AI Is Now Part of Your Attack Surface. Are You Testing It?

A chatbot wired into internal knowledge, a copilot that can pull customer records, an agent allowed to update systems on someone's behalf: each connection brings its own data flows, permissions and trust boundaries. The model being secure says nothing about any of that.

Secarma cover what testers actually look for: how prompt injection and manipulated inputs reach connected systems, where sensitive information leaks through AI-powered applications, and how they assess what a chatbot or agent can access, retrieve and reveal. There's specific coverage of testing permissions, actions and human approval controls.

They also explain how AI testing sits alongside application, API and infrastructure pen testing, and what to have tested before launching or widening an AI feature. Most relevant if you're building something with AI rather than just using off-the-shelf tools.

Worth saying I've attended plenty of Secarma's sessions and they stay genuinely informative rather than veering into sales.