Beijing Institute of AI Safety and Governance (Beijing-AISI)

The Beijing Institute of AI Safety and Governance (Beijing-AISI), whose full Chinese name is 北京前瞻人工智能安全与治理研究院 (Beijing Foresight Institute for AI Safety and Governance), is a government-backed research institute located in the Jingxi Smart Valley within Zhongguancun Mentougou Park in Beijing, China. The institute originated as the Beijing AI Safety and Governance Laboratory, which was established on September 3, 2024, by multiple bureaus within Beijing's local government. After approximately eight months of operation as a preparatory body, the formal research institute received approval from Beijing's Civil Affairs Bureau and was officially inaugurated on May 9, 2025. The Beijing Municipal Commission of Economy and Information Technology serves as its supervising authority. Director Yi Zeng (曾毅) is a professor at the Institute of Automation, Chinese Academy of Sciences, and an internationally recognized expert in AI ethics and safety. He served on the United Nations High-Level Advisory Body on Artificial Intelligence, the UNESCO Ad Hoc Expert Group on AI Ethics, the WHO Expert Group on Ethics and Governance of AI for Health, and China's National Governance Committee for Next Generation AI. He was named to TIME's 100 Most Influential People in AI in 2023. Deputy Director Wei Kai heads the MIIT-backed China Academy of Information and Communications Technology (CAICT) AI Research Institute. The institute conducts cutting-edge fundamental research, develops key technologies, formulates safety standards, and builds platform tools for AI safety and governance. It frames its mission with the analogy that safety provides "guardrails and brakes" for AI while governance acts as "the steering wheel" to guide beneficial development. Core research areas include AI model safety, LLM jailbreak defenses, AI ethics evaluation, and risk assessment frameworks. Key research outputs include PandaGuard, a modular framework for systematically researching LLM jailbreak attacks and defenses (with an accompanying PandaBench benchmark evaluating over 50 LLMs across 19 attack methods and 12 defense mechanisms), and ForesightSafety-Bench, a comprehensive AI safety evaluation benchmark spanning three hierarchical levels, 22 pillars, and 94 granular risk dimensions. Research has been published at AAAI 2025 and ICLR 2025, and the institute's work on super-alignment was cited in the Singapore Consensus on Global AI Safety Research Priorities (SCAI 2025). The institute collaborates with leading Chinese academic institutions including the CAS Institute of Automation, China Academy of Information and Communications Technology, Peking University, Tsinghua University, and Beijing University of Posts and Telecommunications, as well as AI and security companies. Internationally, the institute has participated in the Paris AI Action Summit (February 2025) and SCAI 2025, and has collaborated with Cambridge University on AI safety research priorities.