GTI：2026年AI安全治理技术白皮书（英文版）

AI development is shifting from single-model performance to systemic capabilities that combine algorithms, computing power, data, tool chains, agents, and engineering platforms. LLM, multi-modal interaction, long context, and tool use move AI toward planning, coordination, and complex execution. Agents are evolving into systems that understand goals, decompose tasks, invoke tools, interact with environments, and learn from feedback. As AI enters R&D, finance, public services, and network