diff --git a/src/google/adk/tools/load_artifacts_tool.py b/src/google/adk/tools/load_artifacts_tool.py
index b90219ad57..d84e06f607 100644
--- a/src/google/adk/tools/load_artifacts_tool.py
+++ b/src/google/adk/tools/load_artifacts_tool.py
@@ -111,6 +111,31 @@ def _maybe_base64_to_bytes(data: str) -> bytes | None:
       return None
 
 
+def _decode_xml_reference(match: re.Match[str]) -> str:
+  """Decodes one predefined XML entity or numeric character reference."""
+  reference = match.group(1)
+  if not reference.startswith('#'):
+    return {'amp': '&', 'lt': '<', 'gt': '>', 'quot': '"', 'apos': "'"}[
+        reference
+    ]
+  try:
+    codepoint = (
+        int(reference[2:], 16)
+        if reference.startswith('#x')
+        else int(reference[1:])
+    )
+    # Preserve invalid XML character references instead of aborting extraction.
+    if codepoint in (9, 10, 13) or (
+        0x20 <= codepoint <= 0xD7FF
+        or 0xE000 <= codepoint <= 0xFFFD
+        or 0x10000 <= codepoint <= 0x10FFFF
+    ):
+      return chr(codepoint)
+  except ValueError:
+    pass
+  return match.group(0)
+
+
 def _try_extract_docx_text(data: bytes) -> str | None:
   """Extracts raw text from a DOCX binary."""
   # We use regex instead of standard XML parser to avoid XML bomb vulnerabilities,
@@ -139,7 +164,13 @@ def _try_extract_docx_text(data: bytes) -> str | None:
       for p in re.split(rf'<{p_tag}(?:[^>]*)>', xml_content):
         texts = re.findall(rf'<{t_tag}(?:[^>]*)>([^<]*)</{t_tag}>', p)
         if texts:
-          paragraphs.append(''.join(texts))
+          paragraphs.append(
+              re.sub(
+                  r'&(#x[0-9a-fA-F]+|#[0-9]+|amp|lt|gt|quot|apos);',
+                  _decode_xml_reference,
+                  ''.join(texts),
+              )
+          )
 
       return '\n'.join(paragraphs)
   except (zipfile.BadZipFile, KeyError, struct.error) as e:
diff --git a/tests/unittests/tools/test_load_artifacts_tool.py b/tests/unittests/tools/test_load_artifacts_tool.py
index d0f6e34ad3..a7e8907d47 100644
--- a/tests/unittests/tools/test_load_artifacts_tool.py
+++ b/tests/unittests/tools/test_load_artifacts_tool.py
@@ -171,7 +171,20 @@ async def test_load_artifacts_converts_csv_octet_stream_to_text():
 
 
 @pytest.mark.asyncio
-async def test_load_artifacts_converts_docx_to_text():
+@pytest.mark.parametrize(
+    ('xml_text', 'expected_text'),
+    [
+        ('Hello DOCX', 'Hello DOCX'),
+        ('Research &amp; Development', 'Research & Development'),
+        ('x &lt; 5 &amp;&amp; y &gt; 1', 'x < 5 && y > 1'),
+        ('&quot;Hello&quot; &apos;world&apos;', '"Hello" \'world\''),
+        ('&#20013;&#x6587; &#x1F600;', '中文 😀'),
+        ('Literal &amp;lt; and &amp;#65;', 'Literal &lt; and &#65;'),
+        ('&#x80;', '\x80'),
+        ('&unknown; &#0; &#x110000;', '&unknown; &#0; &#x110000;'),
+    ],
+)
+async def test_load_artifacts_converts_docx_to_text(xml_text, expected_text):
   """DOCX binary payloads are extracted to raw text."""
   artifact_name = 'document.docx'
 
@@ -180,9 +193,9 @@ async def test_load_artifacts_converts_docx_to_text():
   with zipfile.ZipFile(docx_bytes_io, 'w') as zf:
     zf.writestr(
         'word/document.xml',
-        b'<?xml version="1.0" encoding="UTF-8" standalone="yes"?>\n<w:document'
-        b' xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"><w:body><w:p><w:t>Hello'
-        b' DOCX</w:t></w:p></w:body></w:document>',
+        '<?xml version="1.0" encoding="UTF-8" standalone="yes"?>\n<w:document'
+        ' xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main">'
+        f'<w:body><w:p><w:t>{xml_text}</w:t></w:p></w:body></w:document>',
     )
 
   docx_bytes = docx_bytes_io.getvalue()
@@ -216,7 +229,7 @@ async def test_load_artifacts_converts_docx_to_text():
 
   artifact_part = llm_request.contents[-1].parts[1]
   assert artifact_part.inline_data is None
-  assert artifact_part.text == 'Hello DOCX'
+  assert artifact_part.text == expected_text
 
 
 @pytest.mark.asyncio