With high-end computers for hire that have never been cheaper and the wide availability of machine learning tools, the topic of AI has been in everyone's news feed for quite some time now. But even during the 1960s when the term “AI” was first coined, the risks were already being described on what its potential could be when it comes to its use, but also its abuse. Whether you are leveraging AI for your in-house chat-bot to help with on-boarding, or whether you are working on text or image generation algorithms: in the cat and mouse game of attack versus defense, there will always be the potential of subverting the model, extracting training data or abusing the technology for personal gain and profit.

This talk will cover the risks that come with starting out with any form of AI, what you should focus on and how to deal with compliance and threat modeling. The presentation will include examples of real-life AI attacks, how they work as well as what can be done to mitigate them.