文生视频类人工智能的数据风险及其法律规制——以Sora为例

On Data Risk and Legal Regulation of Vincennes Video Artificial Intelligence——Taking Sora as an Example

Sora具有较强的视频生成能力,被OpenAI定位为“物理世界模拟器”,引领着文生视频类AI的发展。数据是文生视频类AI的质料,在AI运行过程中可能存在安全风险,具体包括预训练数据来源合法性风险,人工操纵标注的数据可能导致的意识形态等风险,抓取虚假数据或非法数据的风险等。而这些数据风险可能衍生出国家与个人的数据安全风险、生成视频数据的著作权风险、深度伪造视频等虚假视频数据传播风险以及意识形态渗透风险。建立健全多维法律规制体系成为文生视频类AI数据风险治理的关键所在。我国应以科技向上向善、包容审慎等为原则,明确服务提供者等主体的数据安全法律义务,健全数据分类分级保护制度,完善数据伦理规范,建立健全包容审慎数据监管、风险防控数据监管以及全过程风险预警与应急处置监管相关机制,明确公众数据风险防范与救济路径。

Sora has a strong ability to generate video,and is positioned as a“physical world simulator”by OpenAI,leading the development of text generation video AI.Data is the material of text-generated video AI,and there may be security risks in the operation of AI,including:pre-training data source legitimacy risk;manual manipulation of labeled data can lead to ideological and other risks;the risk of grabbing fake or illegal data.These data risks may result in national and personal data security risks,copyright risks of video data generation,transmission risks of fake video data such as deep forged videos,and ideological penetration risks.The establishment of a sound multi-dimensional legal regulation system has become the key to the risk management of text-generated video AI data.China should follow the principles of advancing science and technology,inclusiveness and prudence,clarify the legal obligations of data security for service providers and other subjects,improve the data classification and classification protection system,improve data ethics,establish and improve the inclusive and prudent data supervision,risk prevention and control data supervision,as well as the whole process of risk early warning and emergency response supervision mechanisms,and clarify the public data risk prevention and relief paths.