AI doesn’t break your security model because it’s clever. It breaks it because it’s given authority: to see data, influence decisions and take actions, often faster than any human can review.
This course teaches a complete, vendor-agnostic method for securing enterprise AI. It’s built on two frameworks I developed, the AI Control Architecture (ACA) and the Recursive Cyber Defense Model (RCDM), and on 30 years building security for some of the most regulated and complex environments there are.
You’ll follow one realistic AI system, a customer refund agent, through the whole method:
Assess: find every AI in your estate and judge its risk by what it can see, decide and do, then assign one of five risk tiers.
Control: apply the ten control pillars and turn risk into specific controls you can build on any stack.
Attack: threat model AI, red team it step by step, and test live systems safely.
Runtime: govern autonomous agents at machine speed by securing the action rather than trusting the actor, and earn autonomy safely through Shadow Mode.
Supply chain: understand what your AI is really made of, build an AI bill of materials, and run evidence-based vendor due diligence.
Prove: produce evidence that controls work, not just that they exist, and report it in terms a board understands.
Every module includes a worksheet and a quiz, and the course ends with a graded capstone. No product is required. The method works with whatever stack you run: your own code, a cloud agent service, or a SaaS product.
About AI use in this course: course materials were developed with AI assistance and reviewed and refined by the instructor. All narration is recorded by the instructor in his own voice.
