Vibe Coding DiscoverADR secures enterprise AI agents through observability, security benchmarking, and threat detection. Deployed at Uber.
Uber's Agentic AI Detection and Response (ADR) secures enterprise AI agents via endpoint discovery, telemetry collection, a 300+ task security benchmark, and a dual-agent detector for prompt injection and risky tool use. Open source covers Discovery, Sensor, Bench and Detector.
Use Cases
Inventory unsanctioned AI tools, CLI agents and MCP servers on employee endpointsCollect and normalize agent telemetry (intent, tool use, traces) across coding agentsBenchmark agent security with 300+ tasks and 134 MCP serversDetect prompt injection and risky agent behavior in production sessionsRed-team agents for unsafe actions before deploymentThreat detection for customer-facing support agents
Built With
- Language
- Python
- Frameworks
- MCP · Anthropic API · OpenAI API · uv
Tags
agent-security · ai-agents · prompt-injection · threat-detection · observability · benchmark · mcp · llm-security · enterprise-security · telemetry · claude-code · cursor · codex · ai-security · red-teaming · detection