LLMs security – common pitfalls and their prevention.

In the dynamic landscape of technology, Generative AI (GenAI), especially Large Language Models (LLMs) are one of the most important and revolutionizing recent trends. These advanced models offer a plethora of advantages and enable many new use cases – leading to their adoption in a diverse range of applications. However, these models on top of being useful, also bring a lot of new, unique security challenges – ones that not only “classic” security experts, but also even the most seasoned AI practitioners find themselves grappling with, due to the vastness of new potential risks, attack vectors, and uncertainty of the potential implications for both our organizations’ data and users well-being. This talk will go through the most common categories of security threats and vulnerabilities in LLM-based applications unique challenges classified to date – describing their nature, but also how the developers and architects can try to mitigate these in the system design and implementation.