The United States administration has implemented export controls on Anthropic's Mythos and Fable 5 models, resulting in a total suspension of public and corporate access to these technologies.

美國政府已對 Anthropic 的 Mythos 和 Fable 5 模型實施出口管制，導致公眾與企業完全無法使用這些技術。

The current regulatory friction originated from concerns regarding a 'jailbreak' vulnerability in Fable 5—a safeguarded iteration of the more potent Mythos model. This vulnerability was reportedly communicated to the administration by Amazon CEO Andy Jassy, following internal research by Amazon. Concurrently, the administration alleged that Anthropic had shared Mythos with SK Telecom, a South Korean entity with purported ties to China. In response to these security concerns, the White House issued a 90-minute ultimatum for the cessation of the model's availability. Upon the expiration of this deadline, the government imposed export controls prohibiting access by foreign nationals, including non-U.S. citizens employed by Anthropic. Due to the inability to implement instantaneous nationality-based filtering, Anthropic deactivated the models for all users.

目前的監管摩擦源於 Fable 5 的「越獄」漏洞——Fable 5 是功能更強大的 Mythos 模型的安全迭代版本。據報導，在 Amazon 進行內部研究後，由 Amazon 執行長 Andy Jassy 將此漏洞告知政府。與此同時，政府指控 Anthropic 與南韓實體 SK Telecom 分享了 Mythos，而該公司據稱與中國有聯繫。針對這些安全疑慮，白宮發出了 90 分鐘的最後通牒，要求停止提供該模型。在期限屆滿後，政府實施出口管制，禁止外國國民（包括受雇於 Anthropic 的非美國公民）訪問。由於無法立即實施基於國籍的過濾，Anthropic 為所有用戶停用了這些模型。

Institutional analysis suggests a significant divergence in the perceived severity of the security flaws. While the administration characterized Anthropic's deployment as reckless, the company maintained that the vulnerability was limited and did not constitute a critical breach. This impasse is further complicated by the administration's broader regulatory posture; having rescinded the technocratic frameworks of the previous administration in favor of a less restrictive environment, the current executive branch is now operating without a standardized protocol for frontier AI risk assessment. Consequently, the administration has transitioned toward an ad hoc licensing regime, despite previous assertions that early-model review processes would remain voluntary.

機構分析顯示，雙方對安全缺陷嚴重程度的認知存在顯著分歧。政府將 Anthropic 的部署描述為魯莽，而公司則堅持該漏洞有限，並不構成嚴重洩漏。由於政府更寬泛的監管姿態，這一僵局更加複雜；現任行政部門廢除了前任政府的技術官僚框架，以追求較少限制的環境，導致目前在進行前沿 AI 風險評估時缺乏標準化協定。因此，儘管先前聲稱早期模型審核過程將保持自願，但政府已轉向一種臨時的許可制度。

Stakeholders within the AI sector have indicated that this event may precipitate a shift in corporate strategy. To mitigate political risk and avoid similar disruptions, industry leaders are considering the implementation of 'advance notice' and 'advance access' protocols for the White House. Furthermore, the incident has prompted some firms to seek alternative hardware and software arrangements outside the U.S. to ensure operational continuity.

AI 領域的利益相關者表示，此次事件可能會促使公司策略發生轉向。為了降低政治風險並避免類似的中斷，業界領袖正考慮為白宮實施「提前通知」與「提前訪問」協定。此外，該事件已促使部分公司在美國境外尋求替代的硬體和軟體安排，以確保營運連續性。

The White House and Anthropic are currently negotiating the establishment of technical benchmarks to evaluate future security flaws and determine the appropriate thresholds for government intervention.

白宮與 Anthropic 目前正在協商建立技術基準，以評估未來的安全缺陷並確定政府干預的適當門檻。