AI Vulnerabilities in 2025 - Some of the darker sides of AI

Generative AI is everywhere, and integrated solutions with Generative AI are implemented in your organization?
Let's look at the risks that can stem from malicious or careless use, and think about what a determined attacker could achieve.

We will examine some currently known attack vectors, discuss typical mitigation strategies, and how you can try to get some insights into the system that typically is not well documented.

Specifically let's consider the current typical set of defenses that you can add considering the base model, fine tuning or alignment, starting with the system prompt, moving on to off-the-shelf filters, and custom implementations.

But part of the conversation is also, how to find those issues through red-teaming.

After this session you should walk away with a better understanding of vulnerabilities that your AI powered systems may have, and things you can try to reduce your risk.