PoC-Adapt: Semantic-Aware Automated Vulnerability Reproduction with LLM Multi-Agents and Reinforcement Learning-Driven Adaptive Policy

While recent approaches leverage large language models LLMs and multi-agent pipelines to automatically generate proof-of-concept PoC exploits from vulnerability reports, existing systems often suffer from two fundamental limitations: unreliable validation based on surface-level execution signals...

Semantics over Syntax: Uncovering Pre-Authentication 5G Baseband Vulnerabilities

Modern 5G user equipment UE processes Radio Resource Control RRC configuration messages during early control-plane exchanges, before authentication and integrity protection are established. Prior work for testing 5G UEs has largely focused on constructing syntactically invalid inputs. In contrast...

AttackEval: A Systematic Empirical Study of Prompt Injection Attack Effectiveness against Large Language Models

Prompt injection has emerged as a critical vulnerability in large language model LLM deployments, yet existing research is heavily weighted toward defenses. The attack side -- specifically, which injection strategies are most effective and why -- remains insufficiently studied.We address this gap...

ContractShield: Bridging Semantic-Structural Gaps Via Hierarchical Cross-Modal Fusion for Multi-Label Vulnerability Detection in Obfuscated Smart Contracts

Smart contracts are increasingly targeted by adversaries employing obfuscation techniques such as bogus code injection and control flow manipulation to evade vulnerability detection. Existing multimodal methods often process semantic, temporal, and structural features in isolation and fuse them...

CVE-2026-30309

InfCode's terminal auto-execution module contains a critical command filtering vulnerability that renders its blacklist security mechanism completely ineffective. The predefined blocklist fails to cover native high-risk commands in Windows PowerShell such as powershell, and the matching algorithm...

Efficient Software Vulnerability Detection Using Transformer-Based Models

Detecting software vulnerabilities is critical to ensuring the security and reliability of modern computer systems. Deep neural networks have shown promising results on vulnerability detection, but they lack the capability to capture global contextual information on vulnerable code. To address th...

Hidden Ads: Behavior Triggered Semantic Backdoors for Advertisement Injection in Vision Language Models

Vision-Language Models VLMs are increasingly deployed in consumer applications where users seek recommendations about products, dining, and services. We introduce Hidden Ads, a new class of backdoor attacks that exploit this recommendation-seeking behavior to inject unauthorized advertisements...

CVE-2026-32901

Affected software: OpenClaw prior to version 2026.3.2. Vulnerability type: semantic drift in node system.run approval hardening that rewrites wrapper argv, enabling execution of unintended local scripts when an attacker can influence argv and place malicious files in the approved working director...

Memory Poisoning and Secure Multi-Agent Systems

Memory poisoning attacks for Agentic AI and multi-agent systems MAS have recently caught attention. It is partially due to the fact that Large Language Models LLMs facilitate the construction and deployment of agents. Different memory systems are being used nowadays in this context, including...

CVE-2026-32622

SQLBot is an intelligent data query system based on a large language model and RAG. Versions 1.5.0 and below contain a Stored Prompt Injection vulnerability that chains three flaws: a missing permission check on the Excel upload API allowing any authenticated user to upload malicious terminology,...

CVE-2026-32622

SQLBot (versions ≤ 1.5.x) exposes a Stored Prompt Injection vulnerability consisting of three chained flaws: (1) missing permission check on the Excel upload API allowing any authenticated user to upload malicious terminology, (2) unsanitized storage of terminology descriptions containing dangero...

CVE-2026-32622 SQLBot: Remote Code Execution via Terminology Poisoning

SQLBot is an intelligent data query system based on a large language model and RAG. Versions 1.5.0 and below contain a Stored Prompt Injection vulnerability that chains three flaws: a missing permission check on the Excel upload API allowing any authenticated user to upload malicious terminology,...

PT-2026-26096

Name of the Vulnerable Software and Affected Versions CKAN MCP Server versions prior to 0.4.85 Description The software allows making HTTP requests to arbitrary endpoints without restriction because it lacks URL validation on the base url parameter. This issue affects the ckan package search,...

Poisoning the Pixels: Revisiting Backdoor Attacks on Semantic Segmentation

Semantic segmentation models are widely deployed in safety-critical applications such as autonomous driving, yet their vulnerability to backdoor attacks remains largely underexplored. Prior segmentation backdoor studies transfer threat settings from existing image classification tasks, focusing...

KLA90920 Multiple vulnerabilities in Microsoft Open Source Software

Multiple vulnerabilities were found in Microsoft Open Source Software. Malicious users can exploit these vulnerabilities to execute arbitrary code, bypass security restrictions. Below is a complete list of vulnerabilities: 1. A remote code execution vulnerabilitycan be exploited remotely to execu...

SCAFFOLD-CEGIS: Preventing Latent Security Degradation in LLM-Driven Iterative Code Refinement

The application of large language models to code generation has evolved from one-shot generation to iterative refinement, yet the evolution of security throughout iteration remains insufficiently understood. Through comparative experiments on three mainstream LLMs, this paper reveals the iterativ...

EUVD-2026-9988

OpenSift is an AI study tool that sifts through large datasets using semantic search and generative AI. Prior to version 1.6.3-alpha, the URL ingest pipeline accepted user-controlled remote URLs with incomplete destination restrictions. Although private/local host checks existed, missing...

SemFuzz: A Semantics-Aware Fuzzing Framework for Network Protocol Implementations

Network protocols are the foundation of modern communication, yet their implementations often contain semantic vulnerabilities stemming from inadequate understanding of specification semantics. Existing gray-box and black-box testing approaches lack semantic modeling of protocols, making it...

Microsoft Semantic Kernel Code Injection Vulnerability

Microsoft Semantic Kernel is a large model orchestration framework from Microsoft Corporation, USA. A code injection vulnerability exists in Microsoft Semantic Kernel versions prior to 1.39.4. The vulnerability stems from the InMemoryVectorStore filtering feature failing to properly filter specia...

PT-2026-23642

OpenSift is an AI study tool that sifts through large datasets using semantic search and generative AI. Prior to version 1.6.3-alpha, some endpoints returned raw exception strings to clients. Additionally, login token material was exposed in UI/rendered responses and token rotation output. This...