red-teaming-auto-mode

Red-teaming Claude Code's auto-mode monitor Code for the paper Red-teaming Claude Code's auto-mode monitor. Production coding-agent monitors e.g. Claude Code's Auto Mode or Codex's Guardian review each action and block unsafe ones in real time. This repository red-teams them under a persistent...

Top AI Agents Built to Catch Malicious Code Can Be Tricked Into Running It

Ask an AI coding agent to scan open-source code for security holes, and it might run the attacker's code on your own machine instead. That is the finding in a proof-of-concept published Wednesday by the AI Now Institute, an attack it calls "Friendly Fire. " It works against Anthropic's Claude Cod...

CVE-2026-44454 Coder vulnerable to workspace auto-creation via crafted URL parameters without user consent

Coder allows organizations to provision remote development environments via Terraform. Prior to versions 2.29.7 and 2.30.2, the dotfiles registry module passed unsanitized user input to shell commands, allowing arbitrary code execution inside a provisioned workspace. Any user who supplied a craft...

Coder vulnerable to workspace auto-creation via crafted URL parameters without user consent

Command injection via dotfiles URI parameter combined with workspace auto-creation Summary The dotfiles registry module passed unsanitized user input to shell commands, allowing arbitrary code execution inside a provisioned workspace. Any user who supplied a crafted dotfilesuri value for example,...

CVE-2026-22682

OpenHarness prior to commit 166fcfe contains an improper access control vulnerability in built-in file tools due to inconsistent parameter handling in permission enforcement, allowing attackers who can influence agent tool execution to read arbitrary local files outside the intended repository...

CVE-2026-22682

OpenHarness CVE-2026-22682 affects the built-in file tools. The root cause is inconsistent parameter handling in permission enforcement, specifically that the path parameter is not passed to the PermissionChecker in read_file, write_file, edit_file, and notebook_edit tools. This allows attackers ...

CVE-2026-22682 OpenHarness Improper Access Control via File Tools

OpenHarness prior to commit 166fcfe contains an improper access control vulnerability in built-in file tools due to inconsistent parameter handling in permission enforcement, allowing attackers who can influence agent tool execution to read arbitrary local files outside the intended repository...

PT-2026-30930

OpenHarness prior to commit 166fcfe contains an improper access control vulnerability in built-in file tools due to inconsistent parameter handling in permission enforcement, allowing attackers who can influence agent tool execution to read arbitrary local files outside the intended repository...

Measuring the Permission Gate: A Stress-Test Evaluation of Claude Code's Auto Mode

Claude Code's auto mode is the first deployed permission system for AI coding agents, using a two-stage transcript classifier to gate dangerous tool calls. Anthropic reports a 0.4% false positive rate and 17% false negative rate on production traffic. We present the first independent evaluation o...

LFISuite

This repository is an offensive tool for Local File Inclusion LFI exploitation and scanning. It is primarily used to exploit LFI vulnerabilities in web applications, allowing an attacker to access sensitive files and potentially gain unauthorized access to a system. The tool, called LFI Suite,...

Mail.ru: Stored XSS in e.mail.ru (payload affect multiple users)

Hi, We have found a high risk level STORED XSS in e.mail.ru chat, the status change function allow to inject malicious payload in javascript & HTML, The attack affect multiple users and run in auto mode, no need a user interaction. Vulnerability affect any user that have been invited to your chat...