CVE-2026-31221

PyTorch-Lightning versions 2.6.0 and earlier contain an insecure deserialization vulnerability (CWE-502) in the checkpoint loading mechanism. The LightningModule.load_from_checkpoint() (and related checkpoint loading paths) call torch.load() without weights_only=True, allowing deserialization of ...

CVE-2026-31218

The CVE concerns the optimate project’s neural_magic_training.py, where _load_model() deserializes a state_dict.pt with torch.load() without enabling weights_only=True. This enables deserialization of arbitrary Python objects via Pickle, allowing a remote attacker to provide a crafted state_dict....

CVE-2026-31219

The loadmodel function in the neuralmagictraining.py script of the optimate project in commit a6d302f912b481c94370811af6b11402f51d377f 2024-07-21 is vulnerable to insecure deserialization CWE-502. When a user provides a single model file path e.g., .pt or .pth via the --model command-line argumen...

CVE-2026-31214

The torch-checkpoint-shrink.py script in the ml-engineering project in commit 0099885db36a8f06556efe1faf552518852cb1e0 2025-20-27 contains an insecure deserialization vulnerability CWE-502. The script uses torch.load to process PyTorch checkpoint files .pt without enabling the security-restrictiv...

CVE-2026-31214

The torch-checkpoint-shrink.py script in the ml-engineering project in commit 0099885db36a8f06556efe1faf552518852cb1e0 2025-20-27 contains an insecure deserialization vulnerability CWE-502. The script uses torch.load to process PyTorch checkpoint files .pt without enabling the security-restrictiv...

CVE-2023-30059

An insecure direct object reference in MK-Auth 23.01K4.9 allows an attacker to access and send support calls for other users by manipulating the chamado parameter via a crafted GET request. The documents do not provide details on exploited versions, specific vectors beyond the parameter manipulat...

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. This vulnerability arises from the model loading process, where the .pt files in the user-specified directory are loaded using torch.load, without enabling...

CVE-2026-31237

The Ludwig framework (up to version 0.10.4) is reported to be vulnerable to insecure deserialization (CWE-502) in its predict() function. If a user supplies a dataset file path to predict(), Ludwig attempts to determine the file format and, when encountering a pickle (.pkl) file, loads it via pan...

Snorkel 安全漏洞

Snorkel is an open-source system developed by Snorkel that uses weak supervision to quickly generate training data. Versions of Snorkel prior to v0.10.0 contain security vulnerabilities. These vulnerabilities stem from the Trainer class’s Trainer.load method, which uses torch.load to load model...

PT-2026-40124

The Ludwig framework thru 0.10.4 is vulnerable to insecure deserialization CWE-502 through its predict method. When a user provides a dataset file path to the predict method, the framework automatically determines the file format. If the file is a pickle .pkl file, it is loaded using pandas.read...

CVE-2026-31219

The loadmodel function in the neuralmagictraining.py script of the optimate project in commit a6d302f912b481c94370811af6b11402f51d377f 2024-07-21 is vulnerable to insecure deserialization CWE-502. When a user provides a single model file path e.g., .pt or .pth via the --model command-line argumen...

Mk-Auth 安全漏洞

Mk-Auth is a Brazilian internet service provider management system developed by Mk-Auth company. It is used to control client access and permissions through a network interface panel. Version 23.01K4.9 of MK-Auth contains a security vulnerability caused by insecure direct object references. This...

PT-2026-39995

Prior to 2025-11-03, well-intended users of Terraform or REST API for Google Cloud AlloyDB for PostgreSQL could have created clusters with an insecure default password which could have been exploited by a remote attacker to gain full administrative access to the database. Exploitation required...

CVE-2026-31238

The Ludwig framework thru 0.10.4 is vulnerable to insecure deserialization CWE-502 in its model serving component. When starting a model server with the ludwig serve command, the framework loads model weight files using torch.load without enabling the security-restrictive weightsonly=True...

CVE-2026-31234

Horovod through v0.28.1 exposes an insecure deserialization vulnerability (CWE-502) in its KVStore HTTP server. The KVStore server lacks authentication/authorization, allowing remote attackers to write arbitrary data via HTTP PUT. When a Horovod worker subsequently reads data from KVStore (via HT...

CVE-2026-31219

The connected documents confirm a concrete vulnerability in the optimate project: the _load_model() (or load_model()) function in neural_magic_training.py deserializes a single model file passed via --model using torch.load() without weights_only=True, enabling arbitrary Python object deserializa...

PT-2026-40057

The load model function in the neural magic training.py script of the optimate project in commit a6d302f912b481c94370811af6b11402f51d377f 2024-07-21 is vulnerable to insecure deserialization CWE-502. When loading a model state dictionary from a state dict.pt file via torch.load, the function does...

CVE-2026-31218

The loadmodel function in the neuralmagictraining.py script of the optimate project in commit a6d302f912b481c94370811af6b11402f51d377f 2024-07-21 is vulnerable to insecure deserialization CWE-502. When loading a model state dictionary from a statedict.pt file via torch.load, the function does not...

OptiMate 安全漏洞

OptiMate is an AI model optimization tool library developed by Nebuly. There is a security vulnerability in OptiMate. This vulnerability stems from the loadmodel function in the neuralmagictraining.py script, which loads model files using torch.load, without enabling the weightsonly=True paramete...

CVE-2026-31222

The Snorkel library prior to v0.10.0 contains an insecure deserialization vulnerability (CWE-502) in Trainer.load(), where model checkpoints are loaded with torch.load() without weights_only=True. This allows deserialization of arbitrary Python objects via Pickle, enabling remote code execution w...