美国政府让Anthropic关了两个模型，理由含糊，还禁止外国人访问。有人猜是因为Anthropic之前没接五角大楼的军事合同。
美国政府以国家安全和越狱漏洞为由，强制Anthropic关闭Fable 5和Mythos 5模型。指令仅基于口头证据，未提供具体技术细节。Anthropic声称该漏洞很窄，且GPT-5.5也能实现相同功能。同时禁止所有外国人访问，包括Anthropic的外籍员工。此事发生在Anthropic拒绝与五角大楼签订军事监控合同四个月之后。
The US government forced Anthropic to shut down Fa…
The US government forced Anthropic to shut down Fable 5 and Mythos 5 yesterday - an export control directive citing national security and a jailbreak. Access was cut for everyone, including internal, behind-closed-doors use.
The “evidence” was verbal only - no specifics, no technical details. Anthropic says the jailbreak is narrow, and that GPT-5.5 can do the same things. The directive also bars any foreign national from access - whether they’re in Paris, Tokyo, or San Francisco - including Anthropic’s own foreign staff.
If this becomes the industry standard, do OpenAI and Google get shut down next? Or is this simply what happens when a frontier lab says no to the Pentagon? Four months after Anthropic walked away from military surveillance contracts, they’re the only ones being blocked.
This looks less like security policy and more like punishment for refusing to play along.