ADR

Overview

ADR (Agentic AI Detection and Response) is an enterprise security system for AI agents. It helps organizations secure employee-facing agents such as coding assistants and customer-facing support agents through discovery, observability, security benchmarking, and threat detection. It is deployed in production at Uber, and the accompanying paper was accepted to MLSys 2026.

Key Features

• Inventories installed AI applications, CLI agents, IDE extensions, local model runtimes, and MCP servers, and flags unknown surfaces for review.
• Captures agent intent, tool use, and execution traces across more than seven AI coding tools on macOS, Linux, and Windows.
• Provides ADR-Bench with 300+ tasks, 134 MCP servers, and coverage of all 17 agent attack techniques.
• Uses a two-tier detection architecture that combines high-recall triage with deeper agentic reasoning for suspicious sessions.
• Includes open-source ADR Discovery, ADR Sensor, ADR-Bench, and ADR Detector components.
• Supports observability for employee-facing agents such as Cursor, Claude Code, Codex, GitHub Copilot CLI, and DeepSeek Harness, as well as customer-facing AI support agents.

Use Cases

• Securing employee-facing AI coding agents and developer tools.
• Monitoring customer-facing AI support agents.
• Evaluating agent defenses against realistic enterprise attack scenarios.
• Detecting risky agent behavior in production sessions.

Getting Started

• Clone the repository, change into Detection, run uv sync, and set ANTHROPIC_API_KEY and OPENAI_API_KEY. The default detector is adr; for keyless smoke tests, use --detector llamafirewall. See the reproducibility guide for the full evaluation workflow.

Deployment & Requirements

• ADR Sensor captures telemetry on macOS, Linux, and Windows.
• The Detection quick start uses uv and requires ANTHROPIC_API_KEY and OPENAI_API_KEY for the default detector.

Before You Adopt

• License: Apache-2.0. Review its terms before using, modifying, or distributing the project.