摘要： 以DeepSeek、Qwen、LLaMA 为代表的开源大模型降低了技术准入门槛，在缩小数字鸿沟与建立主权人工智能方面呈现显著正向效应。然而，人工智能技术赋权效应与安全威胁呈现非线性增长态势，形成技术普惠与风险扩散的“安全悖论”，表现为风险内生 与价值外生的技术本体矛盾、通用性与战略性的应用属性冲突以及技术迭代与治理响应 的时滞困境。在此三重矛盾的复合作用下，各国陷入发展与安全的两难困境，片面追求 开源模型的技术采纳将导致风险敞口的持续扩大，过度强调安全控制则可能丧失技术发 展先机，最终形成技术扩散与安全威胁的循环。因此，后发国家亟须在认知“安全悖论” 的基础上构建适应性治理框架，国际社会需要加强人工智能安全的全球治理，加大对后 发国家的能力建设支持，以此实现人工智能技术扩散效益与安全治理的动态均衡。

关键词: 人工智能, 开源模型, 技术扩散, 安全悖论, 风险治理

Abstract: Open-source large models such as DeepSeek, Qwen, and LLaMA have lowered technological barriers to entry, demonstrating significant positive effects in bridging the digital divide and establishing sovereign artificial intelligence. However, the empowering effects of artificial intelligence technologies and the growing security threats exhibit a nonlinear growth trend, forming a “security paradox” where technological accessibility and risk diffusion coexist. This paradox is manifested in three interrelated contradictions: the inherent risk and externalized value of the technology itself; the conflict between its general and strategic application attributes; and the temporal lag between technological iteration and governance responses. Under the compound effects of these three contradictions, nations face a dilemma between development and security. A one-sided pursuit of technology adoption through opensource models will lead to an ever-expanding risk exposure; an excessive emphasis on security controls may result in losing the initiative in technological advancement; ultimately creating a vicious cycle of technology diffusion and security threats. Therefore, developing countries urgently need to establish adaptive governance frameworks based on an understanding of the “security paradox”. The international community must strengthen global governance of artificial intelligence security and enhance capacity-building support for developing countries to achieve a dynamic equilibrium between the benefits of artificial intelligence technology diffusion and security governance.

Key words: artificial intelligence, open-source models, technology diffusion, security paradox, risk governance