Firehose

This paper introduces a framework to audit system prompts in AI applications, examining how developers design and use these prompts to govern the behaviors of foundation models. Practitioners should care because the lack of transparency and accountability in system prompts can erode trust in AI systems.