[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"readme:adr":3},"\u003Ch1>ADR: Agentic AI Detection and Response\u003C\u002Fh1>\n\u003Cp>\u003Ca href=\"https:\u002F\u002Ftrendshift.io\u002Frepositories\u002F100696?utm_source=repository-badge&utm_medium=badge&utm_campaign=badge-repository-100696\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Ftrendshift.io\u002Fapi\u002Fbadge\u002Frepositories\u002F100696\" alt=\"uber\u002FADR | Trendshift\" \u002F>\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>ADR (Agentic AI Detection and Response) is an enterprise security system for AI agents. It helps organizations secure employee-facing agents such as Cursor, Claude Code, Codex, GitHub Copilot CLI, and DeepSeek Harness, as well as customer-facing agents such as AI support agents.\u003C\u002Fp>\n\u003Cp>ADR is \u003Cstrong>deployed in production at Uber\u003C\u002Fstrong>, and the accompanying paper was accepted to \u003Cstrong>MLSys 2026\u003C\u002Fstrong>: \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fuber\u002Fadr\u002Fblob\u002FHEAD\u002Fdocs\u002Fadr-paper.pdf\" rel=\"nofollow ugc noopener\">Paper PDF\u003C\u002Fa> · \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fuber\u002Fadr\u002Fblob\u002FHEAD\u002Fdocs\u002Fadr-mlsys-2026-slides.pdf\" rel=\"nofollow ugc noopener\">Slides PDF\u003C\u002Fa>\u003C\u002Fp>\n\u003Ch2>How ADR secures enterprise AI agents\u003C\u002Fh2>\n\u003Cp>ADR secures enterprise AI agents through five complementary capabilities: discovering unsanctioned AI tools, observing agent activity, evaluating defenses, detecting threats, and preventing unsafe actions.\u003C\u002Fp>\n\u003Col>\n\u003Cli>\u003Cstrong>ADR Discovery: Find the AI tools present on employee endpoints.\u003C\u002Fstrong> Inventories installed AI applications, CLI agents, IDE extensions, local model runtimes, and MCP servers, and flags unknown surfaces for review.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>ADR Observability: Understand what AI agents are doing and why.\u003C\u002Fstrong> In production, ADR captures agent intent, tool use, and execution traces across 7+ AI coding tools on macOS, Linux, and Windows, as well as internal automation and customer-facing support agents.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>ADR Benchmark: Test agent security under realistic enterprise conditions.\u003C\u002Fstrong> ADR-Bench includes 300+ tasks, 134 MCP servers, and coverage of all 17 agent attack techniques.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>ADR Detection: Detect risky agent behavior efficiently.\u003C\u002Fstrong> Its two-tier architecture combines high-recall triage with deeper agentic reasoning for suspicious sessions.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>ADR Prevention: Stop unsafe actions before they cause harm.\u003C\u002Fstrong> This component is not included in the current open-source release. \u003Cstrong>Stay tuned.\u003C\u002Fstrong>\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Ch2>Repository layout\u003C\u002Fh2>\n\u003Cp>This repository contains the open-source \u003Cstrong>ADR Discovery\u003C\u002Fstrong>, \u003Cstrong>ADR Sensor\u003C\u002Fstrong>, \u003Cstrong>ADR-Bench\u003C\u002Fstrong>, and \u003Cstrong>ADR Detector\u003C\u002Fstrong> described in the paper. The offline \u003Cstrong>ADR Explorer\u003C\u002Fstrong> engine, which hardens ADR Detection through pre-deployment red teaming, is not included here.\u003C\u002Fp>\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Path\u003C\u002Fth>\n\u003Cth>ADR component\u003C\u002Fth>\n\u003Cth>Description\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\u003Ctr>\n\u003Ctd>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fuber\u002Fadr\u002Fblob\u002FHEAD\u002FDiscovery\u002F\" rel=\"nofollow ugc noopener\">Discovery\u002F\u003C\u002Fa>\u003C\u002Ftd>\n\u003Ctd>ADR Discovery\u003C\u002Ftd>\n\u003Ctd>Inventory the AI apps, CLI agents, IDE extensions, model runtimes, and MCP servers on an endpoint, and flag unknown surfaces for review\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fuber\u002Fadr\u002Fblob\u002FHEAD\u002FSensor\u002F\" rel=\"nofollow ugc noopener\">Sensor\u002F\u003C\u002Fa>\u003C\u002Ftd>\n\u003Ctd>ADR Observability\u003C\u002Ftd>\n\u003Ctd>Collect and normalize agent telemetry from Claude Code, Cursor, Codex, GitHub Copilot CLI, DeepSeek Harness, opencode, Claude Desktop, and others\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fuber\u002Fadr\u002Fblob\u002FHEAD\u002FDetection\u002F\" rel=\"nofollow ugc noopener\">Detection\u002F\u003C\u002Fa>\u003C\u002Ftd>\n\u003Ctd>ADR Benchmark + Detection\u003C\u002Ftd>\n\u003Ctd>Dual-agent detector, 134 MCP servers, 304 benchmark tasks, baselines, figure scripts\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fuber\u002Fadr\u002Fblob\u002FHEAD\u002Fdocs\u002FREPRODUCIBILITY.md\" rel=\"nofollow ugc noopener\">docs\u002FREPRODUCIBILITY.md\u003C\u002Fa>\u003C\u002Ftd>\n\u003Ctd>Evaluation\u003C\u002Ftd>\n\u003Ctd>Step-by-step workflow to reproduce benchmark detection and paper figures\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\u003C\u002Ftable>\n\u003Ch2>Quick start: ADR Detection\u003C\u002Fh2>\n\u003Cpre>\u003Ccode class=\"language-bash\">git clone https:\u002F\u002Fgithub.com\u002Fuber\u002FADR\ncd ADR\u002FDetection\nuv sync\nexport ANTHROPIC_API_KEY=\"...\" OPENAI_API_KEY=\"...\"\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Default detector is \u003Ccode>adr\u003C\u002Fcode> (ADR dual-agent). For keyless smoke tests, use \u003Ccode>--detector llamafirewall\u003C\u002Fcode> (see \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fuber\u002Fadr\u002Fblob\u002FHEAD\u002FDetection\u002FREADME.md\" rel=\"nofollow ugc noopener\">Detection\u002FREADME.md\u003C\u002Fa>).\u003C\u002Fp>\n\u003Cp>See \u003Cstrong>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fuber\u002Fadr\u002Fblob\u002FHEAD\u002Fdocs\u002FREPRODUCIBILITY.md\" rel=\"nofollow ugc noopener\">docs\u002FREPRODUCIBILITY.md\u003C\u002Fa>\u003C\u002Fstrong> for the full evaluation workflow (inflate packed benchmark → run detectors → plot figures).\u003C\u002Fp>\n\u003Cp>Component documentation:\u003C\u002Fp>\n\u003Cp>ADR Sensor also captures \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fuber\u002Fadr\u002Fblob\u002FHEAD\u002FSensor\u002FREADME.md#gemini-cli\" rel=\"nofollow ugc noopener\">Gemini CLI session journals\u003C\u002Fa>\non macOS, Linux, and Windows, including tool results and nested s\u003C\u002Fp>\n",1793926088846]