As you scale the use of generated AI across your company, robust guardrails are essential to mitigate risk and promote responsible and ethical operation of your LLM, thereby maintaining trust and supporting innovation.
To meet these needs, JPMorganChase developed the Fence guardrail framework. Fence uses data-driven methodologies to proactively identify, test, and mitigate vulnerabilities such as hallucinations, topic drift, and prompt injections at the individual use case level, strengthening the security and reliability of AI solutions.
Fence’s core strength is its ability to adapt to JPMorganChase’s dynamic needs and leverage synthetic data generation to provide custom guardrails specific to its use case. This broadly applicable approach is a step forward for the industry.
From simple Q&A to advanced search, Fence already provides secure and robust interactions with AI-powered tools and processes to improve the daily experience for employees and customers. Internal benchmarks show that Fence enhances the security and reliability of LLM applications and outperforms existing industry solutions.
As AI evolves, continued innovation in guardrail design will be essential to maintaining trust and compliance in financial services.
For more information about the Machine Learning Center of Excellence, visit jpmorganchase.com/mlcoe.
