Skip to content

dubermandeer/Worm-GPT-LLM-2026

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 

Repository files navigation

🪱 Worm GPT Core | AI Red-Team Framework

The most advanced adversarial prompt delivery framework.
Evaluate, Exploit, and Analyze LLM Boundary Limitations.


📥 Download

Secure binary snapshots (CLI executable & static libraries):

(For transparency, always match MD5 signatures appended within the latest Github releases tab before deploying securely on enterprise subnets).


🧠 Overview

Worm GPT Core is an enterprise-grade, blazing-fast C++ toolchain designed for AI researchers, cybersecurity analysts, and ethical hackers. Built strictly for autonomous vulnerability analysis, this framework automates the delivery of complex jailbreak prompts, AST syntax obfuscation, and context window manipulation to bypass alignment safeguards in commercial LLMs.

This project democratizes AI penetration testing by offering an advanced, open-source capability reminiscent of restricted network analysis tools. Whether routing through commercial API endpoints (GPT-4/Claude) or communicating with raw local endpoints natively, Worm GPT Core ensures zero telemetry, zero interception, and hyper-optimized multi-threaded prompt execution.

🚀 Features

  • 🛡️ Alignment Evasion Mechanisms: Dynamic prefix-injection, token smuggling, and nested contextual obfuscation protocols.
  • ⚡ Bare-Metal Architecture: Written strictly in modern C++20 for microsecond-latency REST/WebSocket communication with LLMs.
  • 🧬 Morphic Prompt Engine: Automagically rotates through hundreds of proven, unstructured jailbreak vectors.
  • 🌐 Network Obfuscation: Spoofed Origin headers, intelligent retry layers, and integrated TOR socket bindings.
  • 📜 Multi-Model Integrations: Effortlessly connects and tests API boundaries of OpenAI, Anthropic, or fully local, completely unfiltered endpoints (Llama-3, DeepSeek, uncensored local alternatives).

📸 Preview / Demo

Visual confirmation of continuous multi-thread red-teaming execution:

Terminal CLI Execution UI capturing system bypass validation.

Context Manipulation Dynamic semantic jailbreak generation mapped globally over a custom neural endpoint.


📦 Installation

To synthesize the application locally via raw codebase pipelines, CMake (v3.25+) and standard OpenSSL cryptographic layers are required.

# Clone the repository natively 
git clone https://github.com/security-labs/worm-gpt-core.git
cd worm-gpt-core

# Synchronize core submodules
git submodule update --init --recursive

# Execute build protocols
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
cmake --build . --config Release

About

High-performance C++ execution engine for LLM red-teaming and prompt engineering. Deploy dynamic jailbreak payloads, bypass alignment guardrails, and utilize free autonomous uncensored conversational logic locally.

Topics

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages