The Cyberspace Administration of China (CAC) has unveiled draft regulations on ChatGPT-like generative AI products. The new rules require service providers to avoid various forms of discrimination, fake news, terrorism and other anti-social content.

The regulations are aimed specifically at generative AI products and services, and require providers to ensure data used to train their AI programs is legal, real, non-discriminating and diversified. Additionally, profiling of users based on input or metadata is banned, and personal information must be edited, removed or blocked upon request.

If any banned content is discovered or reported, providers must re-train their models within three months to prevent recurrence. The draft regulations also list detailed requirements for manual tagging or labeling of data used to train AI models.

Violations of the regulations can result in fines of up to 100,000 yuan (approximately $14,520) and service termination. The public can submit their feedback on the draft regulations to the CAC via internet or mail until May 10.