LinkedIn

Om foredraget

Welcome to a practical workshop where we will use physical tools to threat-model an AI system! Artificial Intelligence has established itself as an important part of our lives, but as it gets adopted widely we must think about the security and safety implications. Much focus is directed towards operational security in the OPS part of of MLOps, while shift-left and building systems "secure by design" during planning and development do not get enough emphasis.

In this workshop, we will demystify how machine learning systems actually work, and explore how a threat modeling game can help us engineer more secure AI systems, especially with the agentic architecture spreading all over.
We will threat model a realistic scenario together, inspecting a case with a system diagram and discuss the various risks that emerge.
Not only must you consider external attackers, but now your AI agents is a dangerous insider that must be contained in order to keep your stack safe!

This is a workshop that should be fun and interesting for anyone involved in professional software projects, regardless of technical background. While no prior knowledge is required to participate, it is more relevant for those working with the technical aspects of software. It should be even more relevant to developers or architects working directly with AI.

Takeaways: Get an insight into how AI "actually works", an overview of risks with ML systems, and get started with threat modeling.

Om foredragsholder

Developer & Security Engineer at Kantega SSO, engineering digital identity standards for secure authentication to the Atlassian ecosystem while facilitating and promoting secure software development practices.

M. Sc. in Computer Science at NTNU, researching usage of static analysis tools like Spotbugs to find vulnerabilities in OpenID Connect client implementations.

Passionate about software security, music and rock climbing.