Ziauddin Sherkar is an attorney and graduate student at the Johns Hopkins School of Advanced International Studies whose work sits at the intersection of advanced compute systems and artificial intelligence, informed by a professional background in law. At SAIS, he serves as a Research Assistant on technology and national security, supporting projects on the strategic evolution of the global semiconductor landscape, including mapping China’s semiconductor ecosystem and tracing the history of U.S. semiconductor policy. He has also served as a teaching assistant for International Economics and World Order and Disorder, and is a Frédéric Bastiat Fellow at the Mercatus Center, where he studies how emerging technologies reshape governance through the lens of the Austrian, Virginia, and Bloomington schools of political economy.

At RSA Conference, Ziauddin will present research examining the structural vulnerability frontier AI systems face from distillation-based model extraction. His work analyzes how the probabilistic outputs that enable high-performance machine learning services, such as confidence scores and soft labels, simultaneously expand the attack surface for adversaries who can convert inference access into competitive student models without access to model weights. The research explores how repeated querying and training can transfer foundational capabilities from frontier models at marginal cost, creating a strategic asymmetry between model providers and adversaries. Using Kepler, an interactive distillation attack simulator, the project evaluates real-world extraction dynamics and examines deterrence-oriented countermeasures, including adaptive output perturbations and distillation-surviving watermark signals designed to degrade unauthorized student models while preserving frontier model performance.