Extended Cyber Kill Chain for AI-Era Threats (github.com)

A significant update to the Lockheed Martin Cyber Kill Chain framework has been announced, designed to enhance defenders' responses to threats posed by large language models (LLMs) and agentic AI. This new version introduces a crucial pre-attack stage focused on model supply chain compromises, addressing the emerging tactics that adversaries leverage to exploit AI systems. Furthermore, the existing seven stages of the framework have been augmented with AI-specific sub-techniques, and the "Actions on Objectives" stage has been split into three distinct peer sub-stages: classical data exfiltration, model extraction, and agentic pivoting. This extension of the Cyber Kill Chain is vital for the AI/ML community as it adapts traditional cybersecurity strategies to the unique challenges posed by AI technologies. By addressing threats like model poisoning and indirect prompt injection, the update provides frameworks for detection and mitigation tailored to the complexities of AI systems. The document serves as a practical guide for security operations center (SOC) analysts and detection engineers, allowing them to reference new detection signals and mitigation strategies directly tied to newly identified sub-techniques. This makes it an essential resource for combating sophisticated AI-era threats, reinforcing the importance of proactive security measures in an increasingly AI-driven landscape.