Field Note #009: The Asymmetric Exposure
This field note explores asymmetric exposure, where one party bears greater risk than another in a system, and how such imbalances can lead to exploitation and fragility. It argues that correcting these exposures is key to building fairer, more resilient systems.
Background
Azimuth is a startup working on "AI verification" — building tools to check whether an AI system's behavior matches its claimed capabilities and safety properties. This is from their internal "Field Notes" series, which shares thinking-in-progress rather than polished product announcements. The post coins a concept called "The Asymmetric Exposure" to describe a structural problem: those who build and deploy AI systems have far less to lose from a system's failure than the people and societies that are exposed to its consequences. This imbalance, the authors argue, undermines standard approaches to safety assurance (like stress-testing or red-teaming) because the incentives to find and fix problems are misaligned. The piece connects to a broader debate in AI governance about who bears risk, the limits of voluntary safety measures, and whether regulation or liability frameworks are needed to rebalance exposure.