China has approved another 487 artificial intelligence (AI) algorithms that can be used for deepfakes, covering products from domestic technology giants such as Baidu, Alibaba Group Holding and Tencent Holdings, along with some foreign firms including Hewlett-Packard (HP).
The new list from the internet watchdog Cyberspace Administration of China (CAC) is the second-largest since the regulation came into effect in January 2023. Included algorithms are those used in products like Baidu’s portrait image diffusion generator that is bundled with its cloud gallery app Yike, and a search algorithm in Tencent’s WeChat, according to the list released by the CAC on Monday.
Alibaba, owner of the South China Morning Post, registered an algorithm to assist with creating documents in its enterprise collaboration tool DingTalk, and a multimodal algorithm designed for corporate clients to generate images and video, which was developed by its research arm Damo Academy.
The efforts are in compliance with the Administrative Provisions on Deep Synthesis for Internet Information Services, which regulates so-called deepfake technologies that use deep learning and augmented reality to generate text, images, audio and video, or to create virtual scenes that mimic people and things in real life. Failure to register an algorithm with the CAC could result in removal from domestic app stores.
The CAC has published six other whitelists, which until this month had been growing longer with each subsequent list. The batch of 492 in June was the largest so far, while the April and February groups included 394 and 266 algorithms, respectively. The first list in June 2023 had just 41.
Cai Peng, partner at Beijing-based Zhong Lun Law Firm, said the growing size of the lists reflects a “clearer” work flow between the regulator and applicants.
The roughly two-month process involves filing and polishing application documents upon the CAC’s request. “It seems that the regulator has established a more efficient process to work with the applicants,” he said.
The latest batch included some other notable tools, including a healthcare knowledge algorithm for Douyin – the Chinese version of TikTok from ByteDance – a music generator from Microsoft’s AI spin-off Xiaoice, and a character dialogue generator for NetEase’s digital version of the popular party game Werewolf.
Multiple foreign brands made the cut too. HP’s computer assistant algorithm was on the list, along with a text generator by Yum China, a spin-off of American fast food giant Yum! Brands, for dine-in and delivery services at KFC and Pizza Hut, among other chains.
China has taken an early lead among countries seeking to regulate AI. The country requires developers to register related technologies with the government. Generative AI (GenAI) models must be registered with the CAC before being provided publicly under a regulation that took effect in August 2023. The regulator has so far released two lists of a combined 188 approved GenAI models.
