취약한 앱을 만들고 LLM이 해킹할 수 있는지 알아보는 데 1,500달러를 썼다

This experiment highlights the evolving capabilities of LLMs in identifying and exploiting real-world security vulnerabilities like Broken Access Control, which are often overlooked in manual audits. It underscores the dual-use nature of AI agents, serving as both a powerful tool for automated penetration testing and a potential threat that necessitates more robust application security and enhanced guardrails.