AI总结
-
核心结论：AI已成为美国地缘政治的新权力工具。美国政府通过管控 frontier 模型及其算力，确立了全球 AI 技术准入的 gatekeeper（守门人）地位。
-
结构逻辑：
- 事件背景：美国政府封禁 Anthropic 模型对海外用户的访问，彰显其通过技术准入掌握全球话语权的意图。
- 历史参照：类比核武器与密码学，指出 AI 技术既具有高度危险性，又具有不可完全封锁的属性。
- 博弈挑战：分析美国面临的经济利益（出口需求）与安全管控（防御优势）之间的两难困境。
- 应对策略：建议美国盟友（欧洲、东亚）通过加强能源与算力基础设施建设，而非盲目抗议，以提升自身的议价能力。
-
关键信息：
- 核心案例：美国政府将 Anthropic 的 Fable 和 Mythos 模型列为管控对象。
- 核心资源：算力（compute）是全球核心竞争壁垒；美国拥有较欧洲约 15 倍的算力储备。
- 行业趋势：AI 逐渐成为外交与贸易的杠杆；全球正从关注能源供给转向关注算力主权。
-
问题导向：
- 问题：美国如何利用 AI 构筑全球控制力？其他国家应如何应对这种不对称的依赖关系？
- 回答：美国通过行政干预定义“安全风险”来筛选准入；他国应通过基础设施（电网、数据中心）建设和技术自主化来增强谈判筹码。
正文
Its government is now the gatekeeper to frontier models—and most compute
Illustration: The Economist/Peter Crowther
THE NEWS is full of how an ignominious peace deal with Iran exemplifies a decline in American power. That conclusion could hardly be more wrong. On June 12th the Trump administration ordered Anthropic to block foreigners from Fable and Mythos, its latest and most capable frontier AI models. In an instant, everyone learned that the American government can decide who may use the world’s most important technology. You don’t get much more powerful than that.
The administration was responding to a supposed jailbreak for Fable, meaning a prompt that circumvents defences against uses such as hacking computers or making bioweapons. The chances are that it wanted Anthropic to switch off the models for everyone, and that targeting foreigners was a means to an end. Sure enough, that is what Anthropic did, while claiming that the concern about its model was overblown. The legal basis of the order remains unclear, and the ban seems unlikely to last.
What matters, though, is the demonstration that global access to the best AI may come down to a decision in the Oval Office. The administration showed in March that it is prepared to trample on the frontier AI companies, when it designated Anthropic a “supply-chain risk”. Now it has shown that it is prepared to trample on users, too.
America must decide how to wield this vast new power. The rest of the world must decide what to do about it. Even as it plans for an unreliable America in everything from defence to trade, it now has to cope with a new way of being captive to the world’s biggest economy.
This is not the first time America has tried to restrict access to frontier technologies. After the second world war it stopped helping Britain’s nuclear-weapons programme. When modern cryptography emerged in the 1970s, it blocked exports, before accepting the trade-off between having secure allies and using secrets to boost its own offensive capabilities. Uncle Sam still refuses to share its best military equipment, even with close allies. America kept the F -22 fighter for itself; allies got the F -35.
To control access to a technology, though, depends on its nature, and rivals’ ability to develop it independently. Nuclear co-operation with Britain resumed in the 1950s after it developed technology of its own; with other countries, America used the Nuclear Non-Proliferation Treaty of 1968. Cryptography methods could not be contained and eventually went public. Many countries are capable of cyber-attacks.
Frontier AI has echoes of all these examples. If the very best models can disable crucial infrastructure or help users create pandemic-ready pathogens then, like nuclear weapons, they are too dangerous for public hands. But as with cryptography algorithms, it will be hard to be sure that advanced proprietary capabilities will never be copied. Open-weight models, which anyone can download, could advance and proliferate. In cyber-security a small imbalance can bring big advantages: if an attacker has version 5 while the defender is stuck with version 4, and the better model uncovers just one more vulnerability, the weaker party will be compromised. As AI is embedded in military hardware, a similar logic may apply on the battlefield.
Yet America has a huge economic interest in leading in AI and selling its tech to foreigners. Many Anthropic staff are not American and so were hit by the ban; to freeze AI research at America’s best lab would be self-defeating. The firm also says that 80% of its consumer use is overseas. As American technology has boomed over the past decade, Europe’s payments to America for intellectual-property products have risen fivefold. America should not want to give the rest of the world a reason to team up with China, the second-ranking AI power.
This may lead to a hierarchy of access. The best capabilities will be closely guarded by America, to provide an edge in cyber-offence and military capability. The next-best alternatives may be available to allies—the equivalent of the F-35. And a sufficiently handicapped model may be sold to the world with the best safety precautions its makers can design.
Such a future would be uncomfortable for America’s allies, which are nowhere close to rivalling the likes of Anthropic. They are already vulnerable to Mr Trump’s bargaining on trade, alliances, the dollar system and more. AI could become the most important lever of the lot. True, some of the dependence is two-way: America needs Dutch lithography machines and Taiwanese fabs. And foreigners could always shun Anthropic or Open AI in favour of good-enough open-weight alternatives.
Yet running models requires computing power, which America has perhaps 15 times more of than Europe, plus much more ongoing investment. If SpaceX, whose share price has surged after listing on June 12th, realises its vision of data centres in space, the gap is unlikely to close. “Europe 2031”, a gloomy essay about the future of AI, imagines vassal status for Europe as its cyber-security, defence and swathes of its economy come to depend on American models and compute.
Many countries will conclude they need America more than ever. They can still strengthen their bargaining positions. Compute is so lacking that it makes sense to build data centres almost anywhere, yet countries show no sense of what is at stake. In the seven months to June 2025 demand for new grid connections in Britain rose from 41 GW to 125 GW, more than twice peak power demand. Open AI paused a data-centre project there in April, citing regulation and costly energy.
Token gestures
Throughout Europe more energy, easier planning and looser rules for modelmakers would help. In East Asia Taiwan, Japan and South Korea must integrate rather than duplicate their efforts. All should avoid the false promise of government attempts at steering investment. The goal should not be protectionism but ensuring that America is not the only economy where the AI ecosystem can thrive. As in trade and defence, the way to cope with Uncle Sam’s transactional turn is not to whine about alliances but to build strength.■
For subscribers only: to see how we design each week’s cover, sign up to our weekly Cover Story newsletter.