4月18日的专栏把人工智能政策定性为“关键时刻”，原因是模型快速迭代使风险上升，专家普遍认为AI导致人类灭绝或其他灾难的概率约为5%–10%。文中批评“AI诱发混乱”等措辞淡化了风险，提到《经济学人》已承认灭绝风险并指向Mythos与递归式自我改进路线，这些都使“让技术自行兜底安全”的赌注更具危险性。
第二段指出，AI治理讨论几乎忽视投资方，而资金决定了应用优先级、扩展速度与技术是否上市，因而投资逻辑在高风险领域正逼近“投资等同于军备投入”。尽管高风险军用AI扩张很快，传统尽职调查跟不上，轻量化而可执行的人权尽调、透明披露与安全红线仍被视为必要，以遏制对公众和地球的伤害。
第三段把大学治理与AI治理作类比：高风险时代既有“有身份无实权”的学术权威，也有“有权无威望”的行政管理，导致官僚化与善意不足；另一位作者则把AI的破坏潜能与曼哈顿计划相比，主张应由政府而非市场单独定轨。与此同时，Malthus相关专栏重申要基于证据反对盲目乐观，而“早餐少浪费”文章的一个观察也被点出：通常拿二次上菜的人会把拿走的食物吃掉，说明并非所有节约方案都按直觉工作。
The April 18th editorial frames this as a pivotal AI-policy moment, with model advances accelerating risk and experts putting a 5–10% chance on AI causing human extinction or another catastrophe. It says phrases like “AI-induced chaos” soften the threat, noting that even *The Economist* acknowledges extinction risk, while highlighting Anthropic’s move away from a safety pledge and its push toward Mythos-driven recursive self-improvement as proof that betting on AI to guarantee its own safety is unacceptable.
The piece also emphasizes that investors are a missing actor in AI accountability: capital decides what gets built, scaled, approved, or excluded, so high-risk AI is increasingly shaped by market incentives akin to war finance. With military AI expanding faster than due diligence systems can adapt, conventional safeguards can still matter—especially minimum standards on human-rights review, transparency of military uses, and clear red lines against catastrophic deployment.
Other commentary links governance failures across sectors. One column depicts universities as suffering a “status without authority” versus “authority without status” mismatch, while another argues that AI’s destructive potential is closer to the Manhattan Project than to ordinary innovation, making state control preferable to laissez-faire. A final pair of essays contrasts Malthusian evidence-first skepticism toward utopian optimism and reports that repeat-buffet diners often consume what they take, suggesting some waste-reduction assumptions are counterintuitive.
Source: Letters
Subtitle: what Mythos means for AI , Thomas Malthus, academic status, measuring buffet portions
Dateline: The Economist May 2nd 2026