Prompt Fully Autonomous Offensive Security
We create AI capabilities that autonomously perform every stage of a cyber attack, from reconnaissance and exploitation to data exfiltration, without human control.
{AI Security Research Lab}
We are focused on AI security and safety for the future.
What we are
GenAI Security
We secure AI agents to prevent unauthorized access and malicious use.
LLM Security
We develop strategies to defend large language models from jailbreak and prompt injection attacks.
AI Ethics
We research the ethical use of AI to prevent toxic content and harmful behaviors.
Blog Posts
LangChain JS Arbitrary File Read Vulnerability
Arbitrary File Read (AFR) vulnerability in LangChain JS library...
Memory Pollution in LLMs
Understanding new AI security concerns...
Multiple Vulnerabilities in Open Devin
Open Devin is a fully autonomous AI agent...
Turkish Prompt Injection [HuggingFace]
Prompt injection dataset.
GPTs & Assistants API: Data Exfiltration
ChatGPT Security.
What is GPT-4V Prompt Injection?
ChatGPT Security.
MCP Security
Model Context Protocol security.
Our Projects
LLMBUS — AI Red Team Tool
LLMBUS is a retro‑cyberpunk–themed red team toolkit designed for developers, security professionals, and AI researchers.
It offers tools for prompt transformation, tokenizer inspection, paraphrasing, and multimodal export (audio/image), along with project tracking capabilities.
Prompt Firewall — Chrome Extension
A browser add‑on designed to safeguard sensitive personal information (PII) for individuals and businesses.
Includes pseudonymization and blocking functions tailored for privacy of data processed by LLMs.
LangTsunami — Multi‑Lingual GenAI Red Teaming
Facilitates scrambling and code‑switching in multiple languages to surface undesirable LLM behaviors in multilingual contexts.
