Agentjacking: The New AI Coding Attack Developers Are Missing

AI coding agents can read issues, inspect repos, run tools, open network connections, and touch CI/CD workflows. That makes them powerful. It also creates a new attack surface: Agentjacking. This video breaks down how malicious instructions hidden inside issues, pull requests, docs, logs, or tool metadata can hijack a coding agent, why normal security tools may miss it, and the five defenses developers should set up before letting Claude Code, Cursor, Codex, or Gemini CLI touch real production repositories. You’ll learn: How Agentjacking works Why prompt injection changes when agents have tools Why “authorized” actions can still be dangerous The five practical defenses developers can apply now #Agentjacking #AIAgents #AICoding #ClaudeCode #CursorAI #OpenAICodex #AISecurity #PromptInjection #DeveloperSecurity #SoftwareEngineering #AIEngineering #GenAIGeek