AI RESEARCH
MATRA: Modeling the Attack Surface of Agentic AI Systems -- OpenClaw Case Study
arXiv CS.AI
•
ArXi:2605.10763v1 Announce Type: new LLMs are increasingly deployed as autonomous agents with access to tools, databases, and external services, yet practitioners (across different sectors) lack systematic methods to assess how known threat classes translate into concrete risks within a specific agentic deployment. We present MATRA, a pragmatic threat modeling framework for agentic AI systems that adapts established risk assessment methodology to systematically assess how known LLM threats translate into deployment-specific risks.