ADR
Detection, telemetry, and benchmarking for securing enterprise AI agents
What it is
A security system from Uber for watching and defending AI agents used inside a company, from coding assistants like Claude Code and Cursor to customer support bots. It inventories AI tools and MCP servers on employee machines, collects normalized agent telemetry, and runs a two-tier detector that flags risky sessions such as prompt injection. It also ships ADR-Bench, a set of 300+ attack tasks across 134 MCP servers. Security teams and researchers use it to monitor agent activity and measure how well defenses hold up.