Before offering generative AI services to the public, conduct required safety assessments and algorithm registration; mark all generated content
If you offer generative AI services with public influence or social mobilization potential to the general public, you must perform a safety assessment and register the algorithm per national regulations before launch. Any changes or shutdowns also require filing updates or cancellation paperwork. All generated images and videos must be clearly marked. Tools meant solely for personal use and not open to the public are exempt from these requirements. The regulation itself does not specify penalties; violations are handled under the Cybersecurity Law, Data Security Law, and Personal Information Protection Law.
No money spent, but you must go through the cyberspace aut…
Providing generative artificial-intelligence services with public-opinion attributes or social-mobilization ca…
No money spent, but you must go through the cyberspace authority’s process, which takes some time. Tools that do not provide services to the domestic public are outside this measure
Providing generative artificial-intelligence services with public-opinion attributes or social-mobilization capacity shall conduct a security assessment per state rules. Algorithm filing under the algorithm-recommendation management rules is also required, plus later amendment and cancellation procedures. Public-opinion attributes or social-mobilization capacity means the ability to affect public views or to organize and mobilize people. Providers shall mark generated content such as images and video. The measure states that when enterprises, research institutions, and others develop or apply generative AI technology and “do not provide generative AI services to the domestic public, this measure does not apply.” Violations of the measure are “punished in accordance with the (中华人民共和国网络安全法), (中华人民共和国数据安全法), (中华人民共和国个人信息保护法), (中华人民共和国科学技术进步法), and other laws and administrative regulations.” The measure took effect on 15 August (8) 2023 (nationwide)
国家互联网信息办公室等七部门 (2023). 生成式人工智能服务管理暂行办法（第二、十二、十七、二十一条）. https://www.gov.cn/zhengce/zhengceku/202307/content_6891752.htm
Open source linkThese measures do not specify fine amounts themselves. Penalties for violations must be located in the Cybersecurity Law, Data Security Law, and Personal Information Protection Law, so the grade is B. This section has not verified original civil cases concerning AI-generated content infringing copyright or likeness rights on the Supreme People’s Court website and does not cite them. For proxy sites reselling foreign large-model API access cheaply, see Item 19 in this section (AI proxy sites).