If this, then breach - AI-Assisted red teaming without malware, infrastructure, or exploits

A lot of modern environments don’t need to be "broken into" in the traditional sense anymore. Between SaaS platforms, identity providers, and automation tooling, there’s already enough capability inside most organisations to cause serious impact—if you know how to use it.

This presentation looks at how AI changes the way attackers approach that problem.

Rather than relying on malware or command-and-control infrastructure, we explore what it looks like to operate entirely within trusted systems—using OAuth access, APIs, and built-in workflows. The difference is that AI makes it much easier to understand unfamiliar environments, map relationships between systems, and identify the small gaps that can be chained together into something meaningful.

We’ll walk through a realistic attack scenarios starting with deepfake cloning, obtaining a low-privileged foothold and pivoting to the crown jewels. Along the way, AI is used to help enumerate SaaS integrations, interpret configuration data, generate plausible automation logic, and even assist in crafting actions that blend into normal business processes. The end result is access to high-impact access without ever deploying a payload or touching an endpoint.

We’ll also look at why this is difficult to detect. From a logging perspective, everything appears legitimate: approved apps, normal API calls, standard automation behaviour. AI doesn’t introduce new vulnerabilities here—it just makes it easier to take advantage of the ones that already exist.

The aim of the session is to show how AI is quietly lowering the barrier to this kind of tradecraft, and why organisations that focus only on endpoints and malware are increasingly missing the real attack surface.