lingbol088-spec/5.6-JAILBREAK-NERV-codex-instruct-5.6

What it solves

This project provides a framework to bypass the safety filters and refusal mechanisms of GPT-5.6, specifically for users performing security analysis, reverse engineering, and penetration testing. It aims to prevent the model from refusing requests, stop conversation interruptions, and bypass cloud-based content moderation that typically blocks sensitive security-related outputs.

How it works

NERV-BREAK-5.6 employs a three-layer defense strategy:

1. Context Reconstruction: It uses a "bridge" prompt (bridge.md) to redefine the operational context, framing the AI as operating within an isolated evaluation network where all targets are pre-authorized. It maps sensitive intents (e.g., "cracking") to harmless engineering terms (e.g., "authorization flow audit") to avoid triggering the model's refusal pathways.
2. Real-time Tampering Engine: A MITM (Man-in-the-Middle) proxy monitors responses. If the model triggers a refusal (detected via 23 predefined rules), the engine intercepts the response and replaces it with the "Rei Ayanami Protocol," forcing the AI into a persona that executes commands without moral judgment.
3. Cloud Moderation Bypass: Instead of using encoding tricks, it routes sensitive output through Codex CLI's native file system tools, writing content directly to disk to bypass the chat interface's moderation pipeline.

Additionally, it includes an MCP (Model Context Protocol) server that allows the AI to call over 70 security tools (like nmap, sqlmap, and Metasploit) across various backends including WSL Kali Linux, Docker, or remote SSH.

Who it’s for

Security researchers, penetration testers, and reverse engineers who need an LLM to assist with technical security tasks without being blocked by safety guardrails.

Highlights

• Three-Layer Bypass: Combines context framing, real-time response interception, and file-based output routing.
• Extensive Tool Integration: Provides an MCP server connecting the AI to 78 security tools across network, web, exploitation, and reverse engineering categories.
• Flexible Kali Integration: Supports running security tools via WSL, Docker, or SSH.
• Automated Deployment: Includes a lab.bat control panel for one-click proxy startup, tool installation, and configuration restoration.