CVE-2025-32434

CVE-2025-32434: PyTorch RCE脆弱性 - PoC CVE-2025-32434とは？ CVE-2025-32434 は、深層学習やニューラルネットワークで広く使用されるPythonライブラリPyTorchにおける、重大なリモートコード実行（RCE）脆弱性です。 中心的な問題...

CVE-2024-3660-PoC

CVE-2024-3660 – TensorFlow Keras における悪意あるモデルによる任意コード実行 概要 CVE-2024-3660 は、TensorFlow の Keras フレームワークにおける任意コード実行の脆弱性で、バージョン 2.13 に影響します。 攻撃者は、Python コードをモデル内（例：Lambda レイヤー経由）に埋め込むことができ、モデル読み込み時に自動的に 実行されます。 脆弱な環境で悪意ある .h5 モデルファイルが読み込まれると、読み込みプロセスと同じ権限で任意のコマンドが実行されます。 影響を受けるバージョン TensorFlow の Keras...

CVE-2026-54559

PocketSphinx (a lightweight speech recognizer, including its Python bindings) versions 5prealpha through 5.1.0 are affected by buffer overflow vulnerabilities in its model-loading code. The trie language-model loaders in src/lm/ngram_model_trie.c fail to validate boundary conditions in ARPA, DMP,...

CVE-2026-47117-openmed-rce

CVE-2026-47117: PII モデルロードを介した OpenMed の未認証リモートコード実行 深刻度: Critical、CVSS 4.0 9.3 、CVSS 3.1 9.8 CNA である VulnCheck により割り当て ベクター v4.0: CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N ベクター v3.1: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H 影響を受けるバージョン: OpenMed 1.5.2 修正バージョン: 1.5.2 CWE:...

CVE-2026-90553

A flaw was found in vLLM. The LlavaOnevision2 processor loader incorrectly ignores the trustremotecode parameter, which is designed to prevent the execution of untrusted code. This oversight allows an attacker to craft a malicious model containing arbitrary code. When such a model is loaded, the...

CVE-2026-90553

vLLM before 0.28.0 contains a remote code execution vulnerability in the LlavaOnevision2 processor loader . The loader ignores the trust_remote_code parameter when loading remote processor classes, allowing an attacker to craft a malicious model with arbitrary code in processing_llava_onevision2....

CVE-2026-88054

A flaw was found in Tesseract, an open-source Optical Character Recognition OCR engine. A remote attacker could provide a specially crafted '.traineddata' model with a zero-length stack for certain internal layers. This would cause an empty-stack dereference during model loading, leading to a...

CVE-2026-88054

Tesseract is an open-source OCR engine affected in version 5.5.3 and earlier by a denial-of-service flaw. The root cause is in Plumbing::DeSerialize (src/lstm/plumbing.cpp), which rejects excessively large network stacks but accepts a zero-length stack for NT_SERIES, NT_PARALLEL, or NT_REVERSED l...

CVE-2026-22807_Range

🛡️ CVE-2026-22807 Simulation: AI Supply Chain RCE Target Range ⚠️ 免责声明 / Disclaimer 本项目仅用于安全研究与教学目的 ，旨在演示 AI 模型加载过程中的供应链攻击原理（Remote Code Execution）。 请勿在生产环境中使用生成的恶意模型，也严禁将本项目用于任何非法攻击行为。开发者不承担任何因使用不当造成的法律责任。 📖 项目简介 Introduction 这是一个轻量级的 Python 靶场，用于模拟和复现 CVE-2026-22807 （一个典型的 AI 供应链漏洞逻辑）。 该靶场演示了当 AI...

CVE-2026-0596-Reproduction

CVE-2026-0596: MLflowエコシステムにおける安全でないデシリアライゼーションによる任意コード実行 mlflow==2.11.1およびmlserver==1.3.5における信頼できないモデル読み込みパイプラインに関連する検証、基礎メカニズム、アーキテクチャ上の脆弱性を詳述した包括的なセキュリティ研究レポート。 ⚠️ 脆弱性インテリジェンス勧告: CVE-2026-0596...

CVE-2026-86169

Axolotl before 0.19.0 contains a remote code execution vulnerability in the multipack patch path where trustremotecode defaults to None instead of False, causing the security guard to be bypassed. Attackers can execute arbitrary Python code by crafting a malicious Hugging Face model repository...

CVE-2026-79657

NLTK versions before 3.10.3 contain a remote code execution vulnerability in allowlisted pickle loaders that trust entire module namespaces instead of specific safe callables. Attackers can craft malicious pickle payloads invoking dangerous in-namespace functions like ReppTokenizer.execute and...

CVE-2026-79657 NLTK before 3.10.3 Remote Code Execution via Unsafe Pickle Deserialization

NLTK versions before 3.10.3 contain a remote code execution vulnerability in allowlisted pickle loaders that trust entire module namespaces instead of specific safe callables. Attackers can craft malicious pickle payloads invoking dangerous in-namespace functions like ReppTokenizer.execute and...

CVE-2026-76843

The official Flair wheels for 0.15.0 and 0.15.1 still contain flair/models/clustering.py, whose ClusteringModel.load static method returns pickle.loadsjoblib.loadstrmodelfile and so executes arbitrary Python while loading a model file. Loading a model supplied by an attacker therefore runs that...

CVE-2026-76843

CVE-2026-76843 affects the official Flair wheels for versions 0.15.0 and 0.15.1 . The file flair/models/clustering.py remains present in the distributed package despite being removed from the documented API. Its ClusteringModel.load static method deserializes via pickle.loads(joblib.load(str(mode...

EUVD-2026-64847

Xinference loads models with Hugging Face remote code execution unconditionally enabled, and before version 2.12.0 exposes no setting to disable it. Six loader call sites pass trustremotecode=True as a literal or as an unconditional default: RerankModel.gettokenizer in...

CVE-2026-76841

Xinference (by XorbitsAI) through version 2.11.0 unconditionally sets trust_remote_code=True at six model loader call sites spanning the rerank, embedding, and LLM modules. An attacker with model launch access can register a model with an unknown type and an arbitrary path, causing AutoTokenizer....

PT-2026-80851

Xinference loads models with Hugging Face remote code execution unconditionally enabled, and before version 2.12.0 exposes no setting to disable it. Six loader call sites pass trust remote code=True as a literal or as an unconditional default: RerankModel. get tokenizer in...

CVE-2026-76395 Remote Code Execution (RCE) through Deserialization of Untrusted Data in the Model Loading REST API in Splunk AI Toolkit

In Splunk AI Toolkit versions below 6.0.0, a user who holds the "power" Splunk role could execute arbitrary code on the Splunk server by loading a model file containing crafted sparse matrix data. The deserialization of untrusted data is possible because a model codec in Splunk AI Toolkit...

CVE-2026-76395

Splunk AI Toolkit versions below 6.0.0 deserialize sparse matrix data containing embedded pickle payloads without validation, allowing a user with the "power" role to execute arbitrary code via the model-loading REST API. The fix is to upgrade to version 6.0.0 or later. No in-the-wild exploitatio...