diff --git a/crypto_kem/ml-kem-512/m4fspeed/poly.c b/crypto_kem/ml-kem-512/m4fspeed/poly.c
index 401b26b..5a6d0ab 100644
--- a/crypto_kem/ml-kem-512/m4fspeed/poly.c
+++ b/crypto_kem/ml-kem-512/m4fspeed/poly.c
@@ -366,6 +366,7 @@ void poly_tobytes(unsigned char *r, poly *a) {
     int i;
     uint16_t t0, t1;
 
+    poly_reduce(a);
     poly_reduce(a);
 
     for (i = 0; i < KYBER_N / 2; i++) {
diff --git a/crypto_kem/ml-kem-512/m4fstack/poly.c b/crypto_kem/ml-kem-512/m4fstack/poly.c
index 443fdba..ed54ec8 100644
--- a/crypto_kem/ml-kem-512/m4fstack/poly.c
+++ b/crypto_kem/ml-kem-512/m4fstack/poly.c
@@ -366,6 +366,7 @@ void poly_tobytes(unsigned char *r, poly *a) {
     int i;
     uint16_t t0, t1;
 
+    poly_reduce(a);
     poly_reduce(a);
 
     for (i = 0; i < KYBER_N / 2; i++) {
diff --git a/crypto_kem/ml-kem-768/m4fspeed/poly.c b/crypto_kem/ml-kem-768/m4fspeed/poly.c
index b52060f..bf549ef 100644
--- a/crypto_kem/ml-kem-768/m4fspeed/poly.c
+++ b/crypto_kem/ml-kem-768/m4fspeed/poly.c
@@ -366,6 +366,7 @@ void poly_tobytes(unsigned char *r, poly *a) {
     int i;
     uint16_t t0, t1;
 
+    poly_reduce(a);
     poly_reduce(a);
 
     for (i = 0; i < KYBER_N / 2; i++) {
diff --git a/crypto_kem/ml-kem-768/m4fstack/poly.c b/crypto_kem/ml-kem-768/m4fstack/poly.c
index 35475ad..a5a694a 100644
--- a/crypto_kem/ml-kem-768/m4fstack/poly.c
+++ b/crypto_kem/ml-kem-768/m4fstack/poly.c
@@ -366,6 +366,7 @@ void poly_tobytes(unsigned char *r, poly *a) {
     int i;
     uint16_t t0, t1;
 
+    poly_reduce(a);
     poly_reduce(a);
 
     for (i = 0; i < KYBER_N / 2; i++) {