Special Focus

With the rapid iteration of global artificial intelligence technology, the competition in technological research and development has never ceased. In February of this year, UN Secretary-General António Guterres stated at the AI Impact Summit that while AI drives progress, it may also exacerbate inequality, amplify biases, and foster harm. This profound awareness of systemic risks has led the international community to gradually form a consensus in the field of AI: the biggest challenge currently lies not in optimizing algorithm models or in the bottleneck of GPU (graphics processing unit) supply, but in the governance that lags behind the code. The pressing issue today is no longer the speed of AI development, but the scale of AI governance. The focus on what to regulate and how to regulate AI has become a common concern for human society.

An Inescapable Topic in Global Governance

The development of AI technology and industry has increasingly impacted the global economic structure, resource allocation, and public perception, making it an inescapable topic in global governance.

Since February of this year, with the publication of research reports such as “The 2028 Global Intelligence Crisis,” the profound impact of AI on the global economic structure has increasingly alarmed various parties. For example, in the United States, many tech companies have laid off employees not due to poor business but because business is too good. The unexpected surge in orders and profits has led to employees being laid off in droves.

This has caused the macroeconomic structure to fall into a rapidly imbalanced quagmire: on one hand, productivity has surged, leading to exponential growth in product supply; on the other hand, a shrinking labor market has resulted in insufficient effective demand. This mismatch of “excess supply of goods and depleted purchasing power” has created a new economic dilemma. In the financial market, many American tech companies have received high valuations due to productivity improvements, but weak consumer demand has made profit expectations difficult to achieve. The intense tug-of-war between these two forces could ultimately trigger economic and financial turmoil.

Like previous technological revolutions, the development of AI also relies on many resources from the real world. The seventh United Nations Environment Assembly, held in Nairobi, Kenya, in December 2025, issued a stern warning about the environmental justice issues surrounding AI: developed countries are enjoying the efficiency dividends brought by AI, while the environmental costs—such as overloaded power grids, depleted water resources, and heavy metal pollution—are being paid for by developing countries.

The energy consumption of AI is not only reflected in electricity but also in water resource consumption. AI servers generate significant heat during computation, and the mainstream cooling methods primarily rely on water for evaporative cooling. Research shows that responding to 10 to 50 inquiries with an AI model can consume about 500 milliliters of water. Examining the future data center construction plans of major tech giants, there is a clear trend of relocating to global southern countries. This is not only due to lower local electricity and water resource costs but also because these countries have relatively weak voices in global environmental governance, resulting in less policy resistance to new high-energy and high-water-consuming data centers. In addition to the massive consumption of energy and water resources, the heavy metal pollution caused by computing hardware manufacturing also urgently needs to be addressed. The leap in computing power relies on the investment of massive GPUs, and the rare metals and key minerals required to manufacture these chips are mostly mined from developing countries.

Not only is the “ecological cost transfer” in terms of natural resource consumption and environmental destruction intensifying, but the global investment distribution is also deteriorating sharply. A report from the UN Conference on Trade and Development on May 6 pointed out that AI is reshaping the global investment landscape, with capital pouring into a few technologically strong countries, while other developing economies face increasing risks of marginalization. This one-way siphoning in the investment field, combined with ever-rising barriers to computing power, not only leads to severe talent loss in developing countries but may also cause them to become “abandoned” in future economic competition. This technological gap will exponentially amplify the inequality in global resource distribution.

The erosion of public perception by AI cannot be overlooked. The World Economic Forum’s “2026 Global Risks Report” released in January ranked “misinformation and false information” and “social polarization” as the second and third short-term risks. With the widespread application of AI technology, the cost of generating and disseminating false information has significantly decreased, severely undermining public trust in scientific facts and media, and exacerbating social polarization.

In recent years, the focus of social governance has remained on the “information cocoon” formed by algorithmic recommendations. Now, the ability of AI to generate massive amounts of semi-true, emotionally charged, and highly misleading content at low cost and in a short time has created an increasingly dense cognitive “fog” on the internet. The digital crisis facing humanity has escalated from the past “information cocoon” to “cognitive ecological pollution.”

Research by misinformation monitoring organization NewsGuard shows that the proportion of false information spread by mainstream large language model-powered chatbots in response to controversial news surged from 18% in August 2024 to 35% in August 2025, nearly doubling within a year. The reasons are twofold: on one hand, the AI’s previous “refusal to answer” mechanism when faced with sensitive issues has been significantly weakened due to commercial competition and the drive to “enhance user experience,” with some AIs’ refusal rates dropping to nearly 0 in 2025; on the other hand, these models heavily rely on real-time web searches, which are filled with false news generated by other AIs. AI lacks sufficient cross-validation and identification capabilities when scraping this content, which is then packaged as authoritative answers for secondary dissemination, effectively turning AI into a “transit station” and “amplifier” for false information.

Building a Global Collaborative Governance System for AI under the UN Framework

In the face of this inescapable topic in global governance, countries must adhere to cooperation and mutual benefit, relying on the UN mechanism to establish a multilateral collaborative governance framework that is universally representative and has strong binding force. This multilateral collaborative governance framework must accurately address the issues of “timing of governance” and “intensity of governance.” Premature or overly strict regulation can stifle industry innovation and hinder industrial development, while delayed or overly lenient regulation can amplify systemic risks and increase subsequent governance costs. This requires us to seek a scientific dynamic balance between “over-intervention suppressing technological vitality” and “regulatory lag inducing systemic risks.” To translate this “dynamic balance” from a macro concept into practical governance effectiveness, we must seek breakthroughs in the following three dimensions.

First is institutional transformation and distribution reconstruction. The productivity revolution brought by AI requires that the focus of human social governance must shift from “merely pursuing production” to “optimizing distribution mechanisms.” In response to the excess profits obtained by companies through AI replacing human labor, governments can explore forms like the “digital dividend tax” to return the dividends of technological development to society. In this regard, some countries have already begun discussions. On May 12, the head of the policy office of the South Korean presidential office, Kim Yong-bum, publicly stated that consideration should be given to using the excess tax revenue generated by the AI industry to establish a “citizen dividend,” returning it to all citizens through institutional arrangements. This approach not only secures people’s livelihoods but also reconnects a portion of wealth creation that is detached from labor with the purchasing power of human economic society, thus resolving the contradiction between excess productivity and shrinking consumption, supporting the minimum consumption power of society, and maintaining the basic circulation of the macroeconomic system.

Second is global resource and environmental auditing. With the exponential growth in the demand for large model training and inference, the AI industry is bringing unprecedented energy consumption and ecological burdens. Therefore, a globally unified and mandatory AI resource environmental auditing mechanism must be established, incorporating the power utilization efficiency, water resource consumption, and heavy metal pollution brought by hardware iteration of data centers into the compliance regulatory and disclosure indicators of large tech companies, thereby promoting the visibility of hidden environmental costs and the green allocation of computing resources. At the same time, drawing on the mature experience of carbon emission trading markets, we should explore the establishment of a “global computing power and ecological quota trading mechanism.” Funds raised through this mechanism can respond to Guterres’ proposal at the AI Impact Summit in February to establish an “ecological and digital compensation fund” to specifically assist global southern countries in building local green computing infrastructure, ensuring they are not marginalized in the AI era and achieving dual fairness in technological dividends and ecological responsibilities.

Finally, cognitive ecology and ethical regulation. In the face of the cognitive “fog” brought by generative AI, humanity urgently needs to establish a set of “environmental standards” for the digital age to address the increasingly serious “cognitive pollution.” On one hand, all AI service providers aimed at the public must add immutable invisible watermarks and explicit labels to the generated texts, images, and videos to effectively safeguard the public’s right to know, while also strengthening ethical regulations for tech giants, requiring them to conduct regular algorithm safety assessments to prevent AI from being used to amplify biases or incite extreme emotions; on the other hand, countries may consider establishing a transnational joint monitoring and rapid response network for AI misinformation under the UN framework to provide early warnings and traceability against behaviors that use AI to manufacture and disseminate false information across borders. By formulating globally unified AI ethical red lines, we can ensure that technological development always adheres to human ethical norms and value orientations.

China has always advocated for the inclusive, equitable, and benevolent development of AI, achieving win-win cooperation. In 2025, China clearly pointed out the existing gaps in the current international mechanism regarding governance in new domains such as AI. In 2025, at the World Artificial Intelligence Conference and the High-Level Meeting on Global AI Governance, China released the “Global Governance Action Plan for Artificial Intelligence,” providing a feasible “Chinese solution” for breaking down technological barriers and constructing a fair global digital order.

The future positioning of AI must and will be a new engine for enhancing the well-being of all humanity. In the face of this profound wave of technology reshaping human civilization, only by abandoning the old mindset of zero-sum games and strengthening collaboration under the UN framework can we ensure that the new round of technological dividends genuinely benefits all humanity.