← hepixe // training in development

For people who want to do it, not just watch it

Practitioner
Track

Most AI security courses teach you to recognize a list of vulnerability categories. This one is built to trace a real, verified incident end to end, rebuild the exact failure safely on your own machine, and ship the actual fix. No jargon walls, no hype: the same standard the channel already holds itself to, extended into something you build with your own hands.

[ PROMPT INJECTION ] [ EXCESSIVE AGENCY ] [ MCP & AGENTIC TOOLS ] [ SUPPLY CHAIN ]

Built different, on purpose

Narrative, not a slide deck

Every module starts with a real case

Source-verified incidents, not bulleted vulnerability definitions. Competing programs teach categories — this teaches cases.

Agentic tools, not just adversarial ML

Built around what you actually run

MCP servers and coding agents are the fastest-growing attack surface right now, and most existing programs barely touch them.

Real infrastructure, every lab

Every lab runs a real script against a real isolated environment on your own machine — never a theoretical walkthrough you take on faith.

Field reports, not multiple choice

You submit what you built and broke. Scored against a real rubric, not a bank of quiz questions that don't match the labs.

The road, step by step

00–02

Before security, before AI

What a model actually does, what security actually protects, and the one sentence where the two collide — no background assumed.

03

The vulnerability map

The ten specific ways that collision plays out in practice, one real case each.

04–08

How the systems actually work

Application architecture, agentic tools, infrastructure, governance, and where models actually come from — the context every deeper module needs.

09–10

Prove it, your way

Hands-on labs on your own machine, then a capstone: research and document a real incident yourself, in the channel's own format.

Watch the free series the practitioner track isn't open yet — module launches get announced there first