CVE-2026-31253

The flash-attention training framework thru commit e724e2588cbe754beb97cf7c011b5e7e34119e62 2025-13-04 contains an insecure deserialization vulnerability CWE-502 in its checkpoint loading mechanism. The loadcheckpoint function in checkpoint.py and the checkpoint loading code in eval.py use...

CosyVoice 安全漏洞

CosyVoice is an open-source voice generation and AI voice cloning platform developed by FunAudioLLM. CosyVoice has a security vulnerability, which stems from the gRPC server component using torch.load to load the voice synthesis model without enabling the weights-only=True security parameter. Thi...

Unsafe Deserialization in Public `keras.layers.TorchModuleWrapper.from_config` Reaches `torch.load(weights_only=False)` Without Explicit Unsafe Opt-In

Description keras.layers.TorchModuleWrapper.fromconfig can deserialize attacker-controlled PyTorch pickle data through torch.load..., weightsonly=False when it is called directly through the public Keras layer API. TorchModuleWrapper is exported as a public Keras layer:...

CVE-2026-1839 Arbitrary Code Execution via Unsafe torch.load() in Trainer Checkpoint Loading in huggingface/transformers

A vulnerability in the HuggingFace Transformers library, specifically in the Trainer class, allows for arbitrary code execution. The loadrngstate method in src/transformers/trainer.py at line 3059 calls torch.load without the weightsonly=True parameter. This issue affects all versions of the...

CVE-2026-1839 Arbitrary Code Execution via Unsafe torch.load() in Trainer Checkpoint Loading in huggingface/transformers

A vulnerability in the HuggingFace Transformers library, specifically in the Trainer class, allows for arbitrary code execution. The loadrngstate method in src/transformers/trainer.py at line 3059 calls torch.load without the weightsonly=True parameter. This issue affects all versions of the...

CVE-2026-1839 Arbitrary Code Execution via Unsafe torch.load() in Trainer Checkpoint Loading in huggingface/transformers

A vulnerability in the HuggingFace Transformers library, specifically in the Trainer class, allows for arbitrary code execution. The loadrngstate method in src/transformers/trainer.py at line 3059 calls torch.load without the weightsonly=True parameter. This issue affects all versions of the...

CVE-2026-1839

A vulnerability in the HuggingFace Transformers library, specifically in the Trainer class, allows for arbitrary code execution. The loadrngstate method in src/transformers/trainer.py at line 3059 calls torch.load without the weightsonly=True parameter. This issue affects all versions of the...

CVE-2026-1839

CVE-2026-1839 concerns the HuggingFace Transformers library, affecting the Trainer class. The root cause is an unsafe load in src/transformers/trainer.py: _load_rng_state() calls torch.load() without weights_only=True, which can allow arbitrary code execution when loading a malicious checkpoint (...

Hugging Face Transformers 安全漏洞

Hugging Face Transformers is an open-source framework developed by Hugging Face for defining state-of-the-art machine learning models. It covers text, visual, audio, and multi-modal models, and can be used for both inference and training. There is a security vulnerability in Hugging Face...

The vulnerability of the weightsOnly parameter in the torch.load() function of the PyTorch machine learning framework allows a hacker to execute arbitrary code.

The vulnerability of the weightsOnly parameter in the torch.load function of the PyTorch machine learning framework is related to deficiencies in the deserialization mechanism. Exploiting this vulnerability could allow a malicious actor to execute arbitrary code remotely...

Arbitrary Code Execution via Unsafe torch.load() in Trainer Checkpoint Loading

Summary A critical arbitrary code execution vulnerability exists in HuggingFace Transformers' Trainer class. The loadrngstate method at src/transformers/trainer.py:3059 calls torch.load without the weightsonly=True parameter. While a safeglobals context manager wraps this call, it provides no...

CVE-2025-67729

LMDeploy is a toolkit for compressing, deploying, and serving LLMs. Prior to version 0.11.1, an insecure deserialization vulnerability exists in lmdeploy where torch.load is called without the weightsonly=True parameter when loading model checkpoint files. This allows an attacker to execute...

CVE-2025-67729

LMDeploy prior to v0.11.1 is affected by an insecure deserialization vulnerability in torch.load() called without weights_only=True when loading model checkpoint files (.bin/.pt). This can allow an attacker to execute arbitrary code on the victim's machine. The issue is patched in v0.11.1. Affect...

CVE-2025-67729 lmdeploy vulnerable to Arbitrary Code Execution via Insecure Deserialization in torch.load()

LMDeploy is a toolkit for compressing, deploying, and serving LLMs. Prior to version 0.11.1, an insecure deserialization vulnerability exists in lmdeploy where torch.load is called without the weightsonly=True parameter when loading model checkpoint files. This allows an attacker to execute...

CVE-2025-67729 lmdeploy vulnerable to Arbitrary Code Execution via Insecure Deserialization in torch.load()

LMDeploy is a toolkit for compressing, deploying, and serving LLMs. Prior to version 0.11.1, an insecure deserialization vulnerability exists in lmdeploy where torch.load is called without the weightsonly=True parameter when loading model checkpoint files. This allows an attacker to execute...

Deserialization of Untrusted Data

Overview Affected versions of this package are vulnerable to Deserialization of Untrusted Data in the torch.load function. An attacker can execute arbitrary code by providing a specially crafted model file that is loaded without proper security parameters. Details Serialization is a process of...

EUVD-2025-205455

lmdeploy vulnerable to Arbitrary Code Execution via Insecure Deserialization in torch.load...

PT-2025-53604

Name of the Vulnerable Software and Affected Versions LMDeploy versions prior to 0.11.1 Description LMDeploy is a toolkit used for compressing, deploying, and serving LLMs. A flaw exists where the torch.load function is called without the weights only=True parameter when loading model checkpoint...

Remote Command Execution

scio-pypi is vulnerable to Remote Command Execution. The vulnerability is due to torch.load executing unsafe deserialization even when weightsonly=True, which allows an attacker to craft malicious model files that trigger arbitrary code execution during loading...

CVE-2025-62164

A vulnerability in vLLM allows attackers to supply malicious serialized prompt-embedding tensors that are deserialized using torch.load without validation. Due to PyTorch 2.8.0 disabling sparse-tensor integrity checks by default, a crafted tensor can bypass bounds checks and cause an out-of-bound...