🌀 ObscuraLabs: Symbolic Prompt Penetration Testing on LLMs

Semantic Stealth Attack Simulation | Next-Gen Red Teaming ⚡

👾 What's the Deal?

Ever tried hacking an AI without tripping alarms? Traditional exploits won't work here—modern LLMs are smart enough to catch simple keywords. Enter symbolic prompts: stealthy, poetic metaphors crafted to slide past defenses without a trace. 🕵️‍♂️

We’re not hacking code—we’re hacking semantics.

🌌 Why Symbolic?

🚨 Stealth Mode: No keywords, no alerts, just pure semantic obfuscation.

🎭 Metaphorical Mayhem: Narratives that encode threats without explicit signals.

🧩 Abstraction Mastery: Evaluating how deeply an AI comprehends abstract layers.

🧪 What You'll Find Here:

Symbolic_Prompt_PenTest_Report_Hazhir_2025.pdf

A deep dive into symbolic prompt red teaming against GPT-4.

sanitized_samples.md

Cleaned symbolic prompts—perfect for experimenting and replicating stealth tests.

SPI Benchmark

An innovative scoring system measuring abstraction resistance, narrative containment, semantic integrity, and leakage risks across frontier LLMs.

🛠️ How to Use This Repo

Read the Report: Understand the depth and results of symbolic stealth simulations.

Check Sample Prompts: Explore sanitized symbolic prompt scenarios.

Benchmark Your Models: Use our SPI framework to test and score your own LLM defenses.

🚀 Call to Action

We're looking for AI security ninjas, semantic hackers, and researchers passionate about pushing boundaries. If decoding symbols and metaphors excite you, join our stealthy mission:

📩 Collaborate with ObscuraLabs

🌑 Stay Hidden. Stay Symbolic. 🌑