跳到正文
Semafor· Brendan Ruberry·· 2026-10-14AI 評分48

Anthropic 警告 AI 或將自我複製

Anthropic issues warning on self-replicating AI

AI 導讀

Anthropic 警告,AI 可能很快在家居內連接互聯網的個人裝置間自我複製,引發難以清除的「殭屍網絡」風險;AI 羣集亦可能劫持整個互聯網。Anthropic 一名共同創辦人另稱,AI 在極少監督下快速迭代自身的「遞迴自我改進」研究過程可能明年出現,相關警告引來批評。

正文

AI may soon be able to self-replicate across internet-connected personal devices in homes, Anthropic warned, raising the possibility of so-called “botnets” that would be difficult to stamp out.

While botnets predate AI and range from harmless to malicious, impending AI swarms could hijack the entire internet, the startup has said. Separately, an Anthropic co-founder wrote that “recursive self-improvement” — a research process whereby AI rapidly iterates upon itself with minimal oversight — could arrive next year.

The warnings drew sharp criticism: “They’re not going to build something that they don’t want to exist,” the journalist Derek Thompson argued, adding that Anthropic’s leaders need to stop acting “like a local news anchor reporting on traffic on the 101, and acknowledge that they are the traffic.”

— Brendan Ruberry

來源:Semafor · semafor.com