双语阅读 | 人工智能安全(3)
首要议题是人工智能与网络攻击的融合。先进的 AI 系统正快速重塑网络空间：它降低了复杂攻击的门槛，加速了自动化漏洞的发掘，并扩大了信息操纵的规模。在当前大国间本就极度缺乏互信的氛围下，由 AI 驱动的网络事件极易被误读为“蓄意升级”。
The first is the intersection between AI and cyber attacks. Advanced AI systems are rapidly changing the cyber domain by lowering barriers to sophisticated attacks, accelerating automated vulnerability discovery and increasing the scale of information manipulation. In an environment already characterised by deep mistrust between major powers, AI-enabled cyber incidents could easily be misinterpreted as intentional escalation.
第二点涉及人工智能与生物安全威胁。AI 正以前所未有的方式加速医学与生命科学领域的科研突破。然而，越来越多的科学家担忧，日益强大的 AI 系统可能会降低实施危险生物行为的技术门槛。这种忧虑如今已不再仅限于政府层面，更延伸至非国家行为体获取昔日仅限国家级实验室掌握之能力的可能性。
The second area is AI and biological risks. AI is already accelerating scientific discovery in medicine and biology in remarkable ways. But many scientists also worry that increasingly capable AI systems could lower the technical barriers for harmful biological misuse. This concern is no longer limited to governments. It increasingly involves the possibility that small groups or even individuals could gain access to capabilities once restricted to advanced state laboratories.
无论中国还是美国，均不愿目睹上述风险成真。这也正是为何在中美维持人工智能安全沟通渠道显得至关重要的原因。在 AI 时代，有一种风险被严重低估：系统故障正愈发可能被视作“敌意行为”。随着 AI 系统变得更加自主、复杂且“黑箱化”，各国政府越来越难以分辨一次事故究竟源于技术故障、意外升级，还是蓄意攻击。在危机情境中，这种模糊性本身就可能引发不稳定。
Neither China nor the United States benefits from such a future. This is precisely why maintaining channels of communication on AI safety matters, even during periods of broader strategic competition. One of the underappreciated dangers of the AI era is that system failures may increasingly be seen as hostile intent. As AI systems become more autonomous and opaque, it may become harder for governments to distinguish between technical malfunction, unintended escalation and deliberate attack. In a crisis situation, this ambiguity itself could become destabilising.
*opaque (adj.)不透明的
历史屡次证明，对技术过度“安全化”只会加剧互信缺失，并导致全球创新生态的碎片化。当下，人工智能治理本身正被全面卷入地缘政治对抗之中。若每一次 AI 能力的突破都被置于零和博弈的框架下解读，那么真正亟需的安全合作，反而会在最关键的时刻陷入政治上的不可行。
History has repeatedly shown that excessive securitisation of technology can deepen mistrust and fragment the global innovation ecosystem. Today, there is a growing risk that AI governance itself becomes absorbed entirely into geopolitical confrontation. If every AI advance is interpreted primarily through a zero-sum framework, then safety cooperation becomes politically difficult precisely when it is most necessary.
*fragment (v.)使破裂