Chinese censors reportedly test AI models to embody 'socialist values'

Chinese authorities are reportedly reviewing the large language models adopted by AI companies in the country to check if their systems “embody core socialist values”.  

According to a report from the Financial Times, China's cyberspace watchdog Cyberspace Administration of China (CAC) has been carrying out the mandatory government review. 

Tech giants and small startups such as ByteDance, Alibaba, Moonshot and 01.AI will be involved in the review.  

Local CAC officials will test the AI models by asking the companies a wide range of questions, mostly related to politically sensitive topics and Chinese president Xi Jinping, according to the report. The model's training data and safety processes will also be reviewed. 

An anonymous source from a Hangzhou-based AI company told Financial Times that its model failed the first round of testing for unknown reasons. It only passed the second time after months of "guessing and adjusting," the person said in the report.  

"Security filtering" is required to meet the censorship policies and the rules have been made complex as Chinese LLMs are still trained mostly on English language content, multiple engineers and industry insiders told the FT. 

According to the report, the authorities will remove “problematic information” from AI models training data and a database of sensitive words and phrases will be created. 

Back in June, CAC, together with Chinese Ministry of Industry and Information Technology, National Development and Reform Commission (NDRC) and Standardisation Administration of China (SAC), rolled out the guidelines for "The Construction of a National Comprehensive Standardisation System for the Artificial Intelligence Industry (2024 Edition)".

Inspired by Chinese president Xi's thoughts on socialism, the guidelines will fully implement the spirit of the 20th National Congress of the Communist Party of China and the second plenary session of the 20th Central Committee of the Communist Party of China, conscientiously implement the deployment requirements of the Central Economic Work Conference and the National Conference on Promotion of New Industrialisation.

This aims to seize the development opportunities of the artificial intelligence industry, improve the top-level design of artificial intelligence standards work, strengthen the coordination of standards work across the entire industry chain, and coordinate and promote the research of standards , formulation, implementation and internationalisation to provide technical support for promoting the high-quality development of the AI industry.

Related articles:

China's cyberspace watchdog fines Didi Global for illegally processing personal data
China cyberspace watchdog revises rules governing mobile apps