On September 19, 2026, the autumn conference co-hosted by the Shenzhen Xiangmihu Institute of International Financial Technology and the Digital Finance Cooperation Forum took place in Futian District, Shenzhen. The event centered on macro-economic conditions and artificial intelligence governance, addressing pressing topics in economic and financial development. Through in-depth discussions, the conference aimed to provide forward-thinking insights and recommendations to support high-quality economic growth and promote beneficial AI applications.

The morning session featured a closed-door seminar analyzing macro-economic trends for the first three quarters of 2026. Experts engaged in comprehensive discussions on new patterns and characteristics emerging in China's economy, examined current difficulties and challenges, and proposed policy measures to maintain economic stability and achieve this year's social and economic development targets.

During the afternoon symposium titled "AI for Good: Governance Challenges and Pathways," distinguished leaders and scholars delivered keynote addresses. These included Wang Jiangping, a national political advisory committee member and former vice minister of industry and information technology; Li Meng, a committee member and former vice minister of science and technology; Zhou Hanmin, a standing committee member and president of the Shanghai Public Diplomacy Association; Luo Zhiquan, a foreign academician of the Chinese Academy of Engineering and vice president of the Chinese University of Hong Kong, Shenzhen; and Xue Lan, a counselor of the State Council and dean of Schwarzman College at Tsinghua University. Expert presentations were also given by Wei Liang from the China Academy of Information and Communications Technology, Shi Jianzhong from China University of Political Science and Law, Xiao Jing from Ping An Group, Wei Jiangbo from China Merchants Bank, Xu Qichang from Digital China Information Service Group, and Zhang Xiaozhi from Alibaba Cloud Intelligence Group. Liu Xiaochun, a member of the academic committee and former president of China Zheshang Bank, moderated the sessions.

Mr. Wang Jiangping delivered a presentation titled "Superior AI: A Grand Journey." He highlighted that current AI ethical risks are diverse, including inherent, application, and systemic risks. As technology evolves and industrial applications expand, these risks transmit across content ecosystems and physical security domains, gradually creating macro-level impacts on economic and social stability. Some frontier models already exhibit significant national security externalities. He noted that AI possesses both constructive and detrimental aspects, and governing AI resembles water management, requiring principles such as altruism, inclusiveness, accessibility, trustworthiness, fairness, justice, and agility. He emphasized that AI differs fundamentally from previous technologies, representing a form of intelligence distinct from humans, and humanity has never before coexisted with two types of intelligent agents. Furthermore, AI is quietly reshaping human civilization's trajectory, and future scenarios require collective exploration. He highlighted four major governance challenges globally: ideological polarization, short-termism in practice, cultural monoculture, and bloc-oriented rulemaking. His book, "Superior AI: Governance with a Human Touch," stems from long-term research and is structured in three parts: the first poses ten questions about AI's personal impacts; the second discusses technical logic in governance, introducing a three-level cultural alignment concept; the third addresses sovereignty issues involving chips, algorithms, and data. He also discussed AI's potential to commodify human emotions, human-machine alignment in the intelligent age, a four-pillar governance framework combining self-discipline, social oversight, government regulation, and judicial supervision, and the need for balanced technological and ethical development. He argued that technology may worsen inequality, and mechanisms ensuring technological accessibility, educational equity, and shared benefits are essential for AI to enrich society equitably. He advocated choosing simple, necessary, and appropriate technological capabilities, returning to clear authentic needs amid technical accumulation.

Mr. Li Meng shared three observations on governing AI amid rapid development. First, governance frameworks and tools should be designed from the perspective of humanity's shared values. China's AI development policy emphasizes independent innovation, application orientation, ecosystem collaboration, open cooperation, and secure controllability, aiming to ensure development toward safety, benefit, fairness, health, orderliness, green practices, sharing, and openness. Second, governance practices should adopt a classification-based approach. AI models and agents carry inherent, embedded risks at different levels: models form the foundation, and governance addresses fundamental quality issues to prevent defects; agents operationalize models in specific contexts, and governance ensures proper conduct and beneficial actions. Third, amid significant uncertainty, sufficient sandbox space should be preserved to maintain governance agility without over-generalizing risks. For vertical applications and derivative risks, a problem-oriented, scenario-based approach is needed, addressing issues as they emerge without overregulation that hampers development. He raised the question of what costs a developing nation is willing to bear for a technological and industrial revolution, calling for reverence in safeguarding AI for good, enabling AI to serve individuals and society within legal, compliant, and cost-effective parameters.

Mr. Zhou Hanmin discussed "Pathway Choices in US-China-EU AI Legal Governance." He noted that AI competition among the three economies encompasses not only technology and industry but also law and rules. AI governance involves accountability, copyright, training data rights, consumer protection, and intersects with civil law, intellectual property law, data law, and national security law. Law must not only regulate technology but also shape new social order. He outlined distinct approaches: China balances development and security through gradual governance, converting legal requirements into technical standards and coordinating legal and technical governance; the United States prioritizes innovation and national competitiveness, reducing pre-constraints in commercial innovation while imposing stringent governance in national security; Europe adopts principles of fundamental rights and risk-based classification, shifting from technology regulation to risk regulation where higher risk entails stricter obligations. To advance China's transition from studying rules to participating in rule-making, he recommended six areas: systemic legislation beyond fragmented rules; full lifecycle regulation beyond model oversight; value chain responsibility beyond single-entity accountability; balancing risk governance with innovation incentives; international compliance alongside domestic compliance; and evolving from rule-taker to rule-shaper. He proposed AI law should encompass three levels: guardrails preventing legal violations, foundations providing stable and transparent institutional environments, and rules for active participation in global AI rule formation.

Mr. Luo Zhiquan shared insights on complex systems intelligence and social value. He observed that AI is evolving from cognitive assistants and task agents to complex systems intelligence, with long-term value in helping humanity understand complex worlds, improve system operations, and enhance social welfare. The Shenzhen Hetao College focuses on understanding and optimizing complex systems, connecting frontier research, talent cultivation, and industrial practice through real-world problems. Its work spans scientific intelligence, engineering intelligence, and social intelligence, sharing model training and verification capabilities. Starting from communication network applications, it is expanding toward social intelligence and societal simulation. He believes social intelligence research should examine individual behavior, group interactions, and system evolution, developing models through three approaches: human behavior models simulating individuals, multi-agent social world models simulating societies, and mathematical modeling with decision models supporting decision-making. The college plans to build AI public infrastructure for complex societies, including economic and social world models and domestic computing platforms, to simulate individual interactions and assess policy impacts. He stressed that AI for good should begin from human needs, verify technology through real experience, and iteratively refine to drive social progress and quality of life improvements.

Mr. Xue Lan presented on "AI for Good: From Technological Innovation to Socio-Technical System Integration." He argued that AI development is transitioning from technological innovation to integration with socio-technical systems, which goes beyond production tool upgrades to systematically challenge modern organizational structures. Despite breakthrough progress, gaps remain between technological advancement and societal value realization, with many industry AI applications falling short of expectations. He proposed that technological progress and social construction are mutually reinforcing processes, and AI's second half involves constructing infrastructure, institutions, organizations, and culture. This includes hard infrastructure such as computing, digital, model, and security systems; soft institutional systems covering data property rights, circulation mechanisms, responsibility frameworks, AI auditing, certification, and regulatory sandboxes; organizational restructuring from flattened structures to AI-native designs; and social-cultural systems where AI for good serves as core guidance, encompassing trust mechanisms, risk perception structures, and capability frameworks. He argued that AI competition extends beyond simple technological innovation to capacity building within socio-technical systems, with sustainable competitiveness belonging to entities that simultaneously advance technological innovation and social system restructuring while embracing AI for good principles.

Mr. Wei Liang shared perspectives on AI risk dynamics and industrial governance. He noted that AI technology is accelerating rapidly, bringing both benefits and risks, with deepening applications creating inherent and derivative security challenges. These include risk chains extending from data, models, and agents; incomplete risk management systems; cascading risks from individual systems to organizations and society; financial sector risks such as model homogeneity, decision convergence, and risk resonance; and extreme risks from frontier model jailbreaks. He observed that globally, governance paths are diversifying by regional context: Europe adopts legislative, risk-based regulation; the United States pursues voluntary commitments for frontier models; China emphasizes people-centered, AI-for-good governance with agile, iterative, and multi-stakeholder approaches. The China Academy of Information and Communications Technology is actively exploring industrial AI safety governance through practical actions: building safety governance systems for industry; promoting AI safety commitments for industrial self-discipline; developing industry-oriented AI risk governance frameworks; continuously tracking technical risks to determine AI safety baselines; and exploring governance pathways for vertical sectors such as finance.

Mr. Shi Jianzhong delivered a themed presentation on AI development and security amid US-China competition and global governance. Observing major AI events in both countries, he offered three judgments: AI capability is not merely commercial but represents national competitiveness; balancing development and security cannot remain rhetoric, as security capacity derives from development; and China's AI legislation is accelerating, which will trigger adjustments across numerous related regulations. He predicted a "competition-primary, limited-coordination" pattern in US-China AI relations. To strengthen China's voice and initiative in global AI governance, he recommended producing thematic white papers to enhance governance explainability, converting domestic governance practices into international rule texts, using compliance predictability to counter extraterritorial jurisdiction, providing governance toolkits to the Global South, transforming opposition to over-generalized national security into operational rule frameworks, incorporating civilizational diversity into large model development and evaluation requirements, and leveraging UN platform draft clauses to promote multilateral negotiations.

Mr. Xiao Jing presented Ping An Group's AI governance practices under the theme "Acting for Good: Governance Challenges and Responses in the AI Era." He noted that AI has entered a new phase characterized by self-planning, self-execution, self-reflection, self-verification, and lifelong learning, forming a complete loop of perception, learning, memory, thinking, action, and reflection. Facing rapid AI advancement, he suggested enterprises focus on strengthening technical foundations, expanding application scenarios, and deepening AI ethics governance to achieve sustainable AI development value encompassing human capital, token capital, and ethical governance. To address core challenges at data, algorithm, and large model levels and comply with latest regulatory requirements, Ping An has built a "1+5+3" next-generation trustworthy AI governance system, implementing practices around data security compliance, model risk prevention, and intelligent agent management, reinforcing full lifecycle safety controls and ensuring legal compliance.

Mr. Wei Jiangbo shared China Merchants Bank's AI governance practices and reflections. He pointed out that national top-level design increasingly emphasizes matching AI governance capabilities with AI development and application capabilities. China Merchants Bank is proactively establishing enterprise-level AI governance mechanisms: strategic and sustainable development committees provide top-level decision-making oversight, coordinated by risk and compliance committees and digital finance committees; institutional guidance converts AI governance requirements into executable systems and processes covering models, scenarios, ethics, and safety; and system tool support strengthens pre-implementation model and data safety assessments, optimizes real-time content and agent safety guardrails, and deepens post-implementation AI application evaluations, driving full-process development through outcomes. He acknowledged current risks including content, data, continuity, explainability, and human-machine collaboration challenges, emphasizing that deterministic governance is key to addressing AI uncertainty, promoting development through governance while refining governance through development.

Mr. Xu Qichang shared Digital China's practices under the theme "Financial AI Governance: One Paradigm Shift, Four Operational Boundaries." He observed that as intelligent agents enter financial operations, governance paradigms are shifting from model governance to agent-system governance. Aligning with regulatory requirements, the basic governance unit is expanding from model invocation to complete decision loops. He proposed four operational boundaries: authority boundaries, pushing agent permissions to operation, tool, and parameter levels; responsibility boundaries, clarifying accountability chains and assigning responsibility to individuals; audit boundaries, ensuring full lifecycle traceability, auditability, and accountability; and evaluation boundaries, optimizing capability assessment standards for financial agents. Digital China has achieved both theoretical research and engineering implementation in financial agent governance. He called for regulators and industry stakeholders to jointly establish financial AI governance standards, collectively advancing AI toward beneficial outcomes.

Mr. Zhang Xiaozhi presented Alibaba Cloud's practices under "AI for Good: Financial AI Governance as a Steering Wheel for Innovation." He noted that with large-scale financial AI adoption, governance has become global regulatory consensus, yet governance capacity building lags behind model deployment speed. Addressing four structural risks in financial AI applications and regulatory requirements, he described Alibaba Cloud's financial intelligent agent platform practices in four areas: constructing a four-layer decoupled architecture connecting models to business operations; building five pillars of financial security trust grounded in technical facts; establishing comprehensive security governance with four layers of defense-in-depth; and developing a three-layer collaborative governance framework covering technology, institutions, and infrastructure. For sustainable financial AI applications, he proposed six actions: completing AI asset inventory and risk classification; establishing full lifecycle model management systems; creating AI ethics and governance committees; building AI supply chain resilience assessment systems; cultivating responsible AI organizational cultures; and jointly developing industry standards and regulatory sandboxes.