The White House Is Making Up Its Rules for AI in Real Time
The White House Is Making Up Its Rules for AI in Real Time
白宫正在实时“现编”人工智能监管规则
It’s been nearly a week since the Trump administration sent an export control directive to Anthropic, forcing one of the world’s leading AI labs to pull its most advanced models offline. After days of negotiations between Anthropic and the White House, the two still remain at odds about how to bring Claude Mythos and Fable 5 back. Why? Well, it depends whom you ask. 距离特朗普政府向 Anthropic 发出出口管制指令已经过去近一周了，这一指令迫使这家全球领先的 AI 实验室将其最先进的模型下线。在 Anthropic 与白宫进行了数天的谈判后，双方在如何让 Claude Mythos 和 Fable 5 恢复上线的问题上仍存在分歧。为什么会这样？这取决于你问的是谁。
Throughout this entire debacle, Anthropic doesn’t believe it violated any concrete procedures or rules laid out by the Trump administration, according to a person close to the company. But the White House contends that Anthropic behaved recklessly, demonstrating that it can’t be trusted to safely roll out frontier technology. 据一位接近 Anthropic 的人士透露，在整个事件中，Anthropic 并不认为自己违反了特朗普政府制定的任何具体程序或规则。但白宫则坚称 Anthropic 的行为鲁莽，证明了其在安全部署前沿技术方面不可信赖。
This saga proved to me that we’re now officially in the Wild West era of American AI regulation. While there are few laws on the books governing frontier AI development, that doesn’t mean companies won’t end up in trouble with Trump’s White House when they cross unspoken lines. 这场风波向我证明，我们现在正式进入了美国人工智能监管的“蛮荒西部”时代。虽然目前几乎没有针对前沿 AI 开发的成文法律，但这并不意味着当企业触碰了那些“潜规则”时，不会惹上特朗普白宫的麻烦。
“The problem here is that the White House has been in this extreme anti-regulatory posture, and they’re now faced with the real AI capabilities that people have been predicting for many years,” says a former White House technology official, who asked to remain anonymous to avoid jeopardizing their professional relationships. “There should have been preparation and policies to systematically deal with this, managing the benefits and risks, but instead it’s just this slap-dash approach that puts the AI industry in a real quandary.” “这里的问题在于，白宫一直处于极端的反监管姿态，而现在他们面对的是人们多年来一直预测的真实 AI 能力，”一位要求匿名以避免影响职业关系的前白宫技术官员表示。“本应有相应的准备和政策来系统地处理这些问题，管理其收益与风险，但现在这种草率的做法让 AI 行业陷入了真正的困境。”
The Trump administration has repeatedly blocked efforts to impose guardrails on the AI industry, often arguing the rules could hamper US innovation and lead the country to fall behind rivals like China. Since he returned to the White House, President Trump has signed executive orders that reversed a Biden-era effort to create a national AI framework and created a federal task force to challenge state laws that could be deemed onerous. 特朗普政府一再阻挠对 AI 行业施加护栏的努力，通常辩称这些规则会阻碍美国创新，并导致美国落后于中国等竞争对手。自重返白宫以来，特朗普总统签署了行政命令，推翻了拜登时期建立国家 AI 框架的努力，并成立了一个联邦工作组，旨在挑战那些被认为过于繁琐的州法律。
While WIRED and other publications have made details of the negotiations between Anthropic and the White House public over the last week, the dispute is still defined by its opaqueness. At no point has the US government clearly stated what Anthropic did wrong—the best we have is a post on X outlining the general situation from White House technology adviser David Sacks. 尽管《连线》（WIRED）和其他媒体在过去一周披露了 Anthropic 与白宫之间谈判的细节，但这场争端依然充满不透明性。美国政府从未明确说明 Anthropic 到底做错了什么——我们目前所能得到的最好信息，也仅仅是白宫技术顾问大卫·萨克斯（David Sacks）在 X 上发布的一篇概述总体情况的帖子。
Ironically, the White House’s actions have likely hampered the very kind of innovation it wants to protect. The Trump administration demanded that Anthropic prohibit all foreign nationals from accessing Mythos and Fable 5, preventing many of the AI lab’s own employees from accessing its most cutting-edge models, which the company says have sped up its research and development in recent months. All of Anthropic’s customers are locked out, too, including Apple, Meta, and much of the Fortune 500. 讽刺的是，白宫的行动很可能阻碍了它想要保护的创新。特朗普政府要求 Anthropic 禁止所有外国公民访问 Mythos 和 Fable 5，这导致该 AI 实验室的许多员工无法使用其最尖端的模型，而公司称这些模型在近几个月内加速了其研发进程。Anthropic 的所有客户也都被拒之门外，包括苹果、Meta 以及大部分财富 500 强企业。
The White House may have had legitimate reasons to be concerned about Anthropic’s models. As my colleagues and I reported on Wednesday, US officials grew concerned when they learned earlier this month that Anthropic shared Mythos with SK Telecom, a South Korean telecom giant they allege has ties to China. Separately, Amazon CEO Andy Jassy raised concerns to US Treasury secretary Scott Bessent that some of the guardrails on Claude Fable 5, a safeguarded version of Mythos, could be circumvented. 白宫对 Anthropic 的模型感到担忧或许有其正当理由。正如我和同事周三报道的那样，美国官员在本月初得知 Anthropic 与韩国电信巨头 SK 电讯（SK Telecom）共享了 Mythos 模型后感到担忧，他们声称该公司与中国有联系。另外，亚马逊首席执行官安迪·贾西（Andy Jassy）也向美国财政部长斯科特·贝森特（Scott Bessent）表达了担忧，称 Claude Fable 5（Mythos 的安全版本）上的某些护栏可能会被绕过。
Even if these worries are legitimate, that doesn’t mean the White House handled them adequately. In the first case, Anthropic says it coordinated with the US government on the rollout of Mythos, suggesting there may have been a chance for officials to raise alarm about SK Telecom ahead of time. Anthropic has also been working with the Korean company for years, and the arrangement seemingly never caused it to run into national security problems before. And when the White House raised concerns about SK Telecom to Anthropic, it revoked access to the model immediately, we reported. 即使这些担忧是合理的，也不代表白宫的处理方式是得当的。在第一种情况下，Anthropic 表示在 Mythos 发布时与美国政府进行了协调，这表明官员们本有机会提前对 SK 电讯提出警示。Anthropic 与这家韩国公司合作多年，这种安排此前似乎从未引发过国家安全问题。而且据我们报道，当白宫向 Anthropic 提出关于 SK 电讯的担忧时，该公司立即撤销了对该模型的访问权限。
In the second case, WIRED and other outlets have written at length about how every large language model can be jailbroken to varying extents. Anthropic and independent cybersecurity researchers argue that solving jailbreaks is not an easy or isolated issue. Because AI models are probabilistic, not deterministic, tech companies can’t guarantee what exactly they will generate in response to a given prompt. If the White House won’t let Anthropic release Claude Fable 5 until the jailbreaking problem is solved, I wouldn’t hold my breath. 在第二种情况下，《连线》和其他媒体已经详细撰文阐述了每一个大语言模型在不同程度上都可以被“越狱”。Anthropic 和独立网络安全研究人员认为，解决越狱问题并非易事，也不是一个孤立的问题。由于 AI 模型是概率性的而非确定性的，科技公司无法保证模型在响应特定提示时究竟会生成什么。如果白宫坚持要等到越狱问题彻底解决才允许 Anthropic 发布 Claude Fable 5，那我劝你别抱太大希望。
But ultimately, the problem here is not that the US government is trying to ensure that advanced AI models have proper safeguards or stay out of the hands of America’s foes. It’s that the Trump administration is now in a position where it has to make these regulatory decisions in real time. 但归根结底，这里的问题不在于美国政府试图确保先进 AI 模型拥有适当的护栏，或防止其落入美国敌对势力手中。问题在于，特朗普政府现在处于一种必须实时做出这些监管决策的境地。
I can assure you that other AI labs like OpenAI, Google, and Meta have been watching Anthropic’s misfortune with bated breath. I’ve been asking AI executives the same question all week: How can you avoid a fate similar to Anthropic’s? 我可以向你保证，OpenAI、谷歌和 Meta 等其他 AI 实验室都在屏息关注着 Anthropic 的遭遇。整个星期我都在问 AI 高管们同一个问题：你们如何才能避免重蹈 Anthropic 的覆辙？
So far, many AI leaders are coming to the same conclusion: They believe they will need to give the White House early access to their latest AI models. Additionally, they say, they need to be extremely proactive about sharing information with the Trump administration about upcoming model launches—the risk of catching officials off guard is simply too great. 到目前为止，许多 AI 领袖得出了相同的结论：他们认为需要让白宫提前访问其最新的 AI 模型。此外，他们表示，必须极其主动地与特朗普政府分享有关即将发布模型的信息——让官员们措手不及的风险实在太大了。
“Advance notice, advance access. I think those are the primary asks that we’ve heard, not just from the US, but others around the world,” Aidan Gomez, CEO of the Canadian AI lab Cohere, said in an interview earlier this week. “I think those are good things in many respects. It shows strong engagement and consideration by authorities on a super important technology.” “提前通知，提前访问。我认为这是我们听到的主要要求，不仅来自美国，也来自世界其他地方，”加拿大 AI 实验室 Cohere 的首席执行官艾丹·戈麦斯（Aidan Gomez）在本周早些时候的一次采访中说道。“我认为在许多方面这些都是好事。这表明当局对一项极其重要的技术表现出了强有力的参与和考量。”
President Trump signed an executive order last month that created a “voluntary” system for AI labs to submit models to the government for early testing. It included a carve-out explicitly stating that this would not become a mandatory licensing regime, a major concern raised by the… 特朗普总统上个月签署了一项行政命令，建立了一个“自愿”系统，供 AI 实验室向政府提交模型进行早期测试。该命令包含一项明确的豁免条款，声明这不会演变成强制性的许可制度，而这正是此前引发的主要担忧……