美国政府已禁止外国人，包括紧密的“五眼联盟”情报盟友和英国人工智能安全研究所，访问Anthropic的Fable 5和Mythos 5模型。这一限制是由国家安全担忧引发的，因为国家安全局（NSA）官员透露，Mythos在数小时而非数周内就入侵了几乎所有美国的机密系统。

尽管美国在人工智能领域保持着明显的领先地位，而中国因芯片管制落后大约一年，但这一单边禁令威胁到了严重依赖外国人才的美国AI企业的全球运营。在历史上，这种孤立主义的做法与1946年的《麦克马洪法案》如出一辙，该法案终止了国际核合作，并迫使盟友寻求本国的技术自主。

尽管特朗普政府努力放松对AI的监管，并在6月2日建立了一个自愿的模型审查框架，但其突然的限制仍然是反复无常且不透明的。因此，这些举措存在着推动黑市上非法规避手段以及阻碍合作研究的风险，却无法成功阻止外国人的获取。

The U.S. government has banned foreigners, including close Five Eyes intelligence allies and Britain's AI Security Institute, from accessing Anthropic's Fable 5 and Mythos 5 models. This restriction was prompted by national security concerns, as NSA officials revealed that Mythos breached almost all U.S. classified systems within hours rather than weeks.

While America maintains a clear lead in AI with China lagging approximately one year behind due to chip controls, this unilateral ban threatens the global operations of American AI firms that rely heavily on foreign talent. Historically, this isolationist approach mirrors the 1946 McMahon Act, which ended international nuclear cooperation and forced allies to pursue their own technological sovereignty.

Despite the Trump administration's efforts to deregulate AI and establish a voluntary model-vetting framework on June 2nd, its sudden restrictions remain capricious and opaque. Consequently, these measures risk driving illicit workarounds on the black market and hampering collaborative research without successfully preventing foreign access.

Source: Donald Trump’s blocking of Anthropic is capricious and chaotic

Subtitle: America’s closest allies are shellshocked

Dateline: 6月 18, 2026 03:46 上午