cve-2025-32434-poc

CVE-2025-32434 Description Technical Details The vulnerability entry point is load in torch/serialization.py when protection weightsonly is enabled: loading a tar file would call these functions in chain: 1. load: torch/serialization.py:1117 2. loadlegacy: torch/serialization.py:1376 to...

alibi-detect

Alibi Detect 是一个源码可用的 Python 库，专注于异常值 、对抗性 和漂移 检测。该包旨在覆盖适用于表格数据、文本、图像和时间序列的在线和离线检测器。漂移检测同时支持 TensorFlow 和 PyTorch 后端。 文档 如需了解更多有关在生产环境中监测异常值和分布的重要性的背景信息，请查看来自 ICML 2020 研讨会 机器学习系统的部署与监测挑战 上的这个演讲，该演讲基于论文生产环境中模型的监测与可解释性并引用了 Alibi Detect。...

saffron

Saffron-1: Inference Scaling for LLM Safety Assurance 📖 Paper 🛠️ Dependencies The code was tested under the following dependencies: Python 3.12.3 CUDA 12.2 typingextensions==4.14.0 numpy==2.2.6 torch==2.5.1 huggingfacehub==0.30.2 accelerate==1.1.1 datasets==3.1.0 evaluate==0.4.3...

CVE-2025-32434

CVE-2025-32434: PyTorch RCE Vulnerability - PoC What is CVE-2025-32434? CVE-2025-32434 is a critical Remote Code Execution RCE vulnerability in PyTorch, a widely used Python library for deep learning and neural networks. The Core Problem The vulnerability exists in the torch.load function, even...

foolbox

home: true heroImage: /logo.png heroText: Foolbox tagline: "Foolbox：在 PyTorch、TensorFlow 和 JAX 中快速发起对抗攻击，以基准测试机器学习模型的鲁棒性" actionText: 开始使用 → actionLink: /guide/ features: title: 原生性能 details: Foolbox 3 基于 EagerPy 构建，可原生运行于 PyTorch、TensorFlow 和 JAX 中。 title: 最先进的攻击方法 details: Foolbox...

fickling

Fickling Fickling 是一个用于 Python pickle 对象序列化的反编译器、静态分析器和字节码重写器。你可以使用 fickling 来检测、分析、逆向工程，甚至创建恶意的 pickle 或基于 pickle 的文件，包括 PyTorch 文件。 Fickling 既可以作为 Python 库 使用，也可以作为 CLI 使用。 安装 保护 AI/ML 环境 通用恶意文件检测 高级用法 跟踪 pickle 执行 Pickle 代码注入 Pickle 反编译 PyTorch 多格式文件（polyglots） 更多信息 联系方式 安装 Fickling 已在 Python...

EXPLOIT-CVE-2024-7804

CVE-2024-7804 — PyTorch 分布式 RPC 不安全反序列化（RCE） 自包含实验环境 + 针对 CVE-2024-7804 的漏洞利用：通过不安全地反序列化不可信数据（CWE-502 ），在 PyTorch 的分布式 RPC 框架（torch.distributed.rpc）中实现远程代码执行。 受影响版本： torch 注意：该公告后来被维护者和 Red Hat 撤回/拒绝 ，理由是“已知功能”——PyTorch RPC 本质上是一个 可信对等 系统。该实验环境仍然是对所报告问题的准确、动手实践演示。 漏洞详情 每次 RPC 调用都会携带一个 PythonUDF...

Occ

Prohibited Items Segmentation via Occlusion-aware Bilayer Modeling ICME 2025 Official PyTorch Implementation of Prohibited Items Segmentation via Occlusion-aware Bilayer Modeling. Paper link: IEEE, Arxiv. Get Started Dependencies Code is successfully tested on Linux with Pytorch 2.1.2 and CUDA...

CVE-2024-5452

CVE-2024-5452 01. Overview of RCE and pytorch-lightning 1 Overview of RCE and pytorch-lightning RCE Remote Code Execution is a vulnerability that allows an attacker to execute arbitrary code remotely within a system or application, and is classified as CWE-94: Improper Control of Generation of...

CVE-2026-94093

DLR-RM stable-baselines3 up to 2.9.0 is affected by unsafe pickle deserialization in the PPO.load, load_replay_buffer, and VecNormalize.load functions of save_util.py, enabling remote attacks with no privileges required (user interaction required). In v2.9.0 the PyTorch tensor load path was harde...

CVE-2025-32434-exploit

Overview This exploit demonstrates a critical remote code execution vulnerability CVE-2025-32434 in PyTorch versions /tmp/whoami" Test the exploit loads the model to verify RCE python3 exploit.py -c "id" -t Advanced exploit variant with alternative bypass python3 exploit.py --advanced -c "curl -s...

CVE-2025-50461

CVE-2025-50461: Remote Code Execution via Unsafe Model Deserialization in Verl Status: RESERVED This CVE ID has been assigned but is not yet published in the official CVE list. A deserialization vulnerability exists in Volcengine's verl 3.0.0, specifically in the scripts/modelmerger.py script whe...

cleverhans

CleverHans（最新版本：v4.0.0） 本仓库包含 CleverHans 的源代码，这是一个用于评估机器学习系统对 对抗样本 漏洞的 Python 库。 你可以在随附的 博客 上了解更多关于此类漏洞的信息。 CleverHans 库持续开发中，始终欢迎 贡献 最新的攻击和防御方法。 特别是，我们始终欢迎帮助解决当前 开放的问题。 自 v4.0.0 起，CleverHans 支持 3 种框架：JAX、PyTorch 和 TF2。我们目前优先在 PyTorch 中实现攻击， 但非常欢迎为所有 3 种框架做出贡献。在 v3.1.0 及更早版本中， CleverHans 支持...

CVE-2023-43654

CVE-2023-43654 ShellTorch is a chain of 3 Critical Vulnerabilities in PyTorch TorchServe - the default inference server of PyTorch. This is an End-to-End POC for an SSRF vulnerability which leads to Arbitrary Remote Code Execution from the network 1. Run TorchServe 0.8.1 root@kitploit: docker run...

Agentic-CLIP-Benchmark

Agentic-CLIP-Benchmark: Zero-Shot Evaluation on CIFAR-10 -green.svg This project implements a robust, automated pipeline for evaluating OpenAI's CLIP ViT-B/32 model on the full CIFAR-10 test set 10,000 images. Developed using an AI Native workflow Trae IDE , it achieves a high-precision zero-shot...

netsentryx

NetSentryx PRO 🛡️ AI-Driven, High-Performance Real-Time Network Observability & Threat Detection Engine NetSentryx PRO is a next-generation, event-driven network monitoring and threat intelligence platform. Combining rule-based signatures with a PyTorch Deep Autoencoder, it provides sub-10ms packe...

CVE-2026-62997

Kedro-Datasets provides data connectors for Kedro. From version 5.0.0 until 9.5.0, kedrodatasetsexperimental.pytorch.PyTorchDataset in kedro-datasets loads .pt model files with torch.load without enforcing weightsonly=True, and user-supplied loadargs are silently dropped. On PyTorch versions...

CVE-2026-62997 Kedro-Datasets: Remote code execution in experimental `PyTorchDataset` via unsafe `torch.load`

Kedro-Datasets provides data connectors for Kedro. From version 5.0.0 until 9.5.0, kedrodatasetsexperimental.pytorch.PyTorchDataset in kedro-datasets loads .pt model files with torch.load without enforcing weightsonly=True, and user-supplied loadargs are silently dropped. On PyTorch versions...

CVE-2026-63632 affecting package pytorch for versions less than 2.2.2-17

CVE-2026-63632 affecting package pytorch for versions less than 2.2.2-17. A patched version of the package is available...

PYSEC-2026-3970 PyTorch Lightning path traversal vulnerability

In lightning-ai/pytorch-lightning version 2.3.2, a vulnerability exists in the LightningApp when running on a Windows host. The vulnerability occurs at the /api/v1/uploadfile/ endpoint, allowing an attacker to write or overwrite arbitrary files by providing a crafted filename. This can lead to...