AI Fight Club
Offense vs Defense — talk edition
33 / 65 confirmed
About
Two halves, two perspectives. One night.
Amir on offense.
AI tooling for offensive security has been moving fast and most of the takes online are either hype or dismissal. This half walks through where it actually finds bugs, where it just burns tokens, and the scaffolding that gets any frontier model behaving the same way. Real cost numbers, the patterns Amir has been using week to week, and an honest answer to when you should put the LLM down and read the code yourself.
KC on defense.
Most blue-team tooling was built on the assumption that a human reads the output. AI breaks that assumption, and once you accept it, every layer of the SOC needs redesigning.
KC's half is a demo of his own AI-native breach detector running against real Wizard Spider attack data, and a look at the principles he has pulled out of building it. What changes when the consumer is an agent, not an analyst.