Cutting access to Anthropic’s Mythos is a gift to China - FT中文网
登录×
电子邮件/用户名
密码
记住我
请输入邮箱和密码进行绑定操作:
请输入手机号码,通过短信验证(目前仅支持中国大陆地区的手机号):
请您阅读我们的用户注册协议隐私权保护政策,点击下方按钮即视为您接受。
FT商学院

Cutting access to Anthropic’s Mythos is a gift to China

Washington’s suspension just strengthened Beijing’s pitch for its own competing AI models
00:00

{"text":[[{"start":6.2,"text":"The writer is senior fellow in geoeconomics at the European Council on Foreign Relations and the author of ‘Backfire’"}],[{"start":13.65,"text":"Last Friday, Washington imposed export controls on Anthropic’s latest frontier AI models Mythos and Fable 5, forcing the company to cut off global access to both. While developers lamented the loss on online forums and Wall Street tried to assess the impact it may have on Anthropic’s upcoming initial public offering, the bigger picture was lost in the noise: Washington has just done more to boost the appeal of Chinese AI models than Beijing could have ever hoped. "}],[{"start":42,"text":"The Trump administration’s implementation of the export ban was chaotic. The notice arrived at 5.21pm on a Friday, attributed to a security issue that Anthropic says is a misunderstanding. For global companies pondering which AI model to pick, such a process is a red flag. If Washington can shut down access at such short notice, it undermines the case for building infrastructure around US AI. The lack of transparency carries a warning that this extraordinary action could be repeated."}],[{"start":74.25,"text":"Washington’s decision to use export controls shows that its concern is with foreign use of Anthropic’s powerful latest models. But nationality is a poor proxy for misuse and if security concerns are genuine then an outright ban would make more sense. Plus, export controls only bite when the target is hard to substitute, such as extreme ultraviolet lithography for chip manufacturing. Although industry benchmarks show that no model is as good as Mythos and Fable 5, the suspension will fuel demand for substitutes readily available from Chinese competitors."}],[{"start":107.85,"text":"The performance of Chinese models like DeepSeek’s V4 Pro, GLM, Kimi, MiniMax and Qwen is increasingly on a par with that of US AI. Crucially, these models are self-hostable (meaning companies can run them from their own servers), making them harder for governments to limit access to. They are also cheap. DeepSeek’s flagship model costs just 87 cents per 1mn output tokens, making it about 60 times cheaper than Anthropic’s Fable 5. "}],[{"start":136.45,"text":"The idea that global companies could turn away from American AI is not far-fetched. In the first two weeks of June, four of the five most popular models on OpenRouter, a platform that provides access to hundreds of AI models, were Chinese. Among the global top 20, Chinese models processed twice as many tokens as their US-developed competitors. If erratic policymaking makes reliance on US models start to look risky, switching to Chinese substitutes may become more common. "}],[{"start":165.6,"text":"Washington cannot ignore this development. Wider adoption of Chinese models supports Beijing’s bid to shape global AI standards. When developers and companies build on Chinese AI architecture, they absorb Chinese technical assumptions — giving Beijing long-term influence and leverage. The process is self-reinforcing. The more developers use Chinese AI models, the more they build custom tools that they share with peers, making Chinese AI models more appealing to other developers. "}],[{"start":196.45,"text":"The suggestion that only America should be able to access US frontier AI models also raises awkward questions for Anthropic, OpenAI and xAI’s parent SpaceX. If these companies cannot sell future models abroad, the rationale underpinning their sky-high valuations rests on shaky ground."}],[{"start":215.14999999999998,"text":"Beijing has spent years telling potential customers that its technology supplies would never be cut off for political reasons. Washington has just strengthened that pitch, making Chinese AI models look less risky even if they remain weaker on capability. Whether the US reinstates access to Mythos and Fable 5 is beside the point. The signal that access to American AI is revocable is a remarkable own goal. Global companies may decide that the best AI is not necessarily the smartest but the one that they can access every Monday morning."}],[{"start":256.55,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1781782957_2858.mp3"}

版权声明:本文版权归FT中文网所有,未经允许任何单位或个人不得转载,复制或以任何其他方式使用本文全部或部分,侵权必究。

谷歌为Anthropic打造的2000亿美元华尔街融资机器

私募信贷、芯片租赁和数据中心担保,支撑起AI支出的全新庞大模式。

“日元干预”等于“美国自保”

美国联手日本支撑日元,不只是出于盟友情谊,更是为了避免日本加息或美国国债遭抛售、导致美债收益率进一步走高。

“诅咒之岛”:科技游民与诈骗犯藏身的千亿美元奢华开发项目

警方的突击搜查再次打击了马来西亚陷入困境的中资“森林城市”项目的声誉。

问题不在因凡蒂诺

马杜罗:应该将国际足联的监管职能与商业活动分开,其治理应真正做到包容并具有代表性,监督必须真正独立。

俄罗斯扩大“影子”液化天然气船队,应对欧盟禁令

随着明年制裁进一步收紧,越来越多的“影子”船舶将帮助俄罗斯继续出口液化天然气。

阿斯利康与百时美施贵宝:大药企有时也不够大

当资产负债表规模扩大、能够押注潜在重磅药物时,规模才会带来优势。
设置字号×
最小
较小
默认
较大
最大
分享×