ADR

Why we included this project

ADR is the security system Uber built to watch its own AI agents, and this release makes the core pieces open source. The sensor normalizes telemetry from coding tools like Claude Code, Cursor, and Codex, the benchmark covers hundreds of tasks across 133 MCP servers, and the two-tier detector triages sessions cheaply before running deeper reasoning on the suspicious ones. That means you can see what your agents are doing in production and test your defenses against realistic attack techniques, rather than treating security as a one-off audit. The prevention layer that blocks unsafe actions is not in this release, so teams will need to bring their own enforcement. If you run agents at scale and want visibility into their behavior, this is a strong place to start.