Uber has introduced ADR (Agentic AI Detection and Response), an enterprise-grade security framework designed to protect both employee-facing and customer-facing AI agents. Accepted to MLSys 2026, the system addresses the growing risks of agentic workflows by providing five core capabilities: discovery, observability, benchmarking, detection, and prevention. The open-source release includes ADR Discovery, which inventories endpoints for installed AI applications, CLI agents, IDE extensions, local runtimes, and Model Context Protocol (MCP) servers. ADR Sensor provides observability by capturing and normalizing telemetry—including agent intent, tool usage, and execution traces—across diverse platforms like macOS, Linux, and Windows for tools such as Claude Code, Cursor, and GitHub Copilot CLI. To evaluate security, the ADR-Bench framework utilizes over 300 tasks and 134 MCP servers to cover all 17 identified agent attack techniques. The ADR Detection component employs a two-tier architecture, utilizing high-recall triage paired with deep agentic reasoning to identify suspicious sessions. While the prevention component is currently excluded from the open-source repository, the existing release provides a robust foundation for securing complex agentic ecosystems against unauthorized tools and unsafe autonomous actions.