社團指南

中國的跨境數據規則與人工智慧培訓

China’s 2024 cross-border data provisions set thresholds and exemptions for exporting personal information and important data.

聆聽此指南AI 語音敘述 · ~3 min

本頁閱讀時間3分鐘
  1. 概述
  2. 深入探討
  3. 現實世界的實施
  4. The Future of China’s Cross-Border Data Rules and AI Training
  5. 付諸實踐
  6. 不斷探索
  7. 常見問題

概述

AI training is not categorically exempt: teams must classify data, identify outbound transfers, and select the applicable assessment, standard-contract, certification, or exemption route.

深入探討

The Cyberspace Administration of China’s Provisions on Promoting and Regulating Cross-Border Data Flow took effect on 22 March 2024. They clarify when data processors must apply for a security assessment, use a standard contract, or obtain personal-information protection certification. Data not identified or publicly announced as important data need not be declared important solely for a security assessment. Some specified business, research, and operational transfers without personal information or important data are exempt from these transfer mechanisms. Other limited personal-information exemptions cover certain contractual needs, cross-border HR management, emergencies, and small volumes. For a non-critical-information-infrastructure operator, exporting important data or at least one million people’s non-sensitive personal information, or at least 10,000 people’s sensitive personal information, generally triggers a data-export security assessment, subject to listed exemptions. Exporting 100,000 to fewer than one million people’s non-sensitive personal information, or fewer than 10,000 people’s sensitive personal information, generally requires a standard contract or certification, again subject to the stated exceptions. Critical information infrastructure operators have separate assessment triggers. Free-trade zones may create approved negative lists that affect which transfers require these mechanisms. AI training creates a data-flow question, not a special blanket category. Uploading Chinese personal information to a foreign cloud, sharing training records with an overseas model provider, or allowing an offshore team to access data may involve an outbound transfer. The processor should identify individuals, sensitivity, volume, purpose, recipient, and whether information is important data; then assess PIPL notices, separate consent, impact assessment, and export mechanism obligations. Synthetic or de-identified information is not automatically exempt if reidentification or personal information remains. Keep a transfer inventory tied to datasets, recipients, model-training stages, and subprocessors. Document exemptions and thresholds, check local free-trade-zone lists, and reassess when training expands to new people or data categories. The 2024 rules simplify selected transfers but retain security, personal-information, and contract duties.

現實世界的實施

A research team confirms that an overseas training vendor cannot access personal information before claiming a data-only exemption.

A company counts individuals and classifies sensitive information before selecting an export mechanism.

A business checks whether a free-trade-zone negative list affects its proposed transfer.

A model team records notices, separate consent, contracts, and impact-assessment evidence for cross-border training data.

The Future of China’s Cross-Border Data Rules and AI Training

China’s CAC continues to refine cybersecurity and data-export administration, while free-trade zones can maintain different approved negative lists. Organizations should check the latest official rules before cross-border training or vendor changes. A threshold exemption removes specified transfer mechanisms only; it does not erase PIPL processing duties, data-security obligations, or other sector restrictions. Maintain an annual threshold counter and refresh it when a dataset or recipient changes. Record the rule used for each dataset and recipient, and check whether annual thresholds have been reached before a new export.

付諸實踐

為什麼這很重要

  • 風險與安全. 災難性和日常的人工智慧危害都取決於誰了解風險以及誰能夠採取行動。
  • 更明確的決策. 民眾和專業素養決定強而有力的安全政策在政治上是否可行。
  • 突破炒作. 清晰的解釋可以減少炒作、實驗室公關和模糊道德劇場的影響。

留意

  • 將存在風險視為科幻小說,同時能力複合。
  • 混淆了表面產品安全與高度自治下的對準。
  • 只給非英語和非專業觀眾留下低品質的資源。

後續步驟

  1. 單獨的產品危害、誤用和失控/失調風險。
  2. 詢問哪些證據會改變您對時間表和嚴重性的看法。
  3. 比起行銷主張,更喜歡主要來源和具體評估。
  4. 確定一條行動路徑:職業、政策、資金或技能——而不僅僅是意識。

不斷探索

免費電子報

獲取每日 AI 簡報

每個工作日上午提供三則經過驗證的 AI 新聞,以淺顯易懂的英文撰寫。永久免費,無廣告。

每個工作日一封電子郵件。一鍵取消訂閱。我們絕不出售或分享您的地址。

測試一下自己

Take the China’s Cross-Border Data Rules and AI Training quiz

每個答案皆有即時回饋,通過課程後可獲得具備可驗證 ID 的可分享證書。

開始測驗

支持免費的 AI 教育。AI Understanding 是一個 501(c)(3) 非營利組織——絕無廣告,絕無付費牆。 捐款

常見問題

What is China’s Cross-Border Data Rules and AI Training?

China’s 2024 cross-border data provisions set thresholds and exemptions for exporting personal information and important data. AI training is not categorically exempt: teams must classify data, identify outbound transfers, and select the applicable assessment, standard-contract, certification, or exemption route.

中國2024年跨國數據規定何時生效?

CAC issued the provisions on 22 March 2024, effective on publication.

每年多少量的非敏感個人資訊通常會觸發非 CIIO 的安全評估?

The rules specify the one-million threshold for non-sensitive personal information.

非敏感個人資訊通常何時需要標準合約或認證?

The provisions set this transfer band for the non-CIIO route.

該規則是否將所有人工智慧訓練資料視為不受出口管制?

AI training has no blanket exemption; ordinary data-transfer provisions apply.

根據規定,自貿區可以做什麼?

Approved FTZ lists can specify data subject to transfer mechanisms.