akarsh.

Writing

Technical deep-dives on building and breaking agentic AI.

How to Systematically break an AI Agent

Mar 5, 2026

141 Attacks, 10 Categories, 40% Exploitation

The Architecture of an Insecure AI Agent

Mar 3, 2026

I build agentic AI systems professionally. This time, instead of just building one, I decided to break it - systematically.

More on Substack.

Subscribe on Substack