joshua

Dias

I help AI companies and LLM developers find security vulnerabilities before attackers do — specifically prompt injection, jailbreak attempts, and data leakage. I've built my own AI red teaming framework with over 2,000 attack techniques based on research by leading labs like NVIDIA and Robust Intelligence[citation:3]. I know how models break because I've tested them — across TinyLlama, Gemma, and Phi models. What you get: a clear CSV report showing exactly where your LLM is vulnerable, with severity ratings and actionable recommendations. Message me with your model details, and I'll send a custom quote