LIGHTNING TALK - From Prompt to Root: Anatomy of an Autonomous Hacking Campaign

LIGHTNING TALK - From Prompt to Root: Anatomy of an Autonomous Hacking Campaign

LIGHTNING TALK - From Prompt to Root: Anatomy of an Autonomous Hacking Campaign

Clifton Wolfe and David Magnotti, AWS

In September 2026, a suspected PRC-nexus actor pointed an open-source autonomous hacking agent at universities in East Asia with a simple instruction: “obtain shell access.” Against an open-source web application used at the universities, the agent needed no help as it fingerprinted an underlying framework, enumerated available APIs, and found a path traversal zero-day on its own. In a later session, the agent exploited the vulnerability to drop a Python reverse shell and execute as root. The human operator barely touched the keyboard. Amazon Threat Intelligence recovered the complete operational logs. This talk covers the zero-day and exploitation chain, our recovery of the operational history, and the greater context on autonomous agents outpacing traditional response workflows built around human approval.