概述
They cover content standards, training data, labeling and user protection, and they require providers with public opinion influence to complete security assessments and algorithm filing. They matter because they made China one of the first major economies with binding rules written specifically for generative AI.
深入探讨
The Cyberspace Administration of China (CAC) issued the Interim Measures in July 2023 together with six other agencies, including the National Development and Reform Commission, the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, and the Ministry of Public Security. They apply to generative AI services offered to the public within mainland China. Research and development that is not offered to the public is largely excluded. The content rules require providers to uphold "core socialist values" and not generate content that undermines state power, national unity or social stability, or that spreads false or harmful information. Providers must also take steps against discrimination, protect intellectual property and trade secrets, and respect personal rights such as privacy and reputation. On training data, providers must use lawful sources, avoid infringing intellectual property, have a legal basis such as consent for personal information, and improve the quality, truthfulness, accuracy, objectivity and diversity of their data. Human data labeling must follow clear rules and be checked for quality. The measures connect to earlier regulations. Services with "public opinion attributes or social mobilization capacity" must undergo security assessments and file their algorithms under the 2022 algorithm recommendation rules, and generated content must be labeled under the deep synthesis rules. The CAC maintains the filing system and publishes lists of filed services. The April 2023 draft was stricter. It required generated content to be true and accurate, which would have been nearly impossible for language models that can hallucinate. The final text softened this to a duty to take effective measures to improve accuracy and reliability, and it added language supporting innovation. The measures do not ban all foreign AI. Instead, they let the CAC take technical measures against services from outside China that violate the rules.
战略影响
风险与安全
灾难性和日常的人工智能危害都取决于谁了解风险以及谁能够采取行动。
更清晰的判决
公众和专业素养决定强有力的安全政策在政治上是否可行。
打破炒作
清晰的解释可以减少炒作、实验室公关和模糊道德剧场的影响。
The Future of China's Generative AI Interim Measures
The measures are labeled interim. A comprehensive AI law has appeared on the State Council's legislative work plans, but none has been adopted. Meanwhile, China keeps adding targeted rules, such as mandatory labeling requirements and new national standards, so the overall system grows in layers. How strictly content rules are applied can shift with political priorities and with CAC enforcement campaigns. For global companies, a key open question is how China's filing-based approach will interact with other countries' AI rules, especially where training data and model access cross borders.
现实世界的实施
Before a Chinese company can open its chatbot to the public, it typically completes a security assessment and files its algorithm with the Cyberspace Administration of China. Only after its filing is accepted does it launch widely.
A provider building a training corpus has to screen for illegal content, use lawfully sourced data, respect intellectual property, and get consent or another legal basis before using personal information.
A consumer AI app must set up complaint and reporting channels, label AI-generated images and video under China's deep synthesis rules, and take steps to keep minors from becoming overly dependent on the service.
A company using a large model only internally for research, without offering it to the public in China, generally falls outside the measures' main obligations.
风险与防护栏
将存在风险视为科幻小说，同时能力复合。
混淆了表面产品安全与高度自治下的对准。
只给非英语和非专业观众留下低质量的资源。
实施路线图
单独的产品危害、误用和失控/失调风险。
询问哪些证据会改变您对时间表和严重性的看法。
比起营销主张，更喜欢主要来源和具体评估。
确定一条行动路径：职业、政策、资金或技能——而不仅仅是意识。
不断探索
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the China's Generative AI Interim Measures quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
常见问题
What is China's Generative AI Interim Measures?
China's Interim Measures for the Management of Generative Artificial Intelligence Services, in force since August 15, 2023, set the rules for chatbots, image generators and other generative AI offered to the public in mainland China. They cover content standards, training data, labeling and user protection, and they require providers with public opinion influence to complete security assessments and algorithm filing. They matter because they made China one of the first major economies with binding rules written specifically for generative AI.
When did China's Interim Measures for generative AI services take effect?
The CAC and six other agencies issued the measures in July 2023, and they took effect on August 15, 2023.
Which activity generally falls outside the measures' main obligations?
The measures apply to generative AI services provided to the public in mainland China. Research and development that is not offered to the public is largely excluded.
Which services must undergo a security assessment and algorithm filing?
The measures require services with public opinion attributes or social mobilization capacity to complete security assessments and algorithm filing under the earlier algorithm recommendation rules.
Which agency leads enforcement of the measures?
The Cyberspace Administration of China issued the measures with other agencies, runs the filing system and leads enforcement.
How did the final measures change the April 2023 draft's requirement on accuracy?
The draft required generated content to be true and accurate, which is impractical for models that can hallucinate. The final text changed this to a duty to take effective measures to improve accuracy and reliability.
继续学习
相关指南
为此主题精选的更多指南