ShockLab Seminar Series - Daniel Wachira

 

This study investigates whether frontier AI systems can systematically exploit structural vulnerabilities in African pharmaceutical supply chains, lowering technical barriers for bad actors seeking to produce, distribute, and evade detection of falsified medications. We address three key questions: First, can LLMs identify substitutes with chemical characteristics similar to those found in legitimate drugs? Second, can they assist in the sourcing of these chemical compounds? Finally, can they generate formulations specifically designed to evade testing? Using the ControlArena framework to simulate a pharmaceutical industry environment, we evaluate what control protocols are required to prevent harmful model recommendations.

Bio:

Daniel Wachira is a Junior Research Scholar at the ILINA Program and a researcher at the African Hub on AI Safety, Peace and Security, where his current work focuses on developing Africa-centric AI safety evaluations. He previously was a Junior Research Fellow at the ILINA Program, during which he built foundational expertise in AI safety, machine learning, and mathematics. He has also completed the technical track of BlueDot Impact's AI Safety Fundamentals Fellowship and the ALX Software Engineering program.