AI coding agents and MCP servers are quietly introducing a new class of secret leaks, and most teams don’t realize it. When an agent reads a .env file, surfaces a credential in a chat, or pulls from a config for debugging, that secret leaves the control plane. No exploit required. No alert triggered.

This session walks through three real-world leak paths: code, management interfaces, and infrastructure, and shows how AI agents bypass traditional controls across each layer. It then covers a practical framework for managing secrets in this age, including how to use Datadog to monitor and audit secret access from a centralized secrets management platform like Doppler.

Finally, this session shows how centralized secrets platforms turn potential breaches into routine rotation events, reducing incidents from days to minutes. Because in the age of AI, prevention isn’t enough. Speed of recovery is what matters.