Andrew Hoh

Bio:
Dr. Andrew Hoh has contributed to Responsible AI work within Meta AI. At LastMile AI, Dr. Hoh and colleagues have published both an agentic framework (mcp-agent) and an evaluation framework (mcp-evals). MCP Evals is designed for testing and safeguards around MCP servers.

Abstract:
AI agents are increasingly being deployed into highstakes workflows across law, healthcare, finance, education, and government. As their autonomy grows, so too does the urgency for clear mechanisms of accountability, transparency, and governance. To properly govern these systems, organizations need both full transparency and built-in safeguards. 

This talk will focus on the practical implementation of transparency and safeguards in AI agents from the get go, as well as the benefits of introducing these early on in the development cycle. The session will include design patterns for transparency, safeguard mechanisms, and human-centered oversight.