uber/ADR: Bringing the EDR Paradigm to AI Agent Security
> Source: https://github.com/uber/ADR > Paper: ADR: An Agentic Detection System for Enterprise Agentic AI Security (MLSys 2026 Industry Track)
An Overlooked Fact: AI Agents Are Already Running in Enterprises
Uber's security team discovered that the company was simultaneously running Cursor, Claude Code, Codex, internal automation agents, and customer service agents — over 7 AI coding/automation tools in total, each capable of calling MCP, reading and writing files, sending HTTP requests, and executing shell commands.
What were these agents doing? Why? Were any hijacked by prompt injection? Nobody could answer.
It's not that Uber lacked security — traditional security tools simply cannot see agents. EDR watches processes, files, and network traffic; DLP watches data exfiltration; SIEM watches logs. But agent behavior happens on a new continent: MCP calls, tool use, prompt injections in context, poisoned tool return values. No sentinels patrol that continent.
Uber's response is the open-source uber/ADR — Agentic AI Detection and Response. The name is no coincidence: it applies the EDR paradigm to the agent world.
Four Pieces: Observe, Benchmark, Detect, Prevent
1. ADR Observability — the "observability layer" for agents. Captures agent intent, tool use, and execution traces from 7+ AI coding tools on macOS/Linux/Windows. Not just "what processes the agent ran," but "what the agent intended, what MCP it called, what the execution chain looks like."
2. ADR Benchmark — a benchmark purpose-built for agent security: 300+ tasks, 133 MCP servers, covering all 17 known agent attack techniques. Not a run-once benchmark but a continuously regressing security test suite.
3. ADR Detection — a two-layer architecture: first, high-recall triage; second, deeper agentic reasoning. Triage fast and coarse, then dig deep — division of labor beats unification.
4. ADR Prevention — not open-sourced. "Stay tuned."
Key Insight: Extending "Verification as a Control Signal" from Code to Agents
ADR doesn't check logs after the agent finishes; it uses verification signals during agent execution to decide the next step. The first layer quickly filters clearly safe sessions and hands suspicious ones to the second layer for deep reasoning — isomorphic to the agent loop of "explore → verify → execute."
The engineering reality behind this choice: agent sessions are too long and complex for a single-layer detector, which would either be too slow (deep-dive everything) or too leaky (coarse-filter everything). Two-layer division is the only design that has run in production for over ten months.
ADR-Bench: 133 MCP Servers and 17 Attack Techniques
- 300+ tasks: from simple tool misuse to complex multi-step attacks
- 133 MCP servers: simulating a real enterprise MCP ecosystem (databases, file systems, APIs, internal tools)
- 17 attack techniques: prompt injection, tool poisoning, context hijacking, credential leakage, and more
- Deployment: 10+ months in Uber production
- Tool coverage: 7+ AI coding tools (Claude Code, Cursor, Codex, etc.) + internal automation + customer service agents
- Benchmark scale: 300+ tasks, 133 MCP servers, 17 attack techniques
- Academic recognition: accepted to MLSys 2026 Industry Track
- License: Apache 2.0
The number 133 wasn't arbitrary — Uber actually runs 133 MCP servers internally; that's the real scale of their production environment. The benchmark was copied from a real battlefield, not built in a lab.
Why It Matters: The "EDR Moment" for the Agent Era
In the 2010s, EDR emerged because traditional antivirus couldn't see fileless and memory attacks. The same story is repeating: traditional security can't see prompt injection, MCP call chains, or agent context hijacking.
Every computing paradigm shift creates a security blind spot, and the tools that fill it borrow paradigms from the old one. EDR filled the endpoint blind spot; ADR fills the agent blind spot.
But ADR has limits — only the Sensor, Benchmark, and Detector are open-sourced; Prevention is not. The community can observe, benchmark, and detect, but not block. Uber kept the sharpest part for itself.
The Numbers
Takeaways for Agent Developers
1. Observability first: build the "what is my agent doing" collection layer before writing the agent itself. 2. Two-layer detection: fast triage + deep reasoning; don't try to solve everything with one model. 3. Benchmarks are assets: add every newly discovered attack to the benchmark for a continuously regressing feedback loop.
---
One-line summary: uber/ADR brings the EDR paradigm to the agent world — observe, benchmark, detect, prevent. The first three pieces are open-sourced; the sharpest, prevention, is not. But those three already draw a starting line for agent security.
> Repo: uber/ADR > Paper: arXiv:2605.17380 > MLSys 2026 Slides: PDF