Explore the world of machine learning model duplication attacks in this conference talk from Hack In The Box Security Conference. Dive into the vulnerabilities of cloud-deployed ML models and learn about a novel approach called GDALR (Gradient Driven Adaptive Learning Rate) for more efficient model stealing. Discover how attackers can exploit MLaaS (Machine Learning as a Service) platforms to clone black box models, potentially undermining business models built around proprietary ML products. Examine the mathematical modifications to current attack methodologies and their implications for MLaaS security. Gain insights into experimental setups, logistic regression, and multi-layer perceptrons (MLPs) as they relate to model duplication. Understand the urgent need for improved countermeasures in the face of these sophisticated attacks, and consider the future of MLaaS security in light of this research.