As we expand the use of generative AI across our firm, robust guardrails are essential to help mitigate risk and promote the responsible and ethical operation of LLMs, thereby maintaining trust and supporting innovation.

To help meet these needs, JPMorganChase developed the Fence guardrail framework. Fence uses data-driven methods to proactively identify, test and mitigate vulnerabilities—such as hallucinations, topic drift, and prompt injection—at the individual use case level, reinforcing the security and reliability of our AI solutions.

A core strength of Fence is its ability adapt to the dynamic needs of JPMorganChase, our teams and clients leveraging synthetic data generation to deliver custom, use case–specific guardrails. This broadly applicable approach is a step forward for the industry.

From simple Q&A to advanced search, Fence is already enhancing daily employee and customer experiences by providing safe, robust interactions with AI-powered tools and processes. Internal benchmarks show that Fence is enhancing the safety and reliability of LLM applications, surpassing existing industry solutions.

As AI evolves, continuous innovation in guardrail design will be essential for maintaining trust and compliance in financial services.

To read more about our Machine Learning Center of Excellence, visit jpmorganchase.com/mlcoe.