LibVulnWatch

For those deeply interested in the architecture, oversight and governance of generative/agentic AI systems, here are three rigorous, publicly-accessible papers that combine technical depth with policy implications: 1. TRiSM for Agentic AI: A Review of Trust, Risk, and Security Management in LLM-based Agentic Multi-Agent Systems Shaina Raza, Ranjan Sapkota, Manoj Karkee & Christos Emmanouilidis – 2025. PDF/Pre-print → https://lnkd.in/eZvDK6Jx arXiv → https://lnkd.in/eGwDKpzM This review presents a structured analysis of trust, risk and security management in large-language-model-based agentic multi-agent systems, introducing a risk taxonomy and oversight strategies. 2. Application-Driven Value Alignment in Agentic AI Systems: Survey and Perspectives Wei Zeng, Hengshu Zhu, Chuan Qin, Han Wu, Yihang Cheng, Sirui Zhang, Xiaowei Jin, Yinuo Shen, Zhenxing Wang, Feimin Zhong & Hui Xiong – 2025. PDF/Pre-print → https://lnkd.in/egYNgP8c arXiv → https://lnkd.in/e9rkKCrc A survey focusing on value-alignment in agentic systems, categorising alignment methods, evaluating multi-agent value coordination, and highlighting application-specific risk implications. 3. LibVulnWatch: A Deep Assessment Agent System and Leaderboard for Uncovering Hidden Vulnerabilities in Open-Source AI Libraries Zekun Wu, Seonglae Cho, Umar Mohammed, Cristian Munoz, Kleyton Costa, Xin Guan, Theo King, Ze Wang, Emre Kazim, Adriano Koshiyama – 2025. PDF/Pre-print → https://lnkd.in/ev9b8AqU arXiv → https://lnkd.in/eJjGTV-c Proposes a graph-based agentic assessment framework that orchestrates specialized agents to examine open-source AI libraries across multiple risk domains (licensing, security, supply chain) — producing governance-aligned scores and continuous monitoring. Practical implications: if you're building agentic systems (LLMs + tool invocation + orchestration) you should design for autonomy tiering, continuous risk scoring, behavioral monitoring, audit logs, and value interplay among agents. 👇 For Discussion: Which of these monitoring/governance strategies (risk taxonomy vs value-alignment vs ecosystem-monitoring) feels most applicable to your system? In your deployment, how would you classify agent autonomy vs risk level? What gaps do you see — e.g., dynamic risk scoring, supply-chain vulnerability monitoring, or value drift across multi-agent workflows? #AI #AgenticAI #Governance #MultiAgentSystems #LLM #HumanAI #Alignment #Research