SecInfer: Preventing Prompt Injection Via Inference-Time Scaling

Prompt injection attacks pose a pervasive threat to the security of Large Language Models LLMs. State-of-the-art prevention-based defenses typically rely on fine-tuning an LLM to enhance its security, but they achieve limited effectiveness against strong attacks. In this work, we propose...

SafeSearch: Automated Red-Teaming for the Safety of LLM-Based Search Agents

Search agents connect LLMs to the Internet, enabling access to broader and more up-to-date information. However, unreliable search results may also pose safety threats to end users, establishing a new threat surface. In this work, we conduct two in-the-wild experiments to demonstrate both the...

Salesforce Patches Critical ForcedLeak Bug Exposing CRM Data via AI Prompt Injection

Cybersecurity researchers have disclosed a critical flaw impacting Salesforce Agentforce, a platform for building artificial intelligence AI agents, that could allow attackers to potentially exfiltrate sensitive data from its customer relationship management CRM tool by means of an indirect promp...

ChatGPT solves CAPTCHAs if you tell it they’re fake

If you’re seeing fewer or different CAPTCHA puzzles in the near future, that’s not because website owners have agreed that they’re annoying, but it might be because they no longer prove that the visitor is human. For those that forgot what CAPTCHA stands for: Completely Automated Public Turing te...

ShadowLeak Exploit Exposed Gmail Data Through ChatGPT Agent

Radware researchers revealed a service-side flaw in OpenAI's ChatGPT. The ShadowLeak attack had used indirect prompt injection to bypass defences and leak sensitive data, but the issue has since been fixed...

ShadowLeak Zero-Click Flaw Leaks Gmail Data via OpenAI ChatGPT Deep Research Agent

Cybersecurity researchers have disclosed a zero-click flaw in OpenAI ChatGPT's Deep Research agent that could allow an attacker to leak sensitive Gmail inbox data with a single crafted email without any user action. The new class of attack has been codenamed ShadowLeak by Radware. Following...

ChatGPT Deep Research zero-click vulnerability fixed by OpenAI

OpenAI has moved quickly to patch a vulnerability known as “ShadowLeak” before anyone detected real-world abuse. Revealed by researchers yesterday, ShadowLeak was an issue in OpenAI’s Deep Research project that attackers could exploit by simply sending an email to the target. Deep Research was...

CVE-2025-59417

Lobe Chat is an open-source artificial intelligence chat framework. Prior to version 1.129.4, there is a a cross-site scripting XSS vulnerability when handling chat message in lobe-chat that can be escalated to remote code execution on the user’s machine. In lobe-chat, when the response from the...

CVE-2025-59417 Lobe Chat Desktop Vulnerable to Remote Code Execution via XSS in Chat Messages

Lobe Chat is an open-source artificial intelligence chat framework. Prior to version 1.129.4, there is a a cross-site scripting XSS vulnerability when handling chat message in lobe-chat that can be escalated to remote code execution on the user’s machine. In lobe-chat, when the response from the...

Early Approaches to Adversarial Fine-Tuning for Prompt Injection Defense: a 2022 Study of GPT-3 and Contemporary Models

This paper documents early research conducted in 2022 on defending against prompt injection attacks in large language models, providing historical context for the evolution of this critical security domain. This research focuses on two adversarial attacks against Large Language Models LLMs: promp...

Securing AI Agents: Implementing Role-Based Access Control for Industrial Applications

The emergence of Large Language Models LLMs has significantly advanced solutions across various domains, from political science to software development. However, these models are constrained by their training data, which is static and limited to information available up to a specific date...

Arbitrary Code Injection

Overview Affected versions of this package are vulnerable to Arbitrary Code Injection via the gonja template parsing process. An attacker can access arbitrary files on the server by injecting malicious template statements into prompts. Allowing an attacker to insert a statement into a prompt to...

CVE-2025-58372

Roo Code is an AI-powered autonomous coding agent that lives in users' editors. Versions 3.25.23 and below contain a vulnerability where certain VS Code workspace configuration files .code-workspace are not protected in the same way as the .vscode folder. If the agent was configured to auto-appro...

Multimodal Prompt Injection Attacks: Risks and Defenses for Modern LLMs

Large Language Models LLMs have seen rapid adoption in recent years, with industries increasingly relying on them to maintain a competitive advantage. These models excel at interpreting user instructions and generating human-like responses, leading to their integration across diverse domains,...

EchoLeak: the First Real-World Zero-Click Prompt Injection Exploit in a Production LLM System

Large language model LLM assistants are increasingly integrated into enterprise workflows, raising new security concerns as they bridge internal and external data sources. This paper presents an in-depth case study of EchoLeak CVE-2025-32711, a zero-click prompt injection vulnerability in Microso...

Decoding Latent Attack Surfaces in LLMs: Prompt Injection Via HTML in Web Summarization

Large Language Models LLMs are increasingly integrated into web-based systems for content summarization, yet their susceptibility to prompt injection attacks remains a pressing concern. In this study, we explore how non-visible HTML elements such as , aria-label, and alt attributes can be exploit...

CVE-2025-58372

Roo Code is an AI-powered autonomous coding agent that lives in users' editors. Versions 3.25.23 and below contain a vulnerability where certain VS Code workspace configuration files .code-workspace are not protected in the same way as the .vscode folder. If the agent was configured to auto-appro...

CVE-2025-58372 Roo Code: Potential Remote Code Execution via .code-workspace

Roo Code is an AI-powered autonomous coding agent that lives in users' editors. Versions 3.25.23 and below contain a vulnerability where certain VS Code workspace configuration files .code-workspace are not protected in the same way as the .vscode folder. If the agent was configured to auto-appro...

CVE-2025-58372 Roo Code: Potential Remote Code Execution via .code-workspace

Roo Code is an AI-powered autonomous coding agent that lives in users' editors. Versions 3.25.23 and below contain a vulnerability where certain VS Code workspace configuration files .code-workspace are not protected in the same way as the .vscode folder. If the agent was configured to auto-appro...

CVE-2025-58372

Roo Code CVE-2025-58372 affects versions ≤3.25.23 where certain VS Code workspace files (.code-workspace) aren’t protected like the .vscode folder. If auto-approve for file writes is enabled and prompts are manipulated (e.g., via prompt injection), an attacker could write malicious workspace sett...