AI is giving threat hunters the possibility to investigate things that, until now, were simply too slow, too manual or almost impossible to do at scale.

In this session, I will present several real threat hunting scenarios where AI becomes part of the investigation: analyzing large amounts of security telemetry, understanding context, correlating different signals, inspecting content and helping the hunter go beyond what a traditional query or detection rule can easily provide.

But there is another side that interests me even more: what happens when the attacker uses AI?

I will also demonstrate different attack scenarios built around AI, using mailboxes and websites as targets and attack vectors, including malicious content designed not only for the user, but for the AI systems and agents processing that content.

This creates some interesting questions for threat hunters. What happens when an email looks harmless to the user but contains instructions for an AI agent? What happens when a trusted agent visits a malicious website? Can an attacker manipulate the context received by an agent, influence its decisions, or abuse the tools and permissions available to it?

We will look at both sides through practical examples: using AI to hunt threats, hunting attacks performed with AI, and identifying abuse against AI agents themselves.

This is a technical threat hunting session based on practical scenarios, experiments and things I have been testing.

Expect KQL, security telemetry, AI-assisted investigations, malicious emails, websites, prompt injection, agent abuse and some scenarios that would have required a painful amount of manual work not long ago.

The goal is simple: understand what AI actually changes for us as threat hunters—and what new things we now need to start looking for.