The United States government has shifted its policy regarding artificial intelligence, moving from a stance of deregulation toward the implementation of federal vetting processes for frontier AI models.

美國政府已調整其人工智慧政策，從放寬監管的立場轉向對前沿 AI 模型實施聯邦審查程序。

The current administration's pivot is characterized by a transition from the repeal of previous safety mandates to the establishment of pre-market review mechanisms. This strategic realignment is primarily attributed to three catalysts. First, the emergence of Anthropic’s Mythos model demonstrated cybersecurity capabilities that the national security apparatus deemed a critical vulnerability, prompting concerns that adversarial actors could utilize such technology to compromise domestic infrastructure. Second, the administration seeks to maintain geopolitical competitiveness in the face of evolving European Union AI regulations, which Treasury Secretary Scott Bessent suggests could inadvertently benefit Chinese technological expansion. Third, the removal of David Sacks from his role as AI and crypto czar eliminated a primary institutional conduit for the 'innovation-at-all-costs' agenda, which had previously alienated various Republican allies through attempts to obstruct state-level AI legislation.

現任政府的轉向特點在於從廢除之前的安全指令，轉向建立上市前審查機制。此次戰略調整主要歸因於三個催化因素。首先，Anthropic 的 Mythos 模型的出現展示了被國家安全機構視為關鍵漏洞的網絡安全能力，引發了對對手可能利用此類技術損害國內基礎設施的擔憂。其次，面對不斷演變的歐盟 AI 法規，政府尋求維持地緣政治競爭力，財政部長 Scott Bessent 指出，歐盟的法規可能會在無意中惠及中國的技術擴張。第三，David Sacks 離開 AI 與加密貨幣負責人一職，消除了「不計成本創新」議程的主要制度管道，該議程此前因試圖阻撓州級 AI 立法而疏遠了多位共和黨盟友。

Operationally, the Department of Commerce has designated the Center for AI Standards and Innovation (CAISI), managed by the National Institute of Standards and Technology (NIST), as the lead entity for pre-deployment testing. Formal agreements have been executed with Microsoft, Google DeepMind, and xAI to grant the government early access to frontier models, including versions with reduced safeguards to facilitate rigorous risk assessment. However, this framework has encountered academic and professional criticism. Experts from Cornell University and other policy analysts argue that CAISI lacks sufficient funding and technical expertise, suggesting that the absence of transparent, standardized evaluation metrics could lead to the politicization of the vetting process. Some proponents advocate for a transition toward independent auditing systems to ensure accountability without administrative interference.

在操作上，商務部已指定由國家標準與技術研究院 (NIST) 管理的 AI 標準與創新中心 (CAISI) 作為部署前測試的主導實體。政府已與 Microsoft、Google DeepMind 及 xAI 簽署正式協議，以獲取前沿模型的早期訪問權限，包括降低安全防護的版本，以利於進行嚴格的風險評估。然而，此框架遭遇了學術與專業批評。康乃爾大學的專家及其他政策分析師認為，CAISI 缺乏足夠資金與技術專業知識，並建議缺乏透明、標準化評估指標的情況可能會導致審查過程政治化。部分支持者則主張轉向獨立審計系統，以確保在無行政干預的情況下實現問責。

The U.S. government has established a collaborative testing framework with major AI firms to mitigate national security risks, marking a departure from its previous deregulatory trajectory.

美國政府已與主要 AI 公司建立協作測試框架以降低國家安全風險，標誌著其脫離了之前的放寬監管軌跡。