diff --git a/src/transformers/models/bart/tokenization_bart.py b/src/transformers/models/bart/tokenization_bart.py
index f674afe1a412..91971a29191c 100644
--- a/src/transformers/models/bart/tokenization_bart.py
+++ b/src/transformers/models/bart/tokenization_bart.py
@@ -18,10 +18,9 @@
 from functools import lru_cache
 from typing import Optional
 
-import regex as re
-
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/bertweet/tokenization_bertweet.py b/src/transformers/models/bertweet/tokenization_bertweet.py
index 3ce1a3182bf9..94e2a7240d1c 100644
--- a/src/transformers/models/bertweet/tokenization_bertweet.py
+++ b/src/transformers/models/bertweet/tokenization_bertweet.py
@@ -21,10 +21,9 @@
 from shutil import copyfile
 from typing import Optional
 
-import regex
-
 from ...tokenization_utils import PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/blenderbot/tokenization_blenderbot.py b/src/transformers/models/blenderbot/tokenization_blenderbot.py
index 76719fa25494..4cdcdde8b016 100644
--- a/src/transformers/models/blenderbot/tokenization_blenderbot.py
+++ b/src/transformers/models/blenderbot/tokenization_blenderbot.py
@@ -19,10 +19,9 @@
 from functools import lru_cache
 from typing import Optional
 
-import regex as re
-
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/blenderbot_small/tokenization_blenderbot_small.py b/src/transformers/models/blenderbot_small/tokenization_blenderbot_small.py
index adb54025ce23..91fa8ffb1dda 100644
--- a/src/transformers/models/blenderbot_small/tokenization_blenderbot_small.py
+++ b/src/transformers/models/blenderbot_small/tokenization_blenderbot_small.py
@@ -18,10 +18,9 @@
 import os
 from typing import Optional
 
-import regex as re
-
 from ...tokenization_utils import PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/clip/tokenization_clip.py b/src/transformers/models/clip/tokenization_clip.py
index 625d26dc6960..7b1a52224423 100644
--- a/src/transformers/models/clip/tokenization_clip.py
+++ b/src/transformers/models/clip/tokenization_clip.py
@@ -20,10 +20,9 @@
 from functools import lru_cache
 from typing import Optional
 
-import regex as re
-
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer, _is_control, _is_punctuation, _is_whitespace
 from ...utils import logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/clvp/tokenization_clvp.py b/src/transformers/models/clvp/tokenization_clvp.py
index 4b0b285561c5..03682e693cbd 100644
--- a/src/transformers/models/clvp/tokenization_clvp.py
+++ b/src/transformers/models/clvp/tokenization_clvp.py
@@ -19,10 +19,9 @@
 from functools import lru_cache
 from typing import Optional
 
-import regex as re
-
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex as re
 from .number_normalizer import EnglishNormalizer
 
 
diff --git a/src/transformers/models/codegen/tokenization_codegen.py b/src/transformers/models/codegen/tokenization_codegen.py
index 4d08c6acd5bb..5906ae77e9f2 100644
--- a/src/transformers/models/codegen/tokenization_codegen.py
+++ b/src/transformers/models/codegen/tokenization_codegen.py
@@ -20,9 +20,9 @@
 from typing import TYPE_CHECKING, Optional, Union
 
 import numpy as np
-import regex as re
 
 from ...utils import logging, to_py_obj
+from ...utils.safe import regex as re
 
 
 if TYPE_CHECKING:
diff --git a/src/transformers/models/ctrl/tokenization_ctrl.py b/src/transformers/models/ctrl/tokenization_ctrl.py
index 5b7935e6404d..0a2c0eff46c7 100644
--- a/src/transformers/models/ctrl/tokenization_ctrl.py
+++ b/src/transformers/models/ctrl/tokenization_ctrl.py
@@ -18,10 +18,9 @@
 import os
 from typing import Optional
 
-import regex as re
-
 from ...tokenization_utils import PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/deberta/tokenization_deberta.py b/src/transformers/models/deberta/tokenization_deberta.py
index 74e958c8030b..858c35bd0870 100644
--- a/src/transformers/models/deberta/tokenization_deberta.py
+++ b/src/transformers/models/deberta/tokenization_deberta.py
@@ -18,10 +18,9 @@
 import os
 from typing import Optional
 
-import regex as re
-
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/deepseek_vl/convert_deepseek_vl_weights_to_hf.py b/src/transformers/models/deepseek_vl/convert_deepseek_vl_weights_to_hf.py
index 3e9b6a37fe09..4c4f2a8c6df6 100644
--- a/src/transformers/models/deepseek_vl/convert_deepseek_vl_weights_to_hf.py
+++ b/src/transformers/models/deepseek_vl/convert_deepseek_vl_weights_to_hf.py
@@ -18,7 +18,6 @@
 import os
 from typing import Optional
 
-import regex as re
 import torch
 from accelerate import init_empty_weights
 from huggingface_hub import snapshot_download
@@ -33,6 +32,7 @@
     DeepseekVLProcessor,
 )
 from transformers.image_utils import IMAGENET_STANDARD_MEAN, IMAGENET_STANDARD_STD
+from transformers.utils.safe import regex as re
 
 
 # fmt: off
diff --git a/src/transformers/models/deepseek_vl_hybrid/convert_deepseek_vl_hybrid_weights_to_hf.py b/src/transformers/models/deepseek_vl_hybrid/convert_deepseek_vl_hybrid_weights_to_hf.py
index 9f377a53c8f3..108a66fae9d4 100644
--- a/src/transformers/models/deepseek_vl_hybrid/convert_deepseek_vl_hybrid_weights_to_hf.py
+++ b/src/transformers/models/deepseek_vl_hybrid/convert_deepseek_vl_hybrid_weights_to_hf.py
@@ -18,7 +18,6 @@
 import os
 from typing import Optional
 
-import regex as re
 import torch
 from accelerate import init_empty_weights
 from huggingface_hub import snapshot_download
@@ -39,6 +38,7 @@
     OPENAI_CLIP_STD,
     PILImageResampling,
 )
+from transformers.utils.safe import regex as re
 
 
 # fmt: off
diff --git a/src/transformers/models/deprecated/jukebox/tokenization_jukebox.py b/src/transformers/models/deprecated/jukebox/tokenization_jukebox.py
index b9948c5c354e..bfaba58a4bae 100644
--- a/src/transformers/models/deprecated/jukebox/tokenization_jukebox.py
+++ b/src/transformers/models/deprecated/jukebox/tokenization_jukebox.py
@@ -22,12 +22,12 @@
 from typing import Any, Optional, Union
 
 import numpy as np
-import regex
 
 from ....tokenization_utils import AddedToken, PreTrainedTokenizer
 from ....tokenization_utils_base import BatchEncoding
 from ....utils import TensorType, is_torch_available, logging
 from ....utils.generic import is_numpy_array
+from ...utils.safe import regex
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/deprecated/tapex/tokenization_tapex.py b/src/transformers/models/deprecated/tapex/tokenization_tapex.py
index fa74d8aa3b55..53f43a36be18 100644
--- a/src/transformers/models/deprecated/tapex/tokenization_tapex.py
+++ b/src/transformers/models/deprecated/tapex/tokenization_tapex.py
@@ -20,12 +20,11 @@
 from functools import lru_cache
 from typing import Optional, Union
 
-import regex as re
-
 from ....file_utils import ExplicitEnum, PaddingStrategy, TensorType, add_end_docstrings, is_pandas_available
 from ....tokenization_utils import AddedToken, PreTrainedTokenizer
 from ....tokenization_utils_base import ENCODE_KWARGS_DOCSTRING, BatchEncoding, TextInput, TruncationStrategy
 from ....utils import logging
+from ...utils.safe import regex as re
 
 
 if is_pandas_available():
diff --git a/src/transformers/models/depth_pro/convert_depth_pro_weights_to_hf.py b/src/transformers/models/depth_pro/convert_depth_pro_weights_to_hf.py
index 655bbdc0230f..8ec66edb804b 100644
--- a/src/transformers/models/depth_pro/convert_depth_pro_weights_to_hf.py
+++ b/src/transformers/models/depth_pro/convert_depth_pro_weights_to_hf.py
@@ -17,7 +17,6 @@
 import os
 from typing import Optional
 
-import regex as re
 import torch
 from huggingface_hub import hf_hub_download
 
@@ -26,6 +25,7 @@
     DepthProForDepthEstimation,
     DepthProImageProcessorFast,
 )
+from transformers.utils.safe import regex as re
 
 
 # fmt: off
diff --git a/src/transformers/models/fastspeech2_conformer/tokenization_fastspeech2_conformer.py b/src/transformers/models/fastspeech2_conformer/tokenization_fastspeech2_conformer.py
index 004a1c36f59c..9255227f997c 100644
--- a/src/transformers/models/fastspeech2_conformer/tokenization_fastspeech2_conformer.py
+++ b/src/transformers/models/fastspeech2_conformer/tokenization_fastspeech2_conformer.py
@@ -18,10 +18,9 @@
 import os
 from typing import Optional
 
-import regex
-
 from ...tokenization_utils import PreTrainedTokenizer
 from ...utils import logging, requires_backends
+from ...utils.safe import regex
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/got_ocr2/convert_got_ocr2_weights_to_hf.py b/src/transformers/models/got_ocr2/convert_got_ocr2_weights_to_hf.py
index 39496fe043ed..a332031ea6e0 100644
--- a/src/transformers/models/got_ocr2/convert_got_ocr2_weights_to_hf.py
+++ b/src/transformers/models/got_ocr2/convert_got_ocr2_weights_to_hf.py
@@ -18,7 +18,6 @@
 import os
 from typing import Optional
 
-import regex as re
 import torch
 from huggingface_hub import snapshot_download
 from safetensors import safe_open
@@ -33,6 +32,7 @@
 )
 from transformers.convert_slow_tokenizer import TikTokenConverter
 from transformers.tokenization_utils import AddedToken
+from transformers.utils.safe import regex as re
 
 
 if is_vision_available():
diff --git a/src/transformers/models/gpt2/tokenization_gpt2.py b/src/transformers/models/gpt2/tokenization_gpt2.py
index 608164ef2d83..8387f71d4e4b 100644
--- a/src/transformers/models/gpt2/tokenization_gpt2.py
+++ b/src/transformers/models/gpt2/tokenization_gpt2.py
@@ -19,10 +19,9 @@
 from functools import lru_cache
 from typing import Optional
 
-import regex as re
-
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/gpt_oss/convert_gpt_oss_weights_to_hf.py b/src/transformers/models/gpt_oss/convert_gpt_oss_weights_to_hf.py
index 736a95247dfb..63d1ccc32500 100644
--- a/src/transformers/models/gpt_oss/convert_gpt_oss_weights_to_hf.py
+++ b/src/transformers/models/gpt_oss/convert_gpt_oss_weights_to_hf.py
@@ -19,7 +19,6 @@
 from pathlib import Path
 from typing import Optional
 
-import regex as re
 import tiktoken
 import torch
 from safetensors.torch import load_file as safe_load
@@ -31,6 +30,7 @@
     PreTrainedTokenizerFast,
 )
 from transformers.convert_slow_tokenizer import TikTokenConverter
+from transformers.utils.safe import regex as re
 
 
 # fmt: off
diff --git a/src/transformers/models/layoutlmv3/tokenization_layoutlmv3.py b/src/transformers/models/layoutlmv3/tokenization_layoutlmv3.py
index 7877c5b4668d..f847757e3952 100644
--- a/src/transformers/models/layoutlmv3/tokenization_layoutlmv3.py
+++ b/src/transformers/models/layoutlmv3/tokenization_layoutlmv3.py
@@ -19,8 +19,6 @@
 from functools import lru_cache
 from typing import Optional, Union
 
-import regex as re
-
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...tokenization_utils_base import (
     BatchEncoding,
@@ -31,6 +29,7 @@
     TruncationStrategy,
 )
 from ...utils import PaddingStrategy, TensorType, add_end_docstrings, logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/led/tokenization_led.py b/src/transformers/models/led/tokenization_led.py
index d110ac30d969..8982a526c2bc 100644
--- a/src/transformers/models/led/tokenization_led.py
+++ b/src/transformers/models/led/tokenization_led.py
@@ -19,11 +19,10 @@
 from functools import lru_cache
 from typing import Optional, Union
 
-import regex as re
-
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...tokenization_utils_base import BatchEncoding, EncodedInput
 from ...utils import PaddingStrategy, logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/longformer/tokenization_longformer.py b/src/transformers/models/longformer/tokenization_longformer.py
index 104bdd7a9b99..63b81e2ad575 100644
--- a/src/transformers/models/longformer/tokenization_longformer.py
+++ b/src/transformers/models/longformer/tokenization_longformer.py
@@ -18,10 +18,9 @@
 from functools import lru_cache
 from typing import Optional
 
-import regex as re
-
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/luke/tokenization_luke.py b/src/transformers/models/luke/tokenization_luke.py
index 4bb19bb5ee73..fcfea6fb1c7e 100644
--- a/src/transformers/models/luke/tokenization_luke.py
+++ b/src/transformers/models/luke/tokenization_luke.py
@@ -22,7 +22,6 @@
 from typing import Optional, Union
 
 import numpy as np
-import regex as re
 
 from ...tokenization_utils import PreTrainedTokenizer
 from ...tokenization_utils_base import (
@@ -38,6 +37,7 @@
     to_py_obj,
 )
 from ...utils import add_end_docstrings, is_torch_tensor, logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/markuplm/tokenization_markuplm.py b/src/transformers/models/markuplm/tokenization_markuplm.py
index 0a6f7c3bd6a0..bec10fc86660 100644
--- a/src/transformers/models/markuplm/tokenization_markuplm.py
+++ b/src/transformers/models/markuplm/tokenization_markuplm.py
@@ -19,8 +19,6 @@
 from functools import lru_cache
 from typing import Optional, Union
 
-import regex as re
-
 from ...file_utils import PaddingStrategy, TensorType, add_end_docstrings
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...tokenization_utils_base import (
@@ -33,6 +31,7 @@
     TruncationStrategy,
 )
 from ...utils import logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/mllama/convert_mllama_weights_to_hf.py b/src/transformers/models/mllama/convert_mllama_weights_to_hf.py
index c773d0514f81..72c73b878446 100644
--- a/src/transformers/models/mllama/convert_mllama_weights_to_hf.py
+++ b/src/transformers/models/mllama/convert_mllama_weights_to_hf.py
@@ -19,7 +19,6 @@
 import os
 from typing import Optional
 
-import regex as re
 import torch
 import torch.nn.functional as F
 
@@ -33,6 +32,7 @@
 from transformers.convert_slow_tokenizer import TikTokenConverter
 from transformers.models.mllama.configuration_mllama import MllamaTextConfig, MllamaVisionConfig
 from transformers.models.mllama.image_processing_mllama import get_all_supported_aspect_ratios
+from transformers.utils.safe import regex as re
 
 
 # fmt: off
diff --git a/src/transformers/models/mvp/tokenization_mvp.py b/src/transformers/models/mvp/tokenization_mvp.py
index f6039df2dc02..4b6397b9f44c 100644
--- a/src/transformers/models/mvp/tokenization_mvp.py
+++ b/src/transformers/models/mvp/tokenization_mvp.py
@@ -18,10 +18,9 @@
 from functools import lru_cache
 from typing import Optional
 
-import regex as re
-
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/pixtral/convert_pixtral_weights_to_hf.py b/src/transformers/models/pixtral/convert_pixtral_weights_to_hf.py
index 373aa6cb6e45..d8cc4c477f01 100644
--- a/src/transformers/models/pixtral/convert_pixtral_weights_to_hf.py
+++ b/src/transformers/models/pixtral/convert_pixtral_weights_to_hf.py
@@ -15,7 +15,6 @@
 import json
 import os
 
-import regex as re
 import torch
 from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
 from safetensors.torch import load_file as safe_load_file
@@ -28,6 +27,7 @@
     PixtralProcessor,
     PixtralVisionConfig,
 )
+from transformers.utils.safe import regex as re
 
 
 """
diff --git a/src/transformers/models/qwen2/tokenization_qwen2.py b/src/transformers/models/qwen2/tokenization_qwen2.py
index be121adb5442..1d935f856d51 100644
--- a/src/transformers/models/qwen2/tokenization_qwen2.py
+++ b/src/transformers/models/qwen2/tokenization_qwen2.py
@@ -20,10 +20,9 @@
 from functools import lru_cache
 from typing import Optional
 
-import regex as re
-
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/recurrent_gemma/convert_recurrent_gemma_to_hf.py b/src/transformers/models/recurrent_gemma/convert_recurrent_gemma_to_hf.py
index 4871311e114b..e056c6c740a0 100644
--- a/src/transformers/models/recurrent_gemma/convert_recurrent_gemma_to_hf.py
+++ b/src/transformers/models/recurrent_gemma/convert_recurrent_gemma_to_hf.py
@@ -19,6 +19,7 @@
 from accelerate import init_empty_weights
 
 from transformers import GemmaTokenizer, RecurrentGemmaConfig, RecurrentGemmaForCausalLM
+from transformers.utils.safe import regex as re
 
 
 try:
@@ -30,8 +31,6 @@
     )
     GemmaTokenizerFast = None
 
-import regex as re
-
 
 """
 Sample usage:
@@ -63,6 +62,7 @@
     num_hidden_layers=26,
 )
 
+
 gemma_7b_config = RecurrentGemmaConfig()
 
 CONFIG_MAPPING = {"2B": gemma_2b_config, "7B": gemma_7b_config}
diff --git a/src/transformers/models/roberta/tokenization_roberta.py b/src/transformers/models/roberta/tokenization_roberta.py
index 67cdcbbf488a..c356fc28b0f7 100644
--- a/src/transformers/models/roberta/tokenization_roberta.py
+++ b/src/transformers/models/roberta/tokenization_roberta.py
@@ -19,10 +19,9 @@
 from functools import lru_cache
 from typing import Optional
 
-import regex as re
-
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex as re
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/transformers/models/whisper/english_normalizer.py b/src/transformers/models/whisper/english_normalizer.py
index 265ea04b5334..0f9594db7342 100644
--- a/src/transformers/models/whisper/english_normalizer.py
+++ b/src/transformers/models/whisper/english_normalizer.py
@@ -20,7 +20,7 @@
 from re import Match
 from typing import Optional, Union
 
-import regex
+from ...utils.safe import regex
 
 
 # non-ASCII letters that are not separated by "NFKD" normalization
diff --git a/src/transformers/models/whisper/tokenization_whisper.py b/src/transformers/models/whisper/tokenization_whisper.py
index 34d9a8965be8..3984aff53096 100644
--- a/src/transformers/models/whisper/tokenization_whisper.py
+++ b/src/transformers/models/whisper/tokenization_whisper.py
@@ -21,10 +21,10 @@
 from typing import Optional, Union
 
 import numpy as np
-import regex as re
 
 from ...tokenization_utils import AddedToken, PreTrainedTokenizer
 from ...utils import logging
+from ...utils.safe import regex as re
 from .english_normalizer import BasicTextNormalizer, EnglishTextNormalizer
 
 
diff --git a/src/transformers/utils/auto_docstring.py b/src/transformers/utils/auto_docstring.py
index 135f20bf4cf9..5f7e4d10f828 100644
--- a/src/transformers/utils/auto_docstring.py
+++ b/src/transformers/utils/auto_docstring.py
@@ -19,8 +19,6 @@
 from pathlib import Path
 from typing import Optional, Union, get_args
 
-import regex as re
-
 from .doc import (
     MODELS_TO_PIPELINE,
     PIPELINE_TASKS_TO_SAMPLE_DOCSTRINGS,
@@ -28,6 +26,7 @@
     _prepare_output_docstrings,
 )
 from .generic import ModelOutput
+from .safe import regex as re
 
 
 PATH_TO_TRANSFORMERS = Path("src").resolve() / "transformers"
diff --git a/src/transformers/utils/safe.py b/src/transformers/utils/safe.py
new file mode 100644
index 000000000000..661e494f5581
--- /dev/null
+++ b/src/transformers/utils/safe.py
@@ -0,0 +1,137 @@
+# Copyright 2025 ModelCloud.ai team and The HuggingFace Inc. team.
+##
+# 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.
+
+"""Thread-safe utilities and module wrappers usable across Transformers."""
+
+from __future__ import annotations
+
+import threading
+from functools import wraps
+from types import ModuleType
+
+import regex as _regex
+
+
+__all__ = ["ThreadSafe", "SafeRegex", "regex"]
+
+
+class ThreadSafe(ModuleType):
+    """Generic proxy that exposes a module through a shared lock."""
+
+    def __init__(self, module: ModuleType):
+        super().__init__(module.__name__)
+        # `_hf_safe_` prefix is used to avoid colliding with the wrapped object namespace.
+        self._hf_safe_module = module
+        # Callable execution lock (re-entrant so wrapped code can re-enter safely)
+        self._hf_safe_lock = threading.RLock()
+        # Cache dict lock
+        self._hf_safe_callable_cache_lock = threading.Lock()
+        self._hf_safe_callable_cache: dict[str, object] = {}
+        # Retain module metadata so introspection tools relying on attributes
+        # like __doc__, __spec__, etc, can see the original values.
+        metadata = {"__doc__": module.__doc__}
+        for attr in ("__package__", "__file__", "__spec__"):
+            if hasattr(module, attr):
+                metadata[attr] = getattr(module, attr)
+        self.__dict__.update(metadata)
+
+    def __getattr__(self, name: str):
+        attr = getattr(self._hf_safe_module, name)
+        if callable(attr):
+            with self._hf_safe_callable_cache_lock:
+                cached = self._hf_safe_callable_cache.get(name)
+                if cached is not None and getattr(cached, "__wrapped__", None) is attr:
+                    return cached
+
+                @wraps(attr)
+                def _hf_safe_locked(*args, **kwargs):
+                    with self._hf_safe_lock:
+                        return attr(*args, **kwargs)
+
+                _hf_safe_locked.__wrapped__ = attr
+                self._hf_safe_callable_cache[name] = _hf_safe_locked
+                return _hf_safe_locked
+        return attr
+
+    def __dir__(self):
+        return sorted(set(super().__dir__()) | set(dir(self._hf_safe_module)))
+
+
+class _ThreadSafeProxy:
+    """Lightweight proxy that serializes access to an object with a shared lock."""
+
+    def __init__(self, value, lock):
+        # `_hf_safe_` prefix is used to avoid colliding with the wrapped object namespace.
+        object.__setattr__(self, "_hf_safe_value", value)
+        object.__setattr__(self, "_hf_safe_lock", lock)
+        object.__setattr__(self, "_hf_safe_cache_lock", threading.Lock())
+        object.__setattr__(self, "_hf_safe_callable_cache", {})
+        object.__setattr__(self, "__wrapped__", value)
+
+    def __getattr__(self, name: str):
+        attr = getattr(self._hf_safe_value, name)
+        if callable(attr):
+            with self._hf_safe_cache_lock:
+                cached = self._hf_safe_callable_cache.get(name)
+                if cached is not None and getattr(cached, "__wrapped__", None) is attr:
+                    return cached
+
+                @wraps(attr)
+                def _hf_safe_locked(*args, **kwargs):
+                    with self._hf_safe_lock:
+                        return attr(*args, **kwargs)
+
+                _hf_safe_locked.__wrapped__ = attr
+                self._hf_safe_callable_cache[name] = _hf_safe_locked
+                return _hf_safe_locked
+        return attr
+
+    def __setattr__(self, name, value):
+        with self._hf_safe_lock:
+            setattr(self._hf_safe_value, name, value)
+
+    def __delattr__(self, name):
+        with self._hf_safe_lock:
+            delattr(self._hf_safe_value, name)
+
+    def __dir__(self):
+        with self._hf_safe_lock:
+            return dir(self._hf_safe_value)
+
+    def __repr__(self):
+        with self._hf_safe_lock:
+            return repr(self._hf_safe_value)
+
+    def __call__(self, *args, **kwargs):
+        with self._hf_safe_lock:
+            return self._hf_safe_value(*args, **kwargs)
+
+
+class SafeRegex(ThreadSafe):
+    """Proxy module that exposes ``regex`` through a shared lock."""
+
+    # We must proxy the shared regex lock to any objects returned here since
+    # compiled patterns expose methods (e.g. pattern.match) that must also be
+    # serialized.
+
+    def compile(self, *args, **kwargs):
+        pattern = self._hf_safe_module.compile(*args, **kwargs)
+        return _ThreadSafeProxy(pattern, self._hf_safe_lock)
+
+    def Regex(self, *args, **kwargs):
+        pattern = self._hf_safe_module.Regex(*args, **kwargs)
+        return _ThreadSafeProxy(pattern, self._hf_safe_lock)
+
+
+regex = SafeRegex(_regex)
diff --git a/tests/utils/test_safe.py b/tests/utils/test_safe.py
new file mode 100644
index 000000000000..a4e83b7fa63b
--- /dev/null
+++ b/tests/utils/test_safe.py
@@ -0,0 +1,180 @@
+# Copyright 2025 ModelCloud.ai team and The HuggingFace Inc. team.
+#
+# 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.
+
+import builtins
+import sys
+import threading
+import types
+
+import pytest
+
+from transformers.utils.safe import ThreadSafe, regex
+
+
+pytestmark = pytest.mark.skipif(
+    not hasattr(sys, "_is_gil_enabled") or sys._is_gil_enabled(),
+    reason="Safe regex test only runs when the GIL is disabled",
+)
+
+
+def _exercise_pattern(pattern_factory):
+    expected = {"group0": "test123", "prefix": "test", "number": "123"}
+
+    num_threads = 32
+    errors = []
+    errors_lock = threading.Lock()
+    ready_group = threading.Barrier(num_threads + 1)
+    run_event = threading.Event()
+
+    def worker():
+        try:
+            ready_group.wait()
+            run_event.wait()
+            for _ in range(100):
+                pattern = pattern_factory()
+                match = pattern.match("test123")
+                assert match is not None
+                assert match.group(0) == expected["group0"]
+                assert match.group("prefix") == expected["prefix"]
+                assert match.group("number") == expected["number"]
+        except Exception as exc:
+            with errors_lock:
+                errors.append(exc)
+
+    threads = [threading.Thread(target=worker) for _ in range(num_threads)]
+    for thread in threads:
+        thread.start()
+
+    ready_group.wait()  # Ensure every thread is ready before triggering execution
+    run_event.set()
+
+    for thread in threads:
+        thread.join()
+
+    return errors
+
+
+def test_regex_thread_safety_under_gil0():
+    def factory():
+        return regex.compile(r"(?P<prefix>test)(?P<number>\d+)")
+
+    errors = _exercise_pattern(factory)
+    assert not errors
+
+
+def test_regex_thread_safety_shared_pattern_under_gil0():
+    shared_pattern = regex.compile(r"(?P<prefix>test)(?P<number>\d+)")
+
+    def factory():
+        return shared_pattern
+
+    errors = _exercise_pattern(factory)
+    assert not errors
+
+
+def test_regex_thread_safety_direct_match_under_gil0():
+    def factory():
+        class _DirectMatcher:
+            def match(self, text):
+                return regex.match(r"(?P<prefix>test)(?P<number>\d+)", text)
+
+        return _DirectMatcher()
+
+    errors = _exercise_pattern(factory)
+    assert not errors
+
+
+def test_regex_threadsafe_allows_reentrant_calls_under_gil0():
+    pattern = regex.compile(r"(?P<prefix>test)(?P<number>\d+)")
+    completed = threading.Event()
+
+    def target():
+        def replace(match):
+            inner = regex.match(r"(?P<prefix>test)(?P<number>\d+)", match.group(0))
+            return inner.group(0)
+
+        result = pattern.sub(replace, "test123")
+        assert result == "test123"
+        completed.set()
+
+    worker = threading.Thread(target=target)
+    worker.start()
+    worker.join(timeout=2)
+
+    assert completed.is_set(), "Re-entrant call should not deadlock"
+
+
+def test_threadsafe_callable_cache_is_serialized_under_gil0():
+    module = types.ModuleType("_threadsafe_test_module")
+
+    counter_lock = threading.Lock()
+    call_counter = {"count": 0}
+
+    def increment(value: int):
+        with counter_lock:
+            call_counter["count"] += 1
+        return value + 1
+
+    module.increment = increment
+    thread_safe_module = ThreadSafe(module)
+
+    num_threads = 32
+    errors = []
+    errors_lock = threading.Lock()
+    ready_group = threading.Barrier(num_threads + 1)
+    run_event = threading.Event()
+
+    def worker():
+        try:
+            ready_group.wait()
+            run_event.wait()
+            for _ in range(256):
+                func = thread_safe_module.increment
+                assert func(1) == 2
+        except Exception as exc:
+            with errors_lock:
+                errors.append(exc)
+
+    threads = [threading.Thread(target=worker) for _ in range(num_threads)]
+    for thread in threads:
+        thread.start()
+
+    ready_group.wait()  # Synchronize thread start to mimic high contention scenarios
+    run_event.set()
+
+    for thread in threads:
+        thread.join()
+
+    assert not errors
+    # Each invocation should be recorded; the exact count confirms every worker executed.
+    assert call_counter["count"] == num_threads * 256
+
+
+def test_threadsafe_copies_existing_module_metadata_under_gil0():
+    thread_safe_regex = ThreadSafe(regex)
+
+    for attr in ("__package__", "__file__", "__spec__"):
+        if hasattr(regex, attr):
+            assert hasattr(thread_safe_regex, attr)
+            assert getattr(thread_safe_regex, attr) == getattr(regex, attr)
+
+
+def test_threadsafe_skips_missing_module_metadata_under_gil0():
+    thread_safe_builtins = ThreadSafe(builtins)
+
+    for attr in ("__package__", "__file__", "__spec__"):
+        if hasattr(builtins, attr):
+            assert getattr(thread_safe_builtins, attr) == getattr(builtins, attr)
+        else:
+            assert not hasattr(thread_safe_builtins, attr)
diff --git a/tests/utils/test_safe_crash.py b/tests/utils/test_safe_crash.py
new file mode 100644
index 000000000000..3a3f3509a24c
--- /dev/null
+++ b/tests/utils/test_safe_crash.py
@@ -0,0 +1,92 @@
+# Copyright 2025 ModelCloud.ai team and The HuggingFace Inc. team.
+##
+# 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.
+
+"""Regression harness that reproduces the raw `regex` crash when the GIL is off."""
+
+import os
+import subprocess
+import sys
+
+import pytest
+
+
+pytestmark = pytest.mark.skipif(
+    not hasattr(sys, "_is_gil_enabled") or sys._is_gil_enabled(),
+    reason="Crash regression only runs when the GIL is disabled",
+)
+
+
+@pytest.mark.xfail(strict=False, reason="Raw regex crashes under PYTHON_GIL=0")
+def test_raw_regex_thread_safety_crashes_under_gil0():
+    script = r"""
+import threading
+import regex
+
+pattern_text = r"(?P<prefix>test)(?P<number>\d+)"
+ready_group = threading.Barrier(33)
+run_event = threading.Event()
+
+
+def worker():
+    ready_group.wait()
+    run_event.wait()
+    for _ in range(100):
+        match = regex.match(pattern_text, "test123")
+        assert match is not None
+        assert match.group("prefix") == "test"
+        assert match.group("number") == "123"
+
+
+threads = [threading.Thread(target=worker) for _ in range(32)]
+for thread in threads:
+    thread.start()
+
+ready_group.wait()
+run_event.set()
+
+for thread in threads:
+    thread.join()
+"""
+
+    env = dict(os.environ, PYTHON_GIL="0")
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        env=env,
+        capture_output=True,
+        text=True,
+    )
+
+    stdout = result.stdout.strip()
+    stderr = result.stderr.strip()
+    message_parts = [
+        "raw regex subprocess run",
+        f"return code: {result.returncode}",
+    ]
+    if stdout:
+        message_parts.append(f"stdout:\n{stdout}")
+    if stderr:
+        message_parts.append(f"stderr:\n{stderr}")
+    message = "\n".join(message_parts)
+
+    if result.returncode == 0:
+        pytest.fail("raw regex unexpectedly behaved thread-safely\n" + message)
+
+    if result.returncode == -11:
+        message += "\nProcess terminated with SIGSEGV (Segmentation fault)."
+
+    if message:
+        sys.stderr.write(message + "\n")
+        sys.stderr.flush()
+
+    pytest.fail(message)