diff --git a/src/google/adk/evaluation/agent_evaluator.py b/src/google/adk/evaluation/agent_evaluator.py
index f52a367950..bdf88fdb16 100644
--- a/src/google/adk/evaluation/agent_evaluator.py
+++ b/src/google/adk/evaluation/agent_evaluator.py
@@ -78,7 +78,7 @@
 
 
 def load_json(file_path: str) -> Union[Dict, List]:
-  with open(file_path, "r") as f:
+  with open(file_path, "r", encoding="utf-8") as f:
     return json.load(f)
 
 
@@ -266,7 +266,7 @@ def migrate_eval_data_to_new_schema(
         old_eval_data_file, eval_config, initial_session
     )
 
-    with open(new_eval_data_file, "w") as f:
+    with open(new_eval_data_file, "w", encoding="utf-8") as f:
       f.write(eval_set.model_dump_json(indent=2))
 
   @staticmethod
@@ -323,7 +323,7 @@ def _get_eval_set_from_old_format(
   def _get_initial_session(initial_session_file: Optional[str] = None):
     initial_session = {}
     if initial_session_file:
-      with open(initial_session_file, "r") as f:
+      with open(initial_session_file, "r", encoding="utf-8") as f:
         initial_session = json.loads(f.read())
     return initial_session
 
diff --git a/src/google/adk/evaluation/evaluation_generator.py b/src/google/adk/evaluation/evaluation_generator.py
index e7f88634d1..165ea7f09f 100644
--- a/src/google/adk/evaluation/evaluation_generator.py
+++ b/src/google/adk/evaluation/evaluation_generator.py
@@ -315,7 +315,7 @@ def generate_responses_from_session(session_path, eval_dataset):
     """
     results = []
 
-    with open(session_path, "r") as f:
+    with open(session_path, "r", encoding="utf-8") as f:
       session_data = Session.model_validate_json(f.read())
       logger.info("Loaded session %s", session_path)
 
diff --git a/tests/unittests/evaluation/test_agent_evaluator.py b/tests/unittests/evaluation/test_agent_evaluator.py
new file mode 100644
index 0000000000..e9ff7aba1a
--- /dev/null
+++ b/tests/unittests/evaluation/test_agent_evaluator.py
@@ -0,0 +1,113 @@
+# Copyright 2026 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+import builtins
+import json
+
+from google.adk.evaluation import agent_evaluator as agent_evaluator_module
+from google.adk.evaluation.agent_evaluator import AgentEvaluator
+from google.adk.evaluation.agent_evaluator import load_json
+
+# Non-ASCII payload (emoji, CJK, accented characters) that only survives a
+# UTF-8 round-trip. Written to eval-data files by the eval subsystem's Pydantic
+# `model_dump_json` (which does not escape non-ASCII).
+_NON_ASCII_TEXT = "\U0001f600 \u4f60\u597d caf\u00e9"
+
+_real_open = builtins.open
+
+
+def _non_utf8_default_open(file, mode="r", *args, **kwargs):
+  """Emulates a platform whose default text encoding is not UTF-8.
+
+  On such platforms (for example Windows, where the default is cp1252),
+  `open()` calls that omit `encoding=` inherit that non-UTF-8 default. This
+  wrapper reproduces that behaviour on any platform by falling back to ASCII
+  when a text-mode open does not specify an encoding, so a missing
+  `encoding="utf-8"` argument raises instead of silently depending on the
+  host locale.
+  """
+  if "b" not in mode and "encoding" not in kwargs:
+    kwargs["encoding"] = "ascii"
+  return _real_open(file, mode, *args, **kwargs)
+
+
+def test_load_json_reads_non_ascii_with_non_utf8_default(tmp_path, mocker):
+  """`load_json` must decode eval data as UTF-8 regardless of platform locale."""
+  file_path = tmp_path / "eval.json"
+  file_path.write_text(
+      json.dumps([{"query": _NON_ASCII_TEXT}], ensure_ascii=False),
+      encoding="utf-8",
+  )
+
+  mocker.patch.object(
+      agent_evaluator_module, "open", _non_utf8_default_open, create=True
+  )
+
+  assert load_json(str(file_path)) == [{"query": _NON_ASCII_TEXT}]
+
+
+def test_get_initial_session_reads_non_ascii_with_non_utf8_default(
+    tmp_path, mocker
+):
+  """`_get_initial_session` must decode the session file as UTF-8."""
+  session_file = tmp_path / "initial_session.json"
+  session_file.write_text(
+      json.dumps({"state": {"city": _NON_ASCII_TEXT}}, ensure_ascii=False),
+      encoding="utf-8",
+  )
+
+  mocker.patch.object(
+      agent_evaluator_module, "open", _non_utf8_default_open, create=True
+  )
+
+  initial_session = AgentEvaluator._get_initial_session(str(session_file))
+
+  assert initial_session == {"state": {"city": _NON_ASCII_TEXT}}
+
+
+def test_migrate_eval_data_round_trips_non_ascii_with_non_utf8_default(
+    tmp_path, mocker
+):
+  """Migration must read the old file and write the new file as UTF-8.
+
+  This exercises both the read (`load_json`) and the write
+  (`model_dump_json`) of eval data, which must stay UTF-8 consistent so that
+  datasets containing non-ASCII characters survive migration on any platform.
+  """
+  old_eval_data_file = tmp_path / "old_format.test.json"
+  old_eval_data_file.write_text(
+      json.dumps(
+          [{
+              "query": _NON_ASCII_TEXT,
+              "reference": _NON_ASCII_TEXT,
+              "expected_tool_use": [],
+          }],
+          ensure_ascii=False,
+      ),
+      encoding="utf-8",
+  )
+  new_eval_data_file = tmp_path / "new_format.json"
+
+  mocker.patch.object(
+      agent_evaluator_module, "open", _non_utf8_default_open, create=True
+  )
+
+  AgentEvaluator.migrate_eval_data_to_new_schema(
+      str(old_eval_data_file), str(new_eval_data_file)
+  )
+
+  migrated = json.loads(new_eval_data_file.read_text(encoding="utf-8"))
+  assert _NON_ASCII_TEXT in json.dumps(migrated, ensure_ascii=False)
diff --git a/tests/unittests/evaluation/test_evaluation_generator.py b/tests/unittests/evaluation/test_evaluation_generator.py
index ea6364cad3..4e03016833 100644
--- a/tests/unittests/evaluation/test_evaluation_generator.py
+++ b/tests/unittests/evaluation/test_evaluation_generator.py
@@ -15,7 +15,9 @@
 from __future__ import annotations
 
 import asyncio
+import builtins
 
+from google.adk.evaluation import evaluation_generator as evaluation_generator_module
 from google.adk.evaluation.app_details import AgentDetails
 from google.adk.evaluation.app_details import AppDetails
 from google.adk.evaluation.conversation_scenarios import ConversationScenario
@@ -33,6 +35,7 @@
 from google.adk.events.event import Event
 from google.adk.events.event_actions import EventActions
 from google.adk.models.llm_request import LlmRequest
+from google.adk.sessions.session import Session
 from google.genai import types
 import pytest
 
@@ -966,3 +969,65 @@ def test_convert_events_preserves_tool_calls_when_skip_summarization():
   assert len(tool_calls) == 1
   assert tool_calls[0].name == "execute_sql"
   assert tool_calls[0].args == {"project_id": "my-proj", "query": "SELECT 1"}
+
+
+_real_open = builtins.open
+
+
+def _non_utf8_default_open(file, mode="r", *args, **kwargs):
+  """Emulates a platform whose default text encoding is not UTF-8.
+
+  Falls back to ASCII when a text-mode open does not specify an encoding, so a
+  missing `encoding="utf-8"` argument raises instead of silently depending on
+  the host locale (for example cp1252 on Windows).
+  """
+  if "b" not in mode and "encoding" not in kwargs:
+    kwargs["encoding"] = "ascii"
+  return _real_open(file, mode, *args, **kwargs)
+
+
+def test_generate_responses_from_session_reads_non_ascii_with_non_utf8_default(
+    tmp_path, mocker
+):
+  """The session file must be read as UTF-8 regardless of platform locale.
+
+  Session files serialized via `model_dump_json` contain raw (unescaped)
+  non-ASCII characters, so reading them without an explicit UTF-8 encoding
+  fails on platforms whose default encoding is not UTF-8.
+  """
+  non_ascii_text = "\U0001f600 \u4f60\u597d caf\u00e9"
+  session = Session(
+      id="s1",
+      app_name="app",
+      user_id="u1",
+      events=[
+          Event(
+              author="user",
+              invocation_id="inv1",
+              content=types.Content(
+                  role="user", parts=[types.Part(text=non_ascii_text)]
+              ),
+          ),
+          Event(
+              author="agent",
+              invocation_id="inv1",
+              content=types.Content(
+                  role="model",
+                  parts=[types.Part(text="response " + non_ascii_text)],
+              ),
+          ),
+      ],
+  )
+  session_path = tmp_path / "session.json"
+  session_path.write_text(session.model_dump_json(), encoding="utf-8")
+
+  mocker.patch.object(
+      evaluation_generator_module, "open", _non_utf8_default_open, create=True
+  )
+
+  results = EvaluationGenerator.generate_responses_from_session(
+      str(session_path), [[{"query": non_ascii_text}]]
+  )
+
+  assert results[0][0]["query"] == non_ascii_text
+  assert results[0][0]["response"] == "response " + non_ascii_text