9. Wojciech Lesicki and Andrzej Agria: Attacking and Defending LLMs in Production Environments

LLMs are going to be increasingly adopted and implemented in production environments by businesses. In this talk Andrzej and Wojtek will showcase some of the dangers associated with these systems as well as some best practices to secure them, based on compelling research by top AI labs and scientists.
In the rapidly evolving landscape of technology, Large Language Models (LLMs) are setting a new frontier for innovation, driving businesses towards their integration into production environments. However, this leap forward comes with its set of challenges and vulnerabilities. Our talk offers an in-depth exploration of the potential risks and security challenges posed by the adoption of LLMs. Drawing on cutting-edge research from top AI labs and scientists, we will unveil the hidden dangers that lurk within these systems and provide a roadmap of best practices for securing them against a spectrum of cyber threats.
Attendees will gain insights into the complexities of LLMs, including data privacy issues, vulnerability to attacks, and the ethical considerations of their use in production. Through real-world examples and research-backed strategies, we aim to equip businesses and cybersecurity professionals with the knowledge and tools needed to implement LLMs safely and responsibly. Whether you're a developer, a cybersecurity expert, or a business leader, this talk will offer critical perspectives and actionable advice to navigate the security landscape of LLMs.