hcaptcha-challenger

hCaptcha Challenger 🚀 Affrontez avec élégance le défi hCaptcha grâce à un modèle de langage multimodal de grande taille. Introduction Ne s'appuie sur aucun script Tampermonkey. N'utilise aucun service anti-captcha tiers. Il suffit d'implémenter quelques interfaces pour rendre possible AI vs AI...

Multimodal-Unlearnable-Examples

Exemples non-apprenables multimodaux : Protéger les données contre l'apprentissage contrastif multimodal ACM MM2024 Résumé : L'apprentissage contrastif multimodal MCL a montré des progrès remarquables en classification zero-shot en apprenant à partir de millions de paires image-légende collectées...

bordair-multimodal

Jeu de données d'injection de prompt multimodal 516 588 échantillons étiquetés 251 782 attaques + 251 576 bénins, plus une partition de validation de 13 230 échantillons réels répartis sur cinq versions du jeu de données plus l'ingestion de jeux de données externes, couvrant les attaques...

vLLM 0.8.3 - 0.14.0 - Information Disclosure

vLLM 0.8.3 to - 0.14.1 contains an information disclosure caused by leaking a heap address in error messages from the multimodal endpoint when processing invalid images, letting remote attackers reduce ASLR entropy, exploit requires sending invalid images. id: CVE-2026-22778 info: name: vLLM 0.8....

ai-captcha-bypass

Solveur de CAPTCHA propulsé par l'IA Ce projet est un outil en ligne de commande basé sur Python qui utilise de grands modèles multimodaux LMM comme GPT-4o d'OpenAI et Gemini de Google pour résoudre automatiquement différents types de CAPTCHA. Il s'appuie sur Selenium pour l'automatisation du...

safety-awareness

Sensibilisation au Transfert de Sécurité pour la Dérive de Sécurité Intermodale Implémentation officielle de Transfer Safety Awareness for Cross-Modal Safety Drift in Multimodal Large Language Models EMNLP 2026 Findings. Ce dépôt contient le pipeline de direction de sensibilisation à la sécurité...

Learning-to-Detect

Learning to Detect Unknown Jailbreak Attacks in Large Vision-Language Models Official implementation of “Learning to Detect Unknown Jailbreak Attacks in Large Vision-Language Models.” This repository contains the data-processing, hidden-state extraction, classifier training, safety-pattern...

vLLM 0.8.3 < 0.14.1 Information Disclosure (GHSA-4r2x-xpjr-7cvv)

The version of the vLLM Python package installed on the remote host is 0.8.3 prior to 0.14.1. It is, therefore, affected by an information disclosure vulnerability. When an invalid image is sent to the vLLM multimodal endpoint, PIL throws an error that vLLM returns to the client, leaking a heap...

CVE-2026-105754

A flaw was found in vLLM. The scale-out inference service fails to properly validate caller-supplied multimodal parameters, such as image geometry and cache identifiers, against the active model's configuration. An authenticated remote attacker can exploit this vulnerability by submitting crafted...

CVE-2026-105753

A flaw was found in vLLM. This vulnerability allows a remote user to cause a Denial of Service DoS on the model serving engine. The issue occurs when a multimodal request is rejected after media processing, causing the frontend and backend caches to become desynchronized. A subsequent request...

EUVD-2026-92843

vLLM: Mirrored multimodal IPC caches desync after a rejected request — a later request reusing the same media hash trips a receiver assertion in the engine core...

vLLM: Mirrored multimodal IPC caches desync after a rejected request — a later request reusing the same media hash trips a receiver assertion in the engine core

Affected - Ecosystem / package: pip / vllm - Affected versions: vLLM ≤ 0.25.1 confirmed on 0.25.1, commit 752a3a504485. The lower bound predates 0.25.1; maintainers can confirm how far back the mirrored sender/receiver cache protocol reaches. Summary vLLM's default multimodal cache...

GHSA-PH3R-5JFG-F84F vLLM: Mirrored multimodal IPC caches desync after a rejected request — a later request reusing the same media hash trips a receiver assertion in the engine core

Affected - Ecosystem / package: pip / vllm - Affected versions: vLLM ≤ 0.25.1 confirmed on 0.25.1, commit 752a3a504485. The lower bound predates 0.25.1; maintainers can confirm how far back the mirrored sender/receiver cache protocol reaches. Summary vLLM's default multimodal cache...

GHSA-PH72-CQR5-QPP7 vLLM: Scale-out disaggregated multimodal transport trusts caller-supplied features

Affected - Ecosystem / package: pip / vllm - Affected versions: vLLM ≤ 0.25.1 confirmed on 0.25.1, commit 752a3a504485. The lower bound predates 0.25.1; maintainers can confirm how far back the scale-out transport path reaches. Summary vLLM's disaggregated scale-out transport splits a multimodal...

PYSEC-2026-4197

vLLM is an inference and serving engine for large language models. Prior to 0.28.0, the default mirrored multimodal LRU cache can commit a media hash in the frontend sender cache during multimodal rendering and before engine admission, while the engine receiver cache never receives the payload if...

CVE-2026-105753

vLLM is an inference and serving engine for large language models. Prior to 0.28.0, the default mirrored multimodal LRU cache can commit a media hash in the frontend sender cache during multimodal rendering and before engine admission, while the engine receiver cache never receives the payload if...

CVE-2026-105754

The vLLM inference and serving engine (prior to version 0.30.0 ) contains a vulnerability in the /inference/v1/generate endpoint within the disaggregated scale-out path. The root cause is that the system trusts caller-supplied features—specifically in the features.kwargs_data , features.mm_hashes...

CVE-2026-105754 vLLM: Scale-out disaggregated multimodal transport trusts caller-supplied features

vLLM is an inference and serving engine for large language models. Prior to 0.30.0, the /inference/v1/generate endpoint in the disaggregated scale-out path accepts caller-supplied tensors in the features.kwargsdata field, cache identifiers in the features.mmhashes field, ranges in the...

CVE-2026-105753 vLLM: Mirrored multimodal IPC caches desync after a rejected request — a later request reusing the same media hash trips a receiver assertion in the engine core

vLLM is an inference and serving engine for large language models. Prior to 0.28.0, the default mirrored multimodal LRU cache can commit a media hash in the frontend sender cache during multimodal rendering and before engine admission, while the engine receiver cache never receives the payload if...

CVE-2026-105753

The vLLM inference and serving engine is vulnerable to a shared-service availability failure in versions prior to 0.28.0 . The root cause is a desynchronization in the mirrored multimodal LRU cache ; if a request is rejected during multimodal rendering, the frontend sender cache commits a media h...