diff --git a/tinyxml2.cpp b/tinyxml2.cpp
index ad5bc870..c436fb24 100644
--- a/tinyxml2.cpp
+++ b/tinyxml2.cpp
@@ -552,12 +552,15 @@ const char* XMLUtil::GetCharacterRef(const char* p, char* value, int* length)
             TIXMLASSERT(digit < radix);
 
             const unsigned int digitScaled = mult * digit;
+            // Reject before adding: if digitScaled alone exceeds MAX_CODE_POINT,
+            // or if adding it to ucs would exceed it (checked without overflow by
+            // testing ucs > MAX_CODE_POINT - digitScaled, safe since digitScaled
+            // <= MAX_CODE_POINT at this point).
+            if (digitScaled > MAX_CODE_POINT || ucs > MAX_CODE_POINT - digitScaled) {
+                return 0;
+            }
             ucs += digitScaled;
-            mult *= radix;       
-            
-            // Security check: could a value exist that is out of range?
-            // Easily; limit to the MAX_CODE_POINT, which also allows for a
-            // bunch of leading zeroes.
+            mult *= radix;
             if (mult > MAX_CODE_POINT) {
                 mult = MAX_CODE_POINT;
             }
@@ -569,11 +572,11 @@ const char* XMLUtil::GetCharacterRef(const char* p, char* value, int* length)
         }
         // convert the UCS to UTF-8
         ConvertUTF32ToUTF8(ucs, value, length);
-		if (length == 0) {
-            // If length is 0, there was an error. (Security? Bad input?)
+        if (*length == 0) {
+            // If *length is 0, ConvertUTF32ToUTF8 rejected the code point.
             // Fail safely.
-			return 0;
-		}
+            return 0;
+        }
         return p + delta + 1;
     }
     return p + 1;
diff --git a/xmltest.cpp b/xmltest.cpp
index d058707b..f8e71962 100644
--- a/xmltest.cpp
+++ b/xmltest.cpp
@@ -2701,6 +2701,56 @@ int main( int argc, const char ** argv )
 		XMLTest("Test attribute encode with a Hex value", value5, "!"); // hex value in unicode value
 	}
 
+	// ---------- Security: numeric character reference bounds ----------
+	{
+		// Regression: U+10FFFF is the last valid Unicode code point and must
+		// parse correctly. The in-loop overflow guard must not reject it.
+		XMLDocument doc;
+		doc.Parse( "<t v='&#x10FFFF;'/>" );
+		XMLTest( "Numeric ref U+10FFFF: no error", false, doc.Error() );
+		const char* v = doc.FirstChildElement()->Attribute( "v" );
+		// U+10FFFF encodes to the 4-byte UTF-8 sequence F4 8F BF BF.
+		const char expected[] = {
+			static_cast<char>(0xF4), static_cast<char>(0x8F),
+			static_cast<char>(0xBF), static_cast<char>(0xBF), 0
+		};
+		XMLTest( "Numeric ref U+10FFFF: correct UTF-8 output", expected, v );
+	}
+	{
+		// Boundary check: U+110000 is one above the maximum code point.
+		// The in-loop overflow guard must catch this before ucs is written,
+		// leaving the entity as a literal (starting with '&').
+		XMLDocument doc;
+		doc.Parse( "<t v='&#x110000;'/>" );
+		XMLTest( "Numeric ref U+110000: no parse error", false, doc.Error() );
+		const char* v = doc.FirstChildElement()->Attribute( "v" );
+		XMLTest( "Numeric ref U+110000: not resolved (left as literal)", true,
+		         v != nullptr && v[0] == '&' );
+	}
+	{
+		// A hex entity with enough digits to overflow uint32_t must
+		// be rejected by the in-loop guard before the accumulator wraps.
+		// Before the fix, ucs could wrap around and pass the post-loop range
+		// check, producing an attacker-chosen character in the parsed output.
+		// Build "&#x" + 300 'F' digits + ";" -- far beyond what fits in uint32_t.
+		const char prefix[] = "<t v='&#x";
+		const char suffix[] = ";'/>";
+		static const int NDIGITS = 300;
+		char xml[sizeof(prefix) + NDIGITS + sizeof(suffix)];
+		strcpy( xml, prefix );
+		memset( xml + strlen(prefix), 'F', NDIGITS );
+		strcpy( xml + strlen(prefix) + NDIGITS, suffix );
+
+		XMLDocument doc;
+		doc.Parse( xml );
+		XMLTest( "Overflow hex entity: no parse error", false, doc.Error() );
+		const char* v = doc.FirstChildElement()->Attribute( "v" );
+		// GetCharacterRef returns 0 for rejected refs; the caller then copies
+		// the literal '&', so the attribute must start with '&', not a char.
+		XMLTest( "Overflow hex entity: not resolved to a character", true,
+		         v != nullptr && v[0] == '&' );
+	}
+
 	// ---------- XMLPrinter Apos Escaping ------
 	{
 		const char* testText = "text containing a ' character";