uber/ADR

ADR secures enterprise AI agents through observability, security benchmarking, and threat detection. Deployed at Uber.

What it solves

ADR provides an enterprise security system designed to protect organizations from the risks associated with AI agents. It addresses the need to monitor and secure both employee-facing tools (like Cursor and Claude Code) and customer-facing support agents, preventing unsafe actions and detecting threats in real-time.

How it works

ADR uses a four-part security framework:

  • Observability: A sensor collects and normalizes telemetry—including agent intent, tool use, and execution traces—across various operating systems and AI coding tools.
  • Benchmarking: The ADR-Bench provides over 300 tasks and 133 MCP servers to test agent security against 17 different attack techniques.
  • Detection: A two-tier architecture uses high-recall triage followed by deeper agentic reasoning to identify suspicious sessions efficiently.
  • Prevention: A component designed to stop unsafe actions (currently not in the open-source release).

Who it’s for

Enterprise security teams and AI researchers who need to monitor, evaluate, and secure AI agents deployed in production environments.

Highlights

  • Production Proven: Deployed at Uber and presented at MLSys 2026.
  • Broad Compatibility: Supports telemetry collection from tools like Claude Code, Cursor, Codex, and Claude Desktop.
  • Comprehensive Testing: Includes a benchmark with 303 tasks covering all 17 known agent attack techniques.
  • Dual-Agent Detection: Employs a sophisticated detection system to identify risky behavior through a triage-and-reasoning approach.

Related

  • Project
    secureagentics/Adrian

    An open-source runtime security engine for AI agents that monitors both actions and reasoning traces to detect and block malicious or out-of-remit behavior in real time.

  • Project
    cordum-io/cordum

    An agent control plane for governance, safety, and trust that provides a deterministic layer to define, enforce, and audit the behavior of autonomous AI agents.

  • Project
    affaan-m/agentshield

    A security auditor for AI agent configurations that scans Claude Code setups for secrets, permission misconfigurations, and prompt injection vectors.

  • Project
    rogue-security/rogue

    An AI agent evaluator and red teaming platform that stress-tests agents against business policies and adversarial attacks to identify security vulnerabilities.

  • Project
    snyk/agent-scan

    A CLI security scanner that discovers AI agents (Claude, Cursor, etc.), their MCP servers, and skills on your machine, and checks them for prompt injection, malicious code, and other vulnerabilities.