(By You Yunting) Recently, many videos of celebrities speaking foreign languages have appeared on the Internet, such as Guo Degang and Zhao Benshan telling English jokes, and Taylor Swift and Donald Trump speaking Chinese, which is actually supported by using AI services of Heygen. I visited Heygen’s website and found that it was not available to domestic users with no WeChat and Alipay in the service payment channel, but in fact Heygen’s founders are two Chinese, both graduating from Tongji University. Today I would like to discuss the formalities Heygen needs to go through in order to have its AI services land in China.
The process of Heygen’s translation service, according to my actual test, is as follows: users need to log into Heygen’s website and upload a video containing human voice with more than 30 seconds, and then Heygen will translate the content of the video and generate a video in which the speaker in the video speaks the target language. It is amazing that the generated video uses the voice of the speaker in the original video, and the mouth shape matches the pronunciation.
It turns out that HeyGen’s service has access to at least three other AI engines[1], ChatGPT-4 is responsible for translating the speech in the video, then 11labs’ voice cloning is used to extract the voice characteristics of the speaker in the video and make the speaker sound in the newly generated video, and wav2lip-2 deals with lip synchronization for the lips of the person in the newly generated video.
I. The services provided by Heygen are deep synthesis services
According to China’s regulations, HEYGEN shall belong to deep synthesis technology and shall be regulated. According to the Provisions on the Administration of Deep Synthesis of Internet-based Information Services, deep synthesis technology refers to the technology that utilizes synthetic algorithms, such as deep learning and virtual reality, to produce text, images, audio, video, virtual scenes and other network information.
ChatGPT-4’s translation of speech in the video belongs to the technology of generating or editing text; 11labs’s voice clone extracting the voice characteristics of the speaker in the video belongs to the technology of generating or editing speech; wav2lip-2’s lip synchronization for the lips of the person in the newly generated video belongs to the technology of generating or editing biometric features in images and videos. All of the aforementioned fall within the business scenarios stipulated in the Provisions on the Administration of Deep Synthesis of Internet-based Information Services. Deep synthesis service providers have many obligations, including specifically:
1. Routine compliance obligations
Such as no activities endangering national security and social public interests and infringing upon the legitimate rights and interests of others; no producing or disseminating false news; establishment and improvement of management systems for user registration, algorithmic mechanism review, scientific and technological ethics review, information release review, data security, personal information protection, anti-telecommunication network fraud and emergency response; and having safe and controllable technical safeguard measures.
2. AI service obligations
Where editing functions for biometric information such as face and voice is provided, the deep synthesis service users shall be reminded to inform the individual being edited according to law, and obtain their separate consent. Information content generated or edited by using their services shall be added signs to.
3. Filing and assessment procedures
Does Heygen need to go through filing and security assessment if its services are to be provided in China? The Measures for the Administration of Generative Artificial Intelligence Services (Exposure Draft) issued in April 2023 stipulates that before using generative artificial intelligence products to provide services to the public, a provider shall apply for security assessment with the Cyberspace Administration of China in accordance with the Provisions on the Security Assessment of Internet Information Services with Public Opinion Attribute or Social Mobilization Capability, and perform the procedures for algorithm filing and modification and cancellation of filing in accordance with the Provisions on the Administration of Algorithmic Recommendation of Internet Information Services. Accordingly, Heygen and ChatGPT-4, 11labs, wav2lip-2 it has access to should go through security assessment and algorithm filing.
However, the corresponding article is amended in the Interim Measures for the Administration of Generative Artificial Intelligence Services issued in July 2023 to read: those who provide generative artificial intelligence services with public opinion attribute or social mobilization capability shall carry out security assessment in accordance with relevant national regulations and perform the procedures for algorithm filing and modification and cancellation of filing in accordance with the Provisions on the Administration of Algorithmic Recommendation for Internet Information Services. In other words, only generative artificial intelligence services with public opinion attribute or social mobilization capability need to go through security assessment and algorithm filing.
According to my understanding of the practice, security assessment and algorithm filing are required for Wenxin Yiyan and other products similar to Chatgpt providing their services in China. The reason why products linked to Chatgpt through API in China are stopped may be that they fail to go through security assessment and algorithm filing. Therefore, if Heygen services having access to Chatgpt wants to land in China, it also needs to undergo security assessment and algorithm filing.
China’s regulation of the Internet is relatively strict. If the application of deep synthesis technology is provided to the public in China, other domestic laws and regulations to be complied with include the Cybersecurity Law of the People’s Republic of China, the Administrative Regulation on Internet Information Services, the Provisions on the Administration of Deep Synthesis of Internet-based Information Services, the Interim Measures for the Administration of Generative Artificial Intelligence Services, the Regulations on the Safety Assessment of Internet Information Services with Public Opinion Attribute or Social Mobilization Capability, etc..
II. Services provided by Heygen involves export of sensitive personal information
There are two types of export of personal information: one is that the data processor transmits and stores the data collected and generated within the territory of China abroad; the other is that the data processor stores the data within the territory of China, but overseas institutions, organizations or individuals can access or procure the data with the identity information blocked. If Heygen would like to provide services in China, the problems of data export can be solved partly by setting up servers in China and restricting identity information from transferring abroad.
However, Heygen’s services requires users to upload a video that directly contains the users’ identifiable portraits and voices, from which people’s physiological characteristics, such as voice prints, facial features and lip shapes, will be extracted upon analysis by wav2lip-2 and 11labs. These are biometric traits and belong to sensitive personal information. If Heygen intends to transfer abroad the sensitive personal information obtained in China, it should also conduct the security assessment of outbound data transfer and list the amount and field of sensitive personal information in its application materials. Since its services targets the public, once provided, the regulatory trigger point of sensitive personal information of more than 10,000 people will certainly be reached.
So even if Heygen’s servers are in China, if the servers of Chatgpt, wav2lip-2, and 11labs that support its services are in foreign countries, Heygen still has to fulfill the compliance obligation of outbound transfer of sensitive personal information. Meanwhile, IP access to Chatgpt services in China is still blocked, and Chatgpt’s AI services inserted in Microsoft’s Bing search engine cannot be used in China. Even if Heygen wants to provide its services for Chinese users, OpenAI, the developer of Chatgpt, may not agree. According to the above analysis, it seems that under the current policy and legal environment, Heygen will face a lot of compliance challenges if it wants to provide services in China.
Lastly, I will discuss the risk of infringement upon personal rights and intellectual property rights. Since Heygen provides services of one-to-one video translation and conversion for users, and the website does not release the converted content to the public, its risk of infringement upon intellectual property rights is low. However, if a user uses Heygen to generate translated content from others’ speech video and publishes it online, a lot of authorization is needed.
For example, if you are to publish Guo Degang’s cross-talk video, you need to obtain authorization from Guo Degang as for the portrait right, the voice right, the producer’s right of the cross-talk video, and the copyright of the cross-talk text content, the lack of any of which may constitute infringement. If you are interested, you can also refer to my previous article What are the IP and other legal issues regarding “AI Sun Yanzi”?[2] and get more detailed discussion. Of course, Guo Degang, as a public figure, may not necessarily protect his right against this kind of harmless dissemination, but if he wants to, the video publisher will certainly lose the lawsuit.
Footnote:
[1] https://www.techopedia.com/definition/heygen
[2] https://mp.weixin.qq.com/s/gN9dK_oLe4rsCBYx23sS1g
Lawyer Contacts
You Yunting
86-21-52134918
Short Link: