Hack the AI agent: Build agentic AI security skills with the GitHub Secure Code Game

Season 4 of the Secure Code Game introduces ProdBot, a deliberately vulnerable agentic AI assistant used inside GitHub Codespaces to teach developers how to find and fix security issues in autonomous AI systems. The course includes five progressive levels that simulate real attack surfaces (sandbox escapes, malicious web content, external tool integrations, skills and memory abuse, and multi-agent orchestration) and aims to build practical instincts for securing agentic applications; it is educational, free, and references real-world risks such as CVE-2026-25253 as illustrative examples.