Vibe-decoding the White House-Anthropic fight over Fable

解读白宫与 Anthropic 关于 Fable 的博弈：一场基于“氛围”的较量

By Tina Nguyen | Jun 17, 2026 作者：Tina Nguyen | 2026年6月17日

Back when I was covering Donald Trump’s first presidency, it was incredibly common to read three different versions of the same story. His administration had split into several factions, all of which had different interests, and all of which hated each other. There was the Reince Priebus traditional GOP faction, the Manhattan society-based Jared Kushner faction, the proto-populist Steve Bannon faction, the deep state John Kelly faction, the conspiracy-MAGA Mike Lindell faction, and so forth. 回想我报道唐纳德·特朗普第一个总统任期时，针对同一件事读到三个不同版本的故事是常有的事。他的政府分裂成多个派系，各方利益诉求不同，且彼此厌恶。当时有 Reince Priebus 代表的传统共和党派系、以曼哈顿社交圈为基础的 Jared Kushner 派系、带有原始民粹主义色彩的 Steve Bannon 派系、代表“深层政府”的 John Kelly 派系，以及阴谋论色彩浓厚的 Mike Lindell 派系等等。

Over time, you could get a sense of which camp was leaking which narratives to the media, either to undermine their rivals or to save their own reputations. In fact, for several decades, media manipulation was a common survival tactic in Trumpworld, which often ran on factionalism and fierce competition for Trump’s approval. As The Associated Press reported on November 26th, 2016, in an article about his pre-presidency business management style, “Aides also often float suggestions to him through the media, knowing that Trump is a voracious watcher of cable TV and might be persuaded by what he sees and hears.” 久而久之，你就能察觉到哪个阵营在向媒体泄露什么样的叙事，目的要么是削弱对手，要么是挽救自己的声誉。事实上，几十年来，操纵媒体一直是“特朗普世界”中常见的生存策略，这个圈子往往依靠派系斗争和对特朗普青睐的激烈争夺来运作。正如美联社在2016年11月26日一篇关于他担任总统前商业管理风格的文章中所报道的那样：“助手们也经常通过媒体向他抛出建议，因为他们知道特朗普是电视新闻的狂热观众，可能会被他所见所闻的内容所说服。”

This tendency toward back-biting and cross-purposes once again gives us a little insight into the White House, this time explaining some of the omnishambles around the Trump admin’s Friday night decision to impose licensing restrictions on Anthropic’s advanced Fable model. Even after several days of statements, reports, and negotiations, it’s still unclear what actually happened behind the scenes, and it’s even less clear who’s responsible for what could have been a massive cybersecurity disaster. 这种背地里捅刀子和目标冲突的倾向，再次让我们得以一窥白宫的内部运作，这也解释了特朗普政府周五晚间决定对 Anthropic 先进的 Fable 模型实施许可限制所引发的一系列混乱。尽管经过几天的声明、报道和谈判，幕后究竟发生了什么仍不清楚，而谁该为这场可能演变成重大网络安全灾难的事件负责，则更加扑朔迷离。

To summarize for the uninitiated: On Friday evening, the White House placed an export control restriction on Anthropic’s Mythos 5 and Fable 5, its most recently released models, preventing foreign governments and nationals from using those two products. It essentially forced Anthropic to completely shut off access to both of those models, throwing their user base into chaos, the company’s future into doubt, and the overall prospects of frontier AI into a gray zone. Could the government simply just tell AI companies to stop operating? 为外行总结一下：周五晚上，白宫对 Anthropic 最近发布的 Mythos 5 和 Fable 5 模型实施了出口管制限制，禁止外国政府和国民使用这两款产品。这实际上迫使 Anthropic 完全切断了对这两个模型的访问，导致其用户群陷入混乱，公司前景存疑，并将前沿人工智能的整体前景推向了灰色地带。政府真的可以随心所欲地叫停人工智能公司的运营吗？

Over the weekend, several conflicting narratives emerged, though a few points remained consistent: The White House’s allies claim that within days of launch, several tech executives, including Amazon president and CEO Andy Jassy, reached out with concerns that Fable and Mythos could be jailbroken, posing an imminent threat to national cybersecurity. The two sides spoke on Friday, though the nature and duration of their calls vary, depending on the account: The Washington Post’s account claimed that Anthropic was given 90 minutes to take down its models, while a White House official told Politico that they had begged Anthropic “for hours.” 周末期间，出现了几种相互矛盾的说法，但有几点是一致的：白宫的盟友声称，在产品发布后的几天内，包括亚马逊总裁兼首席执行官 Andy Jassy 在内的几位科技高管表达了担忧，称 Fable 和 Mythos 可能被“越狱”，从而对国家网络安全构成迫在眉睫的威胁。双方在周五进行了沟通，但根据不同报道，通话的性质和时长各不相同：《华盛顿邮报》称 Anthropic 被要求在90分钟内下架模型，而一位白宫官员则告诉《政治报》，他们曾“苦苦哀求”了 Anthropic 数小时。

And from there, the details become even foggier, at least from the POV of the White House. One party tells The New York Times that Amazon had found a way to “jailbreak” the safety guardrails preventing users from using Fable for cyberattacks. A second party countered to the Times that one could achieve the same results with OpenAI’s ChatGPT 5.5. Semafor reported that it may have been due to a China-linked group accessing Mythos, though no actual jailbreak had been confirmed. And Axios’ sources indicated the admin simply did not like Anthropic’s woke vibe. 此后，细节变得更加模糊，至少从白宫的角度来看是这样。一方告诉《纽约时报》，亚马逊发现了一种“越狱”方法，可以绕过防止用户利用 Fable 进行网络攻击的安全护栏。另一方则向《纽约时报》反驳称，使用 OpenAI 的 ChatGPT 5.5 也能达到同样的效果。Semafor 报道称，这可能是由于一个与中国有关联的组织访问了 Mythos，尽管目前尚未证实发生过实际的越狱。而 Axios 的消息来源则指出，政府只是单纯不喜欢 Anthropic 那种“觉醒”（woke）的氛围。

“Anthropic has not done a great job at trying to speak to the administration and appreciate the ideological differences,” one source familiar with the administration’s thinking told Axios. “It’s like they just speak in different languages.” “Anthropic 在与政府沟通和理解意识形态差异方面做得并不好，”一位熟悉政府想法的消息人士告诉 Axios，“他们就像是在用不同的语言对话。”

The Verge senior AI reporter Hayden Field has a more thorough report here on the view from Anthropic’s world and the AI safety community, which argues that the fears around the jailbreaks are overblown. But inside Washington, the overwhelming consensus I’m hearing is that even if they’re ultimately right, Anthropic and Dario Amodei have created their own political nightmare by opposing the Trump administration — and not in a morally superior #resistance way. “They don’t bend the knee and Dario is stubborn and says what he thinks even when it’s dumb and they have a (justified or not) holier than thou vibe,” one AI policy advocate noted to me. The Verge 的资深 AI 记者 Hayden Field 在此有一篇更详尽的报道，介绍了 Anthropic 和 AI 安全社区的观点，他们认为围绕越狱的担忧被夸大了。但在华盛顿内部，我听到的压倒性共识是：即使他们最终是正确的，Anthropic 和 Dario Amodei 通过对抗特朗普政府，也为自己制造了一场政治噩梦——而且这并非那种道德优越感的“抵抗”。一位 AI 政策倡导者向我指出：“他们不肯屈服，Dario 很固执，即使在想法愚蠢时也会直言不讳，而且他们有一种（无论是否合理）高人一等的氛围。”

One of the most interesting themes emerging in Regulator’s run is how often the hard logic and irrefutable facts inherent to science and technology run headlong into the vibes-based venality of politics. And in the absence of any federal regulatory law, AI regulation is fully reliant on the vibes of whatever statement someone’s convinced Donald Trump to post on Truth Social or executive order he’s been talked into signing. 在《Regulator》专栏中出现的最有趣的主题之一是：科学技术中固有的严密逻辑和不可辩驳的事实，往往会与政治中基于“氛围”的唯利是图发生剧烈碰撞。在缺乏任何联邦监管法律的情况下，人工智能的监管完全依赖于某种“氛围”——即某人说服唐纳德·特朗普在 Truth Social 上发布了什么声明，或者他被说服签署了什么样的行政命令。