b22c9b1485ce 100644
--- a/sycl/test/esimd/gather_scatter_rgba.cpp
+++ b/sycl/test/esimd/gather_scatter_rgba.cpp
@@ -1,6 +1,9 @@
-// RUN: %clangxx -fsycl -fsyntax-only -Wno-unused-command-line-argument %s
+// RUN: %clangxx -fsycl -fsycl-device-only -fsyntax-only -Xclang -verify %s
 
-// This test checks compilation of ESIMD slm gather_rgba/scatter_rgba APIs.
+// This test checks that device compiler can:
+// - successfully compile gather_rgba/scatter_rgba APIs
+// - emit an error if some of the restrictions on template parameters are
+//   violated
 
 #include <CL/sycl.hpp>
 #include <limits>
@@ -20,3 +23,15 @@ void kernel(int *ptr) SYCL_ESIMD_FUNCTION {
 
   scatter_rgba<int, 32, rgba_channel_mask::ABGR>(ptr, offsets, v0);
 }
+
+constexpr int AGR_N_CHANNELS = 3;
+
+void kernel1(int *ptr, simd<int, 32 * AGR_N_CHANNELS> v) SYCL_ESIMD_FUNCTION {
+  simd<uint32_t, 32> offsets(0, sizeof(int) * 4);
+  // only 1, 2, 3, 4-element masks covering consequitive channels starting from
+  // R are supported
+  // expected-error-re@* {{static_assert failed{{.*}}Only ABGR, BGR, GR, R channel masks are valid in write operations}}
+  // expected-note@* {{in instantiation }}
+  // expected-note@+1 {{in instantiation }}
+  scatter_rgba<int, 32, rgba_channel_mask::AGR>(ptr, offsets, v);
+}