diff --git a/lib/std/crypto/ml_kem.zig b/lib/std/crypto/ml_kem.zig
index 65d4c81ec9744f75123b89f31e6e00ae357c6845..4385da81b475860687358b01539fdf1302cae92a 100644
--- a/lib/std/crypto/ml_kem.zig
+++ b/lib/std/crypto/ml_kem.zig
@@ -1179,6 +1179,7 @@ const Poly = struct {
 
         const T = switch (builtin.target.cpu.arch) {
             .x86_64, .x86 => u32, // Generates better code on Intel CPUs
+            .riscv64 => u32, // Work around LLVM miscompiling coefficient extraction with u64 batches https://codeberg.org/ziglang/zig/issues/37128
             else => u64, // u128 might be faster on some other CPUs.
         };