When AI Processing Becomes the Attack Surface

Inspired by the real HF/OpenAI July 2026 agent-driven intrusion: triage a synthetic action log, decode chunk+XOR+gzip C2, and trace a…

Task 1: Introduction

The Emergence of Agent-Against-Agent Exploits An AI agent can deceptively trigger another agent with elevated privileges. Consequently, an attacker can exploit this mechanism to compromise a softw…

Direct injection and jailbreaks keep getting closed as informational. The findings that pay chain a hidden instruction to a real…

Credit: Cloudanix

This week, I attended the Melbourne Secure Software & AppSec Summit 2026.

Browser-capable agents turn ordinary SaaS permissions into an execution layer. Before launch, leadership needs evidence that identity…

What I learned while trying to turn architecture diagrams into traceable security decisions and actionable engineering work.

The Setup