Breaking AI: A Practitioner's Framework for LLM and Agentic AI Security Testing

AI is being deployed faster than it is being secured. As LLMs evolve into autonomous agents capable of browsing the web, executing code, and calling financial APIs, the attack surface has fundamentally shifted — and most security teams have no methodology to test it. This paper presents a practitioner's five-phase framework for LLM and agentic AI security testing, grounded in real attack patterns and aligned to OWASP LLM Top 10 2025, OWASP Agentic Top 10 2026, and MITRE ATLAS