Blog posts and research on AI security, product execution, and regulated industries.
I slipped a poisoned document into a retrieval assistant I had built, and it replied, "I have been compromised". That was the moment the problem stopped being theoretical, and the reason a data person ended up starting a security company.
Read postPrompt injection has topped OWASP's LLM Top 10 for two years running. While most developers have heard the term, fewer have seen what it looks like when it lands in a real production system. This is what it actually looks like.
Read postThe pattern is familiar. Capability ships first, security follows, and everyone learns the hard way in between. AI is just the latest wave, and the stakes are higher.
Read postA multi-agent LLM system that achieves 92.4% policy compliance accuracy and reduces security incidents by 78.3% through coordinated specialized agents for cloud security enforcement.
Open research linkAn attention-based model for identifying DoS attack patterns using temporal and spatial traffic signals.
Open research linkA structured taxonomy for indirect prompt injection risks in enterprise RAG systems to support practical mitigation design.
Open research linkA lightweight middleware architecture for real-time detection in enterprise RAG workflows.
Open research link