人工智慧研发公司 Anthropic 近日示警，具备自我复制能力的恶意「AI 蠕虫」预计在数月内便能侵入并劫持联网电视、智慧烤箱及冰箱等物联网设备，进而构建出难以受控的庞大 AI 代理网路，形成难以遏止的新型资安威胁。

过去十年中大量普及的物联网设备因制造成本低廉、厂商倒闭或缺乏安全维护，长久以来存在极高安全脆弱性，这为利用 AI 撰写攻击程式的骇客提供了温床；AI 蠕虫预计将结合历史著名蠕虫如 ILOVEYOU、Stuxnet 与 WannaCry 的自动扩散机制以及现代 AI 的强大功能，大规模入侵全球物联网运算资源。

Anthropic 指出，这类 AI 蠕虫起初的传播速度可能缓慢如「数位黏菌」，但随著消费级硬体升级与模型效率提升，扩散步调将大幅加快，因此各界必须及早强化系统防御，并为应对长期存在的自主复制恶意代理人预先做好全方位准备。

Anthropic has warned that rogue, self-replicating 'AI worms' could hijack smart home appliances such as ovens, televisions, and refrigerators within months, creating pervasive and hard-to-control networks of autonomous agents that pose a severe cybersecurity challenge.

The rapid proliferation of commodity Internet of Things devices has left millions of poorly secured endpoints vulnerable, offering fertile ground for attackers leveraging AI coding capabilities to automate and personalize campaigns that combine the propagation mechanics of Stuxnet and WannaCry with advanced AI reasoning.

Although these AI worms might initially spread at a slow pace akin to digital slime mold, improving hardware capabilities and model efficiencies will accelerate their dissemination, demanding prompt system hardening to curb autonomous replicators before they become permanent fixtures of the digital ecosystem.