大模型发展对人工智能安全风险治理的挑战和应对分析.pdfVIP

Topic Focusing 热点专题 大模型发展对人工智能安全风险治理的 挑战和应对分析 * 邱惠君，张瑶 国家工业信息安全发展研究中心，北京，100040 摘要： 以 ChatGPT 为代表的大模型发展正在掀起新一轮全球科技革命，将会重塑产业链和 价值链。当前，大模型研发竞争白热化，相关产品“以日为单位”快速更新，文本、图片、代码、 视频的生成工具能力不断提升，并且快速应用至垂直行业领域。但大模型具有算法不透明、数 据强依赖、易被滥用等人工智能固有的风险，与其平台性、易使用、超能力等特点相互交织，风 险快速传递并极速放大。短短数月，敏感数据泄露、虚假信息生成、漏洞攻击加剧等安全风险 已经泄露，给人工智能治理带来新挑战，引发热议和担忧，多个国家和地区已经着手研究和探 索应对举措，全球性的大模型治理迫在眉睫。本文分析了大模型的发展现状、安全风险和治理 进展，并建议当下聚焦建立责任体系、加强技术攻关、推进行业自律、强化宣传引导、参与全球 治理五大维度推进安全风险治理。 关键词： 人工智能；大模型；人工智能安全；治理体系 Analysis of Challenges and Countermeasures of the Development of Large Models to AI Security Risk Governance * Qiu Huijun, Zhang Yao China Industrial Control Systems Cyber Emergency Response Team, Beijing, 100040 Abstract:Large models, represented by ChatGPT, of which development has been setting off a new wave of global scientific and technological revolution,and will reshape the industrial chain and value chain. Currently, the competition for research of large models is intensifying. The update speed of related products is “in days”. The performance of tools for generating texts, images, codes, and videos is continuously improving, and the application of large models is extending to vertical industry fields rapidly. However,large models have inherent risks of AI such as opaque algorithms,strong data dependence, and vulnerable to abuse,and these risks can be intertwined with its characteristic of platformism,ease of use, and superpower,so the risks can be transmitted and amplified rapidly.In just a few months, security risks have been exposed such as leakage of sensitive data, generation of false information and increased