Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery

arXiv:2609.09647v1 Announce Type: new Abstract: Agentic systems are rapidly moving to production, where they read untrusted inputs, call tools with real permissions, and act autonomously, expanding the security surface beyond chat-only models. Yet standard evaluations remain single-turn and fail to capture multi-step agent vulnerabilities. We present a systematic black-box framework for risk-aware agent evaluation requiring only basic system descriptions. Our approach introduces: (1) a seven-domain taxonomy mapping observable behaviors to risk categories, (2) fully automated SAGE-RT red teaming producing 120 adversarial scenarios per domain, and (3) human-validated evaluation using LLM judges. Empirical validation across two agent architectures (CrewAI and AutoGen) with four base models reveals alarming patterns: 56.25% average governance risk, 65% privacy risk in multi-agent configurations, and agent behavior vulnerabilities reaching 85%. Our black-box approach effectively identifies critical architectural vulnerabilities without privileged access, providing a scalable path toward safer agent deployments.

Related

• The Perils of Agency: How Developers Perceive, Prioritize, and Address Risks in Agentic AI Products
• Willful Disobedience: Automatically Detecting Failures in Agentic Traces
• Monitoring Agentic Systems Before They're Reliable
• Silent Failures in Multimodal Agentic Search:A Diagnostic Taxonomy and Cross-Judge Evaluation

Source: arXiv cs.AI | 2026-09-11