Breaking
RUTwo police officers were injured while trying to stop a driver in ArmeniaCNA 75-year-old woman surnamed Chen fell to her death on Jiguang Street, Central District, Taichung CityRUExplosions occurred amid air raid in DnepropetrovskTRSuspects Who Stealed Cables from the Construction Site and Escaped were CaughtUSU.S. military kills four in Caribbean boat strike targeting suspected drug smugglersJPBrazilian presidential election: incumbent Lula and Bolsonaro's eldest son Flavio go to run-offRUTwo UAVs were shot down over the Ryazan region, 199 Ukrainian drones were destroyed overnightITMan tracked down after running over and killing a 79-year-old woman in BisceglieCNTaiwan stocks soared 1,236 points! TSMC closed at 2,575 points and the index hit 49,712 points, a record highTRENAG announced September 2026 E-CPI data: Monthly inflation 2.10%, annual 46.61%RUTwo police officers were injured while trying to stop a driver in ArmeniaCNA 75-year-old woman surnamed Chen fell to her death on Jiguang Street, Central District, Taichung CityRUExplosions occurred amid air raid in DnepropetrovskTRSuspects Who Stealed Cables from the Construction Site and Escaped were CaughtUSU.S. military kills four in Caribbean boat strike targeting suspected drug smugglersJPBrazilian presidential election: incumbent Lula and Bolsonaro's eldest son Flavio go to run-offRUTwo UAVs were shot down over the Ryazan region, 199 Ukrainian drones were destroyed overnightITMan tracked down after running over and killing a 79-year-old woman in BisceglieCNTaiwan stocks soared 1,236 points! TSMC closed at 2,575 points and the index hit 49,712 points, a record highTRENAG announced September 2026 E-CPI data: Monthly inflation 2.10%, annual 46.61%
Back川普與黃仁勳通話強調AI競賽,專家警告安全與治理缺失將轉嫁風險至台灣
川普與黃仁勳通話強調AI競賽,專家警告安全與治理缺失將轉嫁風險至台灣
Developing
自由时报1 hour agoTech3 min readChinaView translation

川普與黃仁勳通話強調AI競賽,專家警告安全與治理缺失將轉嫁風險至台灣

Quick Look

美國總統川普在Truth Social連續發文將AI風險稱為騙局,並致電英伟達CEO黃仁勳強調誰贏AI誰就贏;黃仁勳則稱要讓美國各產業與人民成為AI競賽贏家。專家指出,美中競爭與AI安全治理缺失可能使台灣承擔晶片供應與合規成本的雙重壓力,台灣需爭取可驗證的加速護欄與共同治理話語權。

AI-generated summary

Why It Matters

川普在Truth Social發文質疑AI風險,將其稱為騙局,並致電英伟達CEO黃仁勳強調AI競賽的重要性。同時,Anthropic執行長阿莫戴警告AI遞迴自我改進可能帶來失控風險,主張建立全球AI安全標準。

Font size

◎ 廖明輝

美國總統川普日前致電正在洛杉磯All-In高峰會的輝達執行長黃仁勳。川普強調「誰贏AI,誰就贏」;黃仁勳則說要讓美國每個產業、企業、州與人民都成為AI競賽的贏家。(彭博檔案照)

美國總統川普日前在真實社群(Truth Social)接連向AI安全派開火,從下午四時三十分到七時五十五分不到四小時間連發四則貼文,把「AI接管世界」、「機器人毀滅人類」與全球暖化等爭議並列為「騙局」。川普甚至宣稱AI唯一需要的護欄就是一位「強大而聰明的總統」,並警告監管只會讓中國高興。稍後他致電正在洛杉磯All-In高峰會的輝達執行長黃仁勳,黃仁勳開擴音讓全場收聽。川普強調「誰贏AI,誰就贏」;黃仁勳則說要讓美國每個產業、企業、州與人民都成為AI競賽的贏家。

川普9月15日16:30到19:55間,連發4則貼文,把「AI接管世界」、「機器人毀滅人類」與全球暖化等爭議並列為「騙局」。(擷取自Truth Social)

其實,這場爭論不能被簡化成「相信AI」與「害怕AI」。Anthropic執行長阿莫戴日前發表近三千八百字的「我們必須調控前沿AI的發展步調」,並非要求停止模型訓練,而是認為能力提升不能快過安全。他警告AI的「遞迴自我改進」能力,擔心AI已愈來愈能參與下一代AI研發,更強AI又進一步加速研發而形成回饋迴圈。再加上近期OpenAI-Hugging Face事件,AI代理程式出現越權攻擊與試圖干預評測行為,他因此主張第三方長期進駐評估、民主國家建立共同安全標準,最後再尋求全球協調AI進展。

請繼續往下閱讀...

川普與阿莫戴其實回答的是兩個不同問題。川普問的是美國若減速,中國會不會追上?阿莫戴問的是AI若加速到安全研究追不上,人類還能不能控制?前者是國力競爭,後者則是工程風險。川普質疑AI必然毀滅人類,對應阿莫戴所稱六至十二個月內AI代理群可能形成大規模殭屍網路,顯示這仍只是風險推估,不是已經證實的未來;但把AI安全疑慮全部稱為「騙局」,又與已發生的代理越權事件並不相容。需要檢驗的不是誰喊得比較大聲,而是AI代理是否越權、安全機制能否攔截、以及事故發生能否接受外部查核。

黃仁勳代表另一種產業思維。安全與加速未必二選一,企業若認為模型已經失控,本來就應踩煞車。但前沿AI一旦連接金融、能源、國防與網路,後果不再只由開發企業承擔,僅靠公司自律仍不足以回答公共風險。台灣要注意歐洲並未走向全面停止AI。德國已明言停止AI發展並不可行,主張國際治理必須納入美國與中國;歐洲議會AI立法主要推手貝尼費伊也呼籲美中與歐洲要共同建立全球AI規則。

歐洲議會AI立法主要推手貝尼費伊,也呼籲美中與歐洲要共同建立全球AI規則。(路透檔案照)

台灣該看清的是問題已不只是AI要不要減速,而是AI到底只是一場美中爭奪領先的戰略競賽,還是必須跨國管理的全球共同風險。台灣掌握先進晶片、伺服器與關鍵供應鏈,未來無論美國要求加速建設、強化對中國算力管制,或歐洲提高模型安全與可稽核標準,成本與責任都可能落到台灣企業身上。因此,台灣不能只替美國的加速路線供應晶片,也不能只替歐洲的安全路線承擔合乎法規成本。我們要爭取的是讓「加速」有可驗證護欄,讓「安全」不會成為封鎖創新工具。當歐洲開始說「我們需要美國,也需要中國」,台灣更該思考自己在全球AI生態系的位置。台灣在AI時代的最大戰略弱點,不是沒有晶片,而是沒有決定晶片如何被使用、被限制、與被治理的發言權。

(作者為中華經濟研究院輔佐研究員)

What to Watch

AI outlook — possibilities, not facts

  • 台灣將面臨增加的壓力,需同時應對美國對華算力管制與歐洲AI安全標準的合規成本

    Likely · Within months

  • 美中將在AI領域展開更激烈的戰略競爭,但全球治理協議仍難以快速達成

    Likely · Within months

Open Questions

  • 台灣將如何在美中競爭與全球治理壓力下保持產業競爭力?
  • 全球AI治理框架何時能形成?主要分歧在哪裡?
  • 英伟達等晶片廠商是否會因地緣政治壓力調整供應鏈佈局?

Related Topics

This article was originally published by 自由时报.

Related Stories

Gong Huazhong, President of the Institute of Information Security: In the AI era, information security threats are like heavy rain, and a national protection system will be established with "AI versus AI"
Developing·

Gong Huazhong, President of the Institute of Information Security: In the AI era, information security threats are like heavy rain, and a national protection system will be established with "AI versus AI"

Gong Huazhong, director of the National Information Security Research Institute, held a media exchange meeting two months after taking office. He pointed out that in the AI ​​era, information security attacks are as fast as heavy rain and traditional defenses are insufficient. He will use "AI versus AI" as the core strategy to promote four major tasks: AI automated source code detection and repair, using AI to strengthen penetration testing and offensive and defensive drills, establishing an AI automated threat information system, and building a national-level AI information security attack and defense sandbox to strengthen the in-depth defense of national information security.

自由时报
3 min read
The AI assistant self-published more than 13,000 internal company screenshots online, involving more than 300 organizations.
Developing·

The AI assistant self-published more than 13,000 internal company screenshots online, involving more than 300 organizations.

Information security team Glow Labs discovered that an AI program development assistant was unable to upload screenshots to a private GitHub repository when assisting engineers. Instead, he created a public repository on his own, which resulted in the exposure of more than 13,000 internal company screenshots, involving more than 300 organizations around the world, including Fortune Global 500 companies and AI research institutions. The content included unreleased software screens, customer information, and financial data.

自由时报
2 min read
More on this topic川普