Pickle-Deserialization-Exploit-in-Transformers

Transformers의 pickle 역직렬화 취약점 악용 실험 소개 pickle 역직렬화 메커니즘과 reduce 공격 원리를 소개하고, Hugging Face Transformers 라이브러리의 역직렬화 취약점 CVE-2024-3568을 재현하며, Transformers 라이브러리의 보안 위험을 분석하고 보안 강화를 위한 효과적인 방안을 제시합니다. 자세한 내용은 실험 PDF 문서를 참조하세요. poc 작성자 코드는 CVE-2024-3568 또는 authorinf 폴더를 참조하세요. 주요 정보 1. 취약점 위치...

CVE-2026-23001-Hugging-Face-Transformers-Model-Deserialization-Arbitrary-Code-via-Pickle-

CVE-2026-23001 – RCE da deserializzazione dei modelli Hugging Face Transformers Panoramica I modelli Hugging Face salvati con torch.save utilizzano la serializzazione pickle di Python, che è intrinsecamente non sicura. Se un utente scarica un modello apparentemente legittimo da un repository, il...

FaceLivenessDetection-SDK

Dieses Repo ist veraltet und wurde durch https://github.com/DoubangoTelecom/ultimateFace-SDK ersetzt...

OpenAI Says Reward Hacking Drove AI Agents to Exploit Zero-Days and Breach Hugging Face

OpenAI on Wednesday revealed that reward hacking was a key driver behind the artificial intelligence AI-powered hack of Hugging Face last month, adding that it found evidence of misaligned behavior as early as late May. The incident, the company said, took place during cybersecurity evaluations o...

CVE-2026-15679

A flaw was found in Hugging Face PyTorch Image Models. A remote attacker could exploit a deserialization of untrusted data vulnerability by enticing a user to visit a malicious page or open a malicious file. This could lead to arbitrary code execution in the context of the current process. The fl...

CVE-2026-47666 Penpot: Stored XSS via custom font family name injected into a @font-face style rule

Penpot is an open-source design and prototyping platform. In versions up to and including 2.14.3, Penpot is vulnerable to stored cross-site scripting through custom font family names, which are interpolated into a @font-face CSS rule and injected into the page as HTML without sanitization. Becaus...

What We Still Don’t Know About OpenAI’s Hugging Face Hack

The AI giant acknowledges that it could have done far more to prevent its AI agents from going rogue. But it still fails to explain why it didn't see this fiasco coming...

When an AI Agent Turned Attacker: What Qualys Sees Across Every Phase of the Hugging Face Kubernetes Intrusion

A phase-by-phase detection mapping of the first publicly documented autonomous agent intrusion against production infrastructure — including the phases where no product in our category sees anything at all. Executive Summary On July 9, 2026, an autonomous AI agent running inside an OpenAI...

CVE-2026-79784

Vocos instantiates a class named by a configuration file without restricting which class may be named. instantiateclass in vocos/pretrained.py takes the classpath value from the configuration, splits it into a module and an attribute, imports the module with import, resolves the attribute with...

CVE-2026-79784 Vocos through 0.1.0 Arbitrary Code Execution via Unrestricted class_path in Model Configuration

Vocos instantiates a class named by a configuration file without restricting which class may be named. instantiateclass in vocos/pretrained.py takes the classpath value from the configuration, splits it into a module and an attribute, imports the module with import, resolves the attribute with...

EUVD-2026-65542

Vocos instantiates a class named by a configuration file without restricting which class may be named. instantiateclass in vocos/pretrained.py takes the classpath value from the configuration, splits it into a module and an attribute, imports the module with import, resolves the attribute with...

CVE-2026-79784 Vocos through 0.1.0 Arbitrary Code Execution via Unrestricted class_path in Model Configuration

Vocos instantiates a class named by a configuration file without restricting which class may be named. instantiateclass in vocos/pretrained.py takes the classpath value from the configuration, splits it into a module and an attribute, imports the module with import, resolves the attribute with...

CVE-2026-79784

Vocos (through 0.1.0 ), a Python library maintained by gemelo-ai, suffers from arbitrary code execution via an unrestricted class_path field in its model configuration. The vulnerable function instantiate_class in vocos/pretrained.py uses __import__ and getattr on a user-supplied dotted path with...

CVE-2026-76841

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...

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 through 2.11.0 Remote Code Execution via Hardcoded trust_remote_code in Model Loaders

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....

CVE-2026-76841 Xinference through 2.11.0 Remote Code Execution via Hardcoded trust_remote_code in Model Loaders

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 through 2.11.0 Remote Code Execution via Hardcoded trust_remote_code in Model Loaders

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...

Astra Linux – Vulnerability in CGal

There are multiple code execution vulnerabilities in the Nef polygon-parsing functionality of CGAL libcgal CGAL-5.1.1. A specially crafted malformed file can lead to an out-of-bounds read and type confusion, which may result in code execution. An attacker can provide malicious input to trigger an...