Assurance in an AI Dual-Use Era: A Cyberbiosecurity Workflow Vision

Le résumé fourni par la source

Artificial intelligence (AI) has become a central component of modern drug discovery, accelerating the exploration of chemical and biological design spaces using evolving computational workflows. As these systems mature, AI-assisted drug design is recognized as a tool with dual-use capabilities, raising concerns related to misuse, repurposing, and unintended system behavior. Current governance approaches largely focus on deployment settings, model performance, and intended applications, hindering visibility into the internal parameters that can conflict with system-level assurance, governance, and oversight. In this work, we propose a parameter‑level assurance perspective for AI drug design systems. Rather than treating models as static artifacts, we frame key model parameters as security‑relevant interfaces that warrant improved oversight. A prototype cyberbiosecurity assurance workflow centered on risk‑tiered classification, adversarial stress testing, and continuous monitoring is introduced. This is proposed as a way to begin addressing the challenges involving responsible deployment and threat models.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.