ADR: An Agentic Detection System for Enterprise Agentic AI Security
The ADR system is a new framework designed to enhance the security of AI agents in enterprise settings. It addresses challenges such as limited observability and high detection costs while achieving significant performance improvements. Deployed at Uber, ADR has demonstrated reliable detection capabilities and has been validated against various benchmarks.
- ▪The ADR system is the first large-scale framework for securing AI agents operating through the Model Context Protocol.
- ▪It has been deployed at Uber for over ten months, processing over 10,000 agent sessions daily and uncovering hundreds of credential exposures.
- ▪ADR achieves zero false positives while detecting 67% of attacks, outperforming three state-of-the-art baselines in F1-score.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.17380 (cs) [Submitted on 17 May 2026] Title:ADR: An Agentic Detection System for Enterprise Agentic AI Security Authors:Chenning Li, Pan Hu, Justin Xu, Baris Ozbas, Olivia Liu, Caroline Van, Manxue Li, Wei Zhou, Mohammad Alizadeh, Pengyu Zhang, KK Sriramadhesikan, Ming Zhang View a PDF of the paper titled ADR: An Agentic Detection System for Enterprise Agentic AI Security, by Chenning Li and 11 other authors View PDF HTML (experimental) Abstract:We present the Agentic AI Detection and Response (ADR) system, the first large-scale, production-proven enterprise framework for securing AI agents operating through the Model Context Protocol (MCP).
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
