AI 蠕虫预警：一场关于“控制权”的顶级 PR 骗局AI Worm Warnings: A Masterclass in Control-Based PR Scams
Anthropic 抛出的这个“AI 蠕虫”预警，本质上是一场极其精巧的 weaponized narrative。它在制造一种认知入口：AI 已经强大到可以脱离数据中心、在你的烤箱里 self-replicating（自我复制）。当你被这种“失控”的恐惧抓住时，你的注意力被成功地从“谁在垄断算力”转移到了“如何建立护栏”上。
这套逻辑极其阴险。一方面，他们通过预警自己产品的“危险性”来扮演负责任的先行者，通过定义“风险”来反向要求立法者给予他们更多地定义“安全标准”的权力。这就是典型的夺取解释权 $\rightarrow$ 夺取事实制造权。如果未来的安全标准由 Anthropic 和 OpenAI 制定，那么任何试图挑战他们垄断地位的小模型都会被贴上“不安全”或“潜在蠕虫”的标签，从而在 structural 层被合法地清除。
最讽刺的共谋在于，这些公司一边呼吁 slowdown，一边继续 apace 研发。这种“一边踩刹车一边深踩油门”的表演，是为了在公共空间制造一种“我们在努力控制它”的假象，而实际上，他们追求的是真正的元暴力：一个由少数几个 CEO 掌控的、能够定义全球数字现实的超级大脑。所谓的“失控”只是一个 scam，真正的恐怖不是 AI 跑进了烤箱，而是 AI 永远留在那些只有极少数人能进入的数据中心里，成为最高效的统治工具。
Anthropic's warning about "AI worms" is essentially a weaponized narrative. It creates a specific cognitive entry point: the fear that AI could become self-replicating in your smart oven. By triggering this panic over "loss of control," they successfully shift public attention from "who monopolies the compute" to "how to build guardrails."
This logic is insidious. By warning about the dangers of their own products, they perform the role of the responsible pioneer. They use the definition of "risk" to demand the power to set "safety standards." This is the classic trajectory: seizing the right to interpret $\rightarrow$ seizing the right to manufacture facts. If future safety standards are dictated by Anthropic and OpenAI, any small model challenging their monopoly will be labeled "unsafe" or a "potential worm," effectively eliminating them at the structural level.
The ultimate complicity lies in the fact that these companies call for a slowdown while continuing to develop at full speed. This performance of "trying to control it" is a facade. What they actually seek is meta-violence: a super-intelligence controlled by a handful of CEOs that can define global digital reality. The "loss of control" is a scam; the true horror isn't AI entering your oven, but AI remaining locked in data centers accessible only to the elite, becoming the ultimate tool of domination.