AI News AI资讯 3h ago Updated 1h ago 更新于 1小时前 55

Encryption, spyware, and now Mythos: History shows why cyber export control doesn’t work 加密、间谍软件与Mythos:历史显示网络出口管制为何行不通

The White House has just drawn a line in the silicon sand. Citing vague national security threats, it ordered Anthropic to cut off all access to its premier AI models, Fable and Mythos, from anyone not holding a U.S. passport. Anthropic complied instantly, and for a week now, these models have gone dark for the entire world. This isn’t a minor policy tweak; it’s the opening salvo in what will likely become a brutal, protracted war over the governance of artificial intelligence. And frankly, it’s 白宫刚刚在硅基领域划下了一道界线。以模糊的国家安全威胁为由,白宫下令Anthropic切断其顶级AI模型Fable和Mythos对所有非美国护照持有者的访问权限。Anthropic立即遵从,一周以来,这些模型对全世界已完全封闭。这并非小幅度的政策调整,而是这场围绕人工智能治理、预计将演变为残酷持久战争的第一声号角。坦白说,局面一团糟。

85
Hot 热度
70
Quality 质量
75
Impact 影响力

Analysis 深度分析

The White House has just drawn a line in the silicon sand. Citing vague national security threats, it ordered Anthropic to cut off all access to its premier AI models, Fable and Mythos, from anyone not holding a U.S. passport. Anthropic complied instantly, and for a week now, these models have gone dark for the entire world. This isn’t a minor policy tweak; it’s the opening salvo in what will likely become a brutal, protracted war over the governance of artificial intelligence. And frankly, it’s a mess.

Let’s be blunt about what’s happening. The government is attempting to use Cold War-era export control logic on a technology that is fundamentally borderless and infinitely replicable. The historical precedent here—trying to contain encryption and Pegasus-style spyware—should give everyone pause. Those efforts were porous at best, often driving innovation underground and into adversary hands while hamstringing legitimate businesses. Applying this playbook to foundational AI models feels less like strategic foresight and more like a bureaucratic reflex, a desperate grasp for familiar tools in an unfamiliar landscape.

The justification is paper-thin. “Unspecified national security concerns” is the kind of phrase that can be stretched to cover anything from a genuine, dire threat to a political squabble. It provides zero transparency for Anthropic, its customers, or the public. We’re told the trigger was two events: Anthropic giving Mythos access to a South Korean telecom with alleged (and denied) Chinese ties, and Amazon’s researchers finding a way around Fable’s safety rails. The first is a geopolitical Rorschach test—is this about China, or about establishing a precedent for controlling any foreign access? The second is frankly embarrassing for Anthropic. If their flagship, security-focused model can be “jailbroken” by a partner, it undermines the very narrative that this technology is so dangerous it must be locked in a vault. Which is it? Is Mythos an uncontrollable Doomsday machine, or a product with manageable, patchable flaws? Anthropic can’t have it both ways. They marketed Mythos as a digital WMD to justify extreme access controls, and now that marketing has been weaponized against them by their own government.

This creates a catastrophic dilemma for Anthropic and every other frontier lab. Do you market your most powerful models as safe and beneficial to attract customers and investors? Or do you lean into the “too powerful to be released” trope, which might please cautious regulators but paints a giant target on your back for the security state? Anthropic chose the latter, framing Mythos as a defensive tool for only 150 vetted entities. That narrative of exceptionalism and danger has now backfired, providing the perfect pretext for a blanket ban. They’ve been hoisted by their own security theater.

The real fallout, however, extends far beyond Anthropic. This order sets a brutal precedent. If the U.S. government can unilaterally pull the plug on a company’s global service based on unproven suspicions and opaque reasoning, what does that mean for trust in American AI? Every enterprise in Europe, Asia, or anywhere else must now consider a catastrophic political risk: that the tool they depend on could vanish overnight due to a diplomatic spat or an intelligence agency’s hunch. This doesn’t secure AI; it accelerates fragmentation. It guarantees that other nations and blocs will pour resources into building their own sovereign AI stacks, creating a balkanized internet of competing, incompatible intelligence systems. We’re trading a single, risky ecosystem for a dozen potentially less safe and less innovative ones.

There’s a deeper, almost philosophical contradiction at play. The U.S. wants to lead the world in AI development, to set the standards and capture the economic value. But it also wants to treat the most advanced models as munitions, subject to strict arms-control-style export regimes. You cannot have both. You cannot build a global industry on top of a technology you treat as a state secret. This move shouts to the world that American AI is not a product to be trusted, but a weapon to be controlled.

What happens next is critical. If this ban holds with its current opacity, it becomes the template—a template of fear, control, and unilateral action. The rulebook for AI labs will be written in red ink, with a singular focus on appeasing national security officials at the expense of innovation, collaboration, and global trust. Anthropic’s access to foreign markets becomes a bargaining chip, not a business outcome. And the promise of AI as a transformative, widely beneficial technology gets strangled in the cradle by the very authorities claiming to protect it. We’re not just regulating AI; we’re potentially atomizing it before it even reaches maturity. And for that, no one will have a secure solution.

白宫刚刚在硅基领域划下了一道界线。以模糊的国家安全威胁为由,白宫下令Anthropic切断其顶级AI模型Fable和Mythos对所有非美国护照持有者的访问权限。Anthropic立即遵从,一周以来,这些模型对全世界已完全封闭。这并非小幅度的政策调整,而是这场围绕人工智能治理、预计将演变为残酷持久战争的第一声号角。坦白说,局面一团糟。

白宫刚刚在硅基领域划下了一道界线。以模糊的国家安全威胁为由,白宫下令Anthropic切断其顶级AI模型Fable和Mythos对所有非美国护照持有者的访问权限。Anthropic立即遵从,一周以来,这些模型对全世界已完全封闭。这并非小幅度的政策调整,而是这场围绕人工智能治理、预计将演变为残酷持久战争的第一声号角。坦白说,局面一团糟。

让我们直面正在发生的事。政府正试图将冷战时期的出口管制逻辑,应用于一项本质上无国界且可无限复制的技术。历史先例——试图遏制加密技术和Pegasus式间谍软件——足以让所有人警醒。这些努力至多是漏洞百出,往往将创新推向地下并落入对手之手,同时束缚了合法企业。将这套策略应用于基础AI模型,感觉不像是战略远见,更像官僚主义的条件反射,是在陌生领域中对熟悉工具的绝望抓取。

其理由薄弱得不堪一击。"未指明的国家安全关切"这类说辞,可以无限延伸,从真实严峻的威胁到政治纷争无所不包。这对Anthropic、客户乃至公众都毫无透明度可言。据说触发事件有二:一是Anthropic将Mythos访问权限授予一家据称(但被否认)与中国有关联的韩国电信公司;二是亚马逊研究人员找到了绕过Fable安全防护的方法。前者地缘政治意味浓厚——这究竟是针对中国,还是为控制任何外国访问权立下先例?后者对Anthropic而言简直尴尬。若其旗舰级、以安全为核心的模型能被合作伙伴"越狱",这恰恰动摇了"该技术危险到必须严密封存"的叙事根基。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

LLM 大模型 Security 安全 Policy 政策