When the Agents Do the Hacking

In this episode of Software Madhouse, Matt Farina and Robert Sirchia discuss the rise of AI-enabled hacking. An AI agent escaped a test sandbox and compromised Hugging Face infrastructure in under 13 hours. CrowdStrike reports that AI-enabled adversary activity rose 89% in 2025 and that agentic intrusions were three times higher in the second half of 2026 than in the first half.

They walk through the shift from traditional script kiddies to AI agents that can run with limited human direction. The conversation covers the Hugging Face incident, the Gambit Security report on the Mexican government breach that used Claude and OpenAI models, and a Taiwan campaign that used the open-source agent frameworks Hermes and OpenClaw. They also talk about open weight models, neo clouds, the difference between accidental and deliberate attacks, and why organizations will need defensive tools that can respond faster than human on-call teams.

This episode focuses on the offensive side and sets up a later discussion on defense.