2025人工智能治理年度报告：迈向“可衡量”的AI治理

Stride Toward the Measurable AI Governance: 2025 Annual Overview &Action Recommendations Institute for AI International Governance Contents Review of AI Development in 2025......................................................................4 Technological Exploration: AI Advances into the Physical World, Multi-Path Explorationof AGI..................................................................................................................................4Infrastructure: Accelerated AI Infrastructure Development Faces Dual Bottlenecks ofEnergy Constraints and Computing Power Competition....................................................5Industrial Ecosystem: Open-Source Model Boom and Application Expansion AccelerateIndustrial Intelligence Transformation...............................................................................5 Review of AI Governance in 2025....................................................................6 National Level: Accelerated Global AI Governance Deployment......................................6Intergovernmental Level: Deepening Cooperation and Differentiation Coexist in GlobalGovernance.........................................................................................................................6Non-State Actors: Consensus Exploration and Industry Self-Regulation in AI SafetyGovernance.........................................................................................................................7 Summary and Recommendations....................................................................8 Current Status Summary: The Gap Between AI Development and Governance Widens...8Action Recommendations....................................................................................................9⚫Recommendation 1: Support Exploration of Adaptive Governance Frameworks.9⚫Recommendation 2: Explore and Anchor Collaborative "Anchors" WithinCompetitive Landscapes................................................................................................10⚫Recommendation 3: Strengthen the Governance Role and Capacity Building of Non-State Actors....................................................................................................................10⚫Recommendation 4: Toward Measurable AI Governance........................................10 目录 2025年人工智能发展回顾...........................................................................11 技术探索：AI迈向物理世界，多路线探索AGI...........................................................11基础设施：AI基础设施建设提速，能源与算力成为双瓶颈......................................12产业生态：开源模型繁荣与应用场景扩张，产业智能化转型提速...........................12 2025年人工智能治理回顾...........................................................................13 国家层面：人工智能治理布局全球提速.......................................................................13政府间层面：全球治理合作深化与分化并存...............................................................14非国家行为体：AI安全治理的共识探索与行业自律..................................................14 总结与建议...................................................................................................15 现状总结：人工智能发展与治理的实质差距拉大.......................................................15行动建议...........................................................................................................................17⚫建议一：支持适应性治理框架探索................................................................18⚫建议二：在竞争格局中探索并锚定合作“锚点”........................................18⚫建议三：强化非国家行为体的治理角色与能力建设....................................18⚫建议四：迈向可衡量的治理............................................................................18 Review of AI Development in2025 research.As scaling laws driven by textualdata reached diminishing marginal returns,the focus shifted from purely digital domainstoward understanding and interacting withphysicalenvironments."Spatialintelligence" and "world models" becamenew directions, attracting top AI scientistslike Fei-Fei Li and Yann LeCun. AI nolonger merely predicts text tokens but beginslearning to infer the next frame of an image,theforces acting on objects,and theevolution of three-dimensional space. Theexploration of spatial intelligence and worldmodels will jointly provide the technologicalfoundationfor embedding AI into thephysical world. Technological Exploration: AI Advancesinto the Physical World, Multi-PathExploration of AGI In 2025, AI advances beyond the digital realmintothe physical world.Scientists haveacceleratedexploration of Artificial GeneralIntelligence (AGI) in this regard via multipletechnological pathways which further include AIagent, world models, spatialintelligence, as wellas embodied intelligence. Moreover, the infusingof AI and the physical world has also profoundlytransformed the nature and positioning of human-machine interaction. ⚫In 2025, models evolve from "cognitiveintelligence" to "embodied intelligence."In cutting-edge fields like humanoid roboticsandautonomousdriving,embodiedintelligence has moved beyond laboratory