Phase 5: Security Testing AI-Generated Code · 50 min · Python · AI models (Claude, GPT-4o) · prompt engineering

The Generation-Review Asymmetry

AI generates vulnerabilities 55.8% of the time but detects them 78.7% of the time in review mode. Exploit the gap.

Real-world application: Engineers who understand the generation-review asymmetry can exploit it for dual-mode security testing, catching 78.7% of vulnerabilities that AI introduces during generation.

What you will learn

Explain the generation-review asymmetry: 55.8% generation rate vs 78.7% review detection rate
Apply the dual-mode testing approach: generate code, then have the same model review it
Understand why 'write secure code' prompts only reduce vulnerabilities by 4 points (insufficient)
Quantify the gap between generation and review modes for security testing

Introduction

The Generation-Review Asymmetry

Objective

Run the generation-review asymmetry simulation with 1000 artifacts. Compare default generation, 'write secure code' prompt, dual-mode testing, and the full pipeline (review + SAST + Z3). Observe the 44-point reduction from dual-mode vs 4-point from prompt engineering.

AI generates vulnerabilities 55.8% of the time but detects them 78.7% of the time in review mode. Exploit the gap.

Unlock the full lesson

You've read the first 2 sections. The rest of this lesson covers The Problem, The Concept, Build, Check Yourself, Key Terms & Next — plus a hands-on lab, quiz, and project artifact.

Create a free account — the opening phases of 29 of 35 courses are free, no credit card.