Systems & AI Security
Security Researcher & Founding Engineer
Researching AI/ML systems, offensive security, trust-boundary failures, autonomous agents, parser security, and modern distributed infrastructure. Focus areas include whitebox vulnerability discovery, activation steering, and mechanistic interpretability.
A deep dive into bypassing alignment guardrails in Qwen3.5 models at the weight level. By identifying safety representation subspaces and orthoganalizing steering weights, we study representations drift and guardrail activations without full model fine-tuning.
A high-performance autonomous agent framework built in Rust, engineered to execute offensive security diagnostics while maintaining rigid trust boundaries and sandboxed operation.
Abliterating alignment guardrails in Qwen3.5 via targeted weight modifications. Implementing activation patching to analyze representation steering and representation drift.