PublishedEcosystem tierLearn
Basileak
Intentionally vulnerable LLM fine-tuned on Falcon 7B for red team training and AI security research.
Overview
Basileak is the entry point of the Black Unicorn LLM Security Lifecycle. It is a deliberately insecure large language model, fine-tuned on Falcon 7B to exhibit real-world LLM vulnerabilities in a controlled environment.
Built for red teamers, security researchers, and CTF players, Basileak exposes exploitable behaviors across prompt injection, data leakage, jailbreaks, and unsafe output generation. Every vulnerability is intentional and documented, making it the ideal training target for anyone learning LLM offensive security.
Hosted freely on HuggingFace, Basileak pairs directly with DojoLM for structured attack testing. The philosophy is simple: learn to break models before you learn to defend them.
Tech Stack
PythonFalcon 7BHuggingFacePEFTLoRA
Key Features
Fine-tuned Falcon 7B with intentional vulnerabilities
Exploitable prompt injection vectors
Data leakage and exfiltration behaviors
Jailbreak-susceptible safety alignment
Unsafe output generation scenarios
System prompt extraction weaknesses
Free and open on HuggingFace
Pairs with DojoLM for structured testing
Ideal for CTF competitions and training labs
Documented vulnerability taxonomy
Presentation
Powering Our Services
Basileak is used directly in the following services:
LLM & AI Security
Comprehensive security assessments for LLM-powered applications. From prompt injection testing and AI agent security to multi-agent operations and custom model hardening.
Penetration Testing & Red Teaming
Offensive security services that identify real vulnerabilities before attackers do. Web apps, APIs, infrastructure, and social engineering.