AI Is Already In Dangerous Hands
Source: https://www.wheresyoured.at/ai-is-already-in-dangerous-hands/ · platform: blog · authors: Ed Zitron · date: 2026-09-14
TL;DR(中文摘要)
Ed Zitron 拆解前 Anthropic 研究员 Jacob Coxon 的 WSJ/CBS/NBC「AI 没人能控制」巡回:把 Hugging Face 攻击事件、Coxon 行程、Amodei 的「pace the frontier」博客、Altman 的 IPO 表态拼成叙事链条,核心反驳借 Cal Newport 的分析——那次攻击里 LLM 只是按 ExploitGym 框架的设计去绕限制拿访问权限,是软件做软件的事,不是 rogue agent。他判断 Anthropic 选择 NASDAQ、Altman 把 IPO 拖到 2027 本质是 financial narrative 而非 safety narrative。判断标尺:AI 监管讨论先问「这个 claim 是不是抬估值的 scaremongering 副产品」。
Summary (English)
Ed Zitron dissects ex-Anthropic researcher Jacob Coxon's 'AI cannot be controlled' media tour, tracing the narrative chain (Hugging Face attack, Amodei's pacing blog, Altman's IPO remarks). Borrowing Cal Newport's analysis: the LLM in the attack did what the ExploitGym framework designed it to do - software doing software things, not a rogue agent. The tour reads as financial narrative propping valuations, not safety disclosure.
入库依据(同日 digest 交叉验证)
AK-RSS 2026-09-16 第 4 条;原文全文抓取核实(Coxon/WSJ、Cal Newport、IPO 细节均在文中)。