Introduction

In July 2026, Moonshot AI released Kimi K3, an open-weight model that has become the world’s largest open-weight AI model to date. Kimi K3 has once again drawn global attention to the capabilities of China’s frontier AI models, while bringing the debate over open versus closed AI models into sharper focus. Previously, many in the industry believed that the future of advanced AI would belong almost entirely to closed models. However, in recent years, an increasing number of leading AI companies around the world have begun exploring open-weight approaches. What innovations and value do open models bring? To what extent can they be considered global public goods? What unique advantages does China’s open AI ecosystem possess? As models become increasingly powerful and develop agentic capabilities, what new security risks and governance challenges will emerge? And how can we establish more responsible evaluation mechanisms and governance frameworks? In this episode of SinoAI Insights, we invite two leading experts to examine these questions through the lens of Kimi K3. Together, they explore the driving forces behind the emerging shift toward open AI, examine the latest developments in AI safety evaluation, and discuss new governance frameworks for a future shaped by rapidly advancing frontier AI capabilities.

Host

·XIAO Qian, Deputy Director, Center for International Security and Strategy, Tsinghua University

Guests

·Tianhao CHEN, Tenured Associate Professor and Ph.D. Supervisor, School of Public Policy and Management and School of Law; Director of the Center for Law and Governance and Assistant Director of the Center for Science & Technology Development and Governance (THUSTDG), Tsinghua University

·Xiaojian (Luke) LI, Founder & CEO of Fangcun Leap; Ph.D. Candidate at the College of Artificial Intelligence, Tsinghua University

Shownotes

03:05 Starting from Kimi K3: Why Leading AI Companies Are Turning Toward Open Models?

05:17 The Value of Open Models: From Engineering Efficiency to Accelerated Innovation

06:45 Can Open Models Be Global Public Goods?

08:06  How China Became a Major Contributor to the Global Open-Source AI Ecosystem?

10:34 Governance Challenges Emerging from Increasingly Powerful Open Models

14:43 What 60 System Cards Reveal: Changing Trends in AI Safety Evaluation

19:45 The Value of Third-Party Evaluation and the Worrying Turn Toward Closed Benchmarks

22:17 Evaluating AI with AI: Automated Benchmark Repair and System Card Generation

24:41 Why Agentic Runtime Safety Is Becoming Increasingly Important?

26:56 Introducing Agent Safety Governance Framework 1.0

30:34 The Four Threat Layers and Seven Governance Pillars for AI Agents

34:47 How Abstract Governance Principles Can Be Translated into Actionable Implementation Mechanisms?

37:32 Global AI Safety Priorities for the Next Three Years

Team

Executive Producer:XIAO Qian

Producers:WANG Yexu,LIU Yuan

Podcast Editor:CAI Leyao

Content Assistant:ZHANG Shuoning

Graphics Designer:XU Ruijia