diff --git a/ports/cxd56/common-hal/os/__init__.c b/ports/cxd56/common-hal/os/__init__.c
index e1024108b07..9797e641ae4 100644
--- a/ports/cxd56/common-hal/os/__init__.c
+++ b/ports/cxd56/common-hal/os/__init__.c
@@ -4,23 +4,28 @@
 //
 // SPDX-License-Identifier: MIT
 
-#include <stdlib.h>
-
 #include "genhdr/mpversion.h"
 #include "py/objstr.h"
 #include "py/objtuple.h"
 
+// The CXD5602 has no true hardware random number generator wired up in this
+// port. This function used to fill the buffer from the C library rand(), a
+// deterministic PRNG, and return true -- meaning os.urandom() handed back
+// predictable bytes while presenting itself as a cryptographically secure
+// source. That is the same class of defect behind the 2026 Coldcard wallet
+// key-recovery incident, where a silent PRNG fallback stood in for hardware
+// entropy.
+//
+// Fail closed instead: report that no secure entropy is available, so that
+// os.urandom() raises NotImplementedError ("No hardware random available")
+// rather than silently returning non-random data. This matches the broadcom,
+// silabs and litex ports, which also lack a hardware RNG.
+//
+// The correct long-term fix is to wire a real hardware entropy source for the
+// CXD5602 if the silicon exposes one; that requires the chip and datasheet and
+// is left as a follow-up.
 bool common_hal_os_urandom(uint8_t *buffer, mp_uint_t length) {
-    uint32_t i = 0;
-
-    while (i < length) {
-        uint32_t new_random = rand();
-        for (int j = 0; j < 4 && i < length; j++) {
-            buffer[i] = new_random & 0xff;
-            i++;
-            new_random >>= 8;
-        }
-    }
-
-    return true;
+    (void)buffer;
+    (void)length;
+    return false;
 }