Chinese govt limits what biometrics and data can be used to train generative AI

Data used to train AI should be subject to a security assessment to identify “illegal and harmful information” and organizations should seek consent from individuals whose biometric data is used in training, say new requirements proposed by a committee made up of officials from two Chinese government agencies and the police.

According to a report by Reuters, the draft requirements would see any body of data assessed to have more than 5 percent of said undesirable content blacklisted from being used to train public-facing AI models.

Among the kinds of information deemed harmful or illegal is anything that advocates terrorism or violence – a standard security concern. But in keeping with Beijing’s tightly-monitored, autocratic security stance, the draft requirements also flag content that involves “overthrowing the socialist system,” “damaging the country’s image,” and “undermining national unity and social stability,” and bans any content that has been censored on the Chinese internet.

Among members of the National Information Security Standardization Committee, which published the draft, are officials from the Cyberspace Administration of China (CAC), which has been lobbying since April to have security assessments for firms developing generative AI intended for public use. Representatives from the Ministry of Industry and Information Technology and the police are also on the committee.

While censorship watchdogs may see another mechanism of government control, the requirements are intended to support China’s drive to rival the U.S. in the AI sector and become a world leader in the field by the end of the decade. As such, they also include detailed guidelines on how to avoid intellectual property violations.

China already exports AI-based tech to roughly twice as many countries as the U.S.

Article Topics

AI  |  biometric data  |  biometrics  |  China  |  generative AI  |  standards  |  training