The United States government has introduced strict rules regarding the release of advanced artificial intelligence models. These measures primarily affect companies like Anthropic and OpenAI due to growing national security concerns and changes in the market.

美國政府針對發布先進人工智慧模型引入了嚴格規則。由於國家安全憂慮增加以及市場變動，這些措施主要影響如 Anthropic 和 OpenAI 等公司。

The current regulatory environment is moving toward a system of forced licensing. For example, following a government directive in June, the Trump administration required Anthropic to stop foreign nationals from accessing its Fable 5 and Mythos 5 models. Because the company could not verify the citizenship of all users, it suspended these systems globally. Although some access was later restored for a small number of approved partners, this situation has hurt the company's expected revenue and its plans to go public. Similarly, OpenAI has slowed the release of its GPT-5.6 suite, limiting access to government-approved partners. OpenAI claims this is a temporary step to create a reliable release process, even though some staff have doubts about the long-term effects of this oversight.

目前的監管環境正趨向於一種強制許可制度。例如，根據六月份的政府指令，川普政府要求 Anthropic 禁止外國國民使用其 Fable 5 和 Mythos 5 模型。由於公司無法核實所有使用者的國籍，因此在全球範圍內暫停了這些系統。雖然隨後為少數獲准的合作夥伴恢復了部分權限，但此情況已損害了公司的預期營收及其上市計劃。同樣地，OpenAI 減慢了 GPT-5.6 系列的發布速度，將使用權限限制在政府批准的合作夥伴內。OpenAI 聲稱這是一個臨時步驟，旨在建立可靠的發布流程，儘管部分員工對此監管的長期影響存有懷疑。

At the same time, the economic situation for AI models is changing. Business clients are now focusing more on 'intelligence per dollar,' which means they are looking for cheaper alternatives or open-source models. For instance, Zhipu's GLM 5.2 performs as well as Anthropic's Opus 4.8 but costs significantly less. Furthermore, companies are moving away from general test scores and instead using task-specific evaluations, as general scores often hide serious failures in critical areas like healthcare and law.

與此同時，AI 模型的經濟狀況正在發生變化。企業客戶現在更關注「每美元智能」，這意味著他們正在尋找更便宜的替代方案或開源模型。例如，智譜的 GLM 5.2 性能與 Anthropic 的 Opus 4.8 相當，但成本顯著降低。此外，公司正逐漸捨棄通用測試分數，轉而採用針對特定任務的評估，因為通用分數往往掩蓋了在醫療和法律等關鍵領域的嚴重缺陷。

Finally, there are significant risks because AI technology can be used for both helpful and harmful purposes. Reports from the University of Cambridge suggest that advanced models could make cyberattacks and biological threats more dangerous. Consequently, the US has started the 'Pax Silica' initiative to secure AI and semiconductor supply chains across 17 nations to compete with China. Meanwhile, the industry is experiencing a 'talent war,' with some technical experts earning over $1 million in salary, showing how important skilled workers are during this unstable period.

最後，由於 AI 技術可用於有益或有害的目的，因此存在重大風險。劍橋大學的報告指出，先進模型可能會使網路攻擊和生物威脅變得更危險。因此，美國啟動了「Pax Silica」計劃，以確保 17 個國家的 AI 和半導體供應鏈安全，以與中國競爭。與此同時，產業內正經歷一場「人才大戰」，部分技術專家的薪資超過 100 萬美元，顯示出在這個不穩定時期，熟練勞工的重要性。

The AI industry is currently struggling to balance fast technological growth and the need for large investments with a US government that is increasingly intervening to protect national security.

AI 產業目前正努力在快速的技術成長、龐大的投資需求，以及日益介入以保護國家安全的美國政府之間取得平衡。