Introduction: How the future is delivered to everyone
On August 10, 2026, Meta founder and CEO Mark Zuckerberg published an article titled (The Future is for Everyone) (“The Future is for Everyone”), with the subtitle “The Path to a Positive AI Future.” Just a day earlier, superintelligence still mostly existed in the realm of technology-company predictions, lab reports, and public imagination—but Zuckerberg’s piece brought it into the center of politics and everyday life. He expected that over the coming years, people would use “superintelligence beyond human capabilities” to create new things, discover new knowledge, build new businesses, and improve health and quality of life. The article repeatedly emphasized that superintelligence should be “delivered to everyone,” and that everyone should have their own intelligent agent, creative tools, mentors, and research assistants. Zuckerberg also argued that AI safety should not rely on some single superintelligence to make decisions for humankind, but instead on checks and balances among individuals, businesses, governments, and multiple AI systems.
This may look like a wonderful vision. But besides the outlook above, the article has another, equally clear line. Zuckerberg says the United States and democratic countries must maintain leadership in artificial intelligence. The front-line labs should cooperate with the government by providing the government with intermediate training checkpoints and technological resources for advanced models. The United States should continue building chips, energy, and data centers, using export controls to delay competitors’ development while also letting the United States lead the global open-source AI ecosystem. Here arises a question worth probing: how can superintelligence belong to “everyone” while still being led and controlled by the United States and its allies? I think we cannot simply dismiss this as a contradiction in Zuckerberg. We should ask further: in the phrase “everyone,” who exactly are they—AI users, beneficiaries, or political subjects who have the power to jointly decide AI’s direction? From using superintelligence to owning the future, how far is that gap, really?
I. Who are “everyone”?
“Everyone” looks like the most inclusive word, but in actual use it is often the most ambiguous. When a person encounters AI, they can at least occupy four different positions: they can be a user; they can benefit from AI’s development; they can participate in governing AI; and they can participate in deciding the direction of AI. The first two positions involve access and consumption; the latter two involve power and sovereignty. Zuckerberg’s article is very clear about the first two promises: everyone will have a personal agent, able to use creative tools, private tutors, and scientific assistants; the tools should be free or priced low enough to meet the needs of billions of people. As for whether ordinary people can decide how the model is trained, which data can be used, which values should be written into the system, which groups should receive priority care—his article does not give clear answers. A person can use AI but may not own AI; a person can benefit from intelligence but may not be able to change the conditions under which intelligence operates.
Personal AI is exciting because it creates a strong sense of agency. An ordinary person can make AI act as a teacher, translator, programmer, business consultant, and even become a life assistant that stays with them for the long term. Zuckerberg’s examples are very concrete: a personal agent can help users monitor sleep, prepare recipes for the household, and can also assist a person in conceiving products and developing applications. Such technology does indeed expand the range of what individuals can do. The problem, however, is that capability and sovereignty are not the same. Users can command agents to complete a task, but cannot decide what the agent learns, what it refuses, or how it understands what is “beneficial to the user.” Users can pose questions, but may not be able to decide which questions are worth asking; they can get answers, but may not be able to further question the institutional and interest relationships behind the answers. Thus we see that individuals with super AI form a new identity: on the one hand, people are granted stronger action capacity; on the other, people are still merely users within the system.
In addition, understanding superintelligence as a product that can be produced, priced, and delivered will also obscure its institutional attributes. Once AI enters education, healthcare, labor, media, the judiciary, and the military, what it affects is not only personal efficiency—it also helps arrange the ways knowledge circulates, redraws the boundaries between professional and non-professional, influences who can find jobs and who finds it easier to get loans. And once the issue is framed as “whether everyone can afford it,” many power relationships slip out of sight. Even if everyone can use a model for free, the training data behind the model, chips, cloud computing, interfaces, and security standards still remain concentrated in the hands of a few companies. It is like how cars may be widespread, but roads are still built and managed by a small number of institutions; the internet is used worldwide, yet undersea cables, servers, and communication standards still have clear controllers.
Therefore, when Zuckerberg says “AI belongs to everyone,” it is actually closer to bringing everyone into an intelligent order built by a small number of institutions. This judgment does not erase the real benefits of AI adoption. Low-cost educational tools may help students in impoverished areas, and personal agents may reduce the barriers ordinary people face when accessing knowledge. But real-world benefits cannot make us ignore the fact that expanding access does not automatically expand decision-making power. So the following questions deserve continued scrutiny: when everyone becomes a user, who can still set the rules of the system? When the future is delivered to the public as a product, who is responsible for producing the future? Who is responsible for explaining it? And who has the right to refuse this product? At that time, does the public still have the ability to imagine the future?
Second, when power oriented toward broad distribution becomes concentrated
If we only see the surface conflict between “everyone uses it” and “U.S. leadership,” it is easy to interpret Zuckerberg’s argument too simply. Superintelligence will indeed bring two categories of opposite risks. One risk comes from concentration—if the strongest models are held only by one company, one government, or a small group of experts, society may form a new class of technological aristocrats. The other risk comes from uncontrolled diffusion—high-intensity network attacks, biological risks, military applications, and large-scale fraud could all become easier as advanced models become widely available. Zuckerberg tries to find an arrangement between the two: distribute superintelligence across individuals, enterprises, and society, while the United States and its allies control key technologies and security initiative. This plan has some grounding in reality and also clear institutional consequences. This article, for now, focuses its critique on the latter.
According to Zuckerberg’s vision, superintelligence will form a vertically separated structure. The top layer appears highly distributed: individual users, developers, small businesses, government agencies, and multiple AI agents can all gain capabilities. But the bottom layer requires highly centralized resources: the chips, energy, and data centers that large-scale model training depends on cannot exist without continuous capital input. Following this logic, there is an entry point for individuals to get superintelligence, and once the platform controls the entry point, it controls the road behind it. A small business can use AI to do work that used to require large teams, but it still must rely on the platform’s interfaces, computing power, and update services. A country can deploy models, but it may be unable to independently train systems of comparable scale. As more and more people enter the same AI ecosystem, reliance on a small number of infrastructure providers may become even deeper. From the outside, intelligence seems to flow to all corners of society, but deep down it creates broader dependence.
“Openness” has the same two-sided nature. Open-source code or open models can lower research barriers, encourage developer participation, and reduce a single company’s technical lock-in—these values are worth acknowledging. But even if models can be downloaded, it does not mean training capability has been distributed equally. Even if a system is public, ordinary researchers may not have the computing power, data, talent, and funding needed to replicate it. Developers can build applications on top of a model, but they may not be able to change the model’s most basic classification system and value boundaries. Whoever owns the original architecture will control the largest user base; whoever decides the update rhythm will likely retain the strongest position in an open ecosystem. Open-source code may weaken a certain monopoly, or help a platform push its standards to the world, but the replicability of technology cannot directly be inferred as replicability of power.
The freedom of personal AI use also needs to be distinguished from the freedom to participate in rule-making. The former means people can use tools to generate content; the latter requires people to participate in deciding how AI should develop, which risks are acceptable, how workers should be compensated, and how platforms should be supervised—an array of issues more tightly tied to social development. Zuckerberg mainly discusses the former. He wants people to have personal superintelligence so they can live according to their own goals and create. But when users can only freely choose within the boundaries set by existing platforms, what they have is closer to consumer freedom. Genuine political freedom also includes the ability to object to the institution itself, modify rules, and even refuse certain technological arrangements. Zuckerberg’s desired freedom for users clearly does not include the power to shape an AI order.
This explains a seemingly contradictory phenomenon: AI is widely distributed, yet power can still be—and will further be—concentrated. Open layers of use expand the size of platform users; increased individual capability deepens people’s reliance on underlying services; and requirements for national security provide further justification for concentrating key technologies. This leads to distributed use and centralized dependence happening at the same time. What Zuckerberg proposes is an order that enables more people to act, but the power to control the conditions of action still lies with a small number of companies and governments. The more deeply AI enters society, the more opportunity infrastructure owners have to become intermediaries that society cannot do without.
Third, the boundaries of universalism
Since the power of infrastructure owners will keep increasing, it becomes ever more important to determine who has the right to own AI infrastructure, define AI infrastructure, and regulate AI infrastructure. Zuckerberg clearly understands this. In the text, he repeatedly places leadership in the United States alongside leadership in democratic countries. His basic point is straightforward: in the future, the direction of advanced AI development will affect the international order—whoever controls the leading systems may shape how freedom, prosperity, and security are distributed. Therefore, the United States and its allies need to maintain their lead: build more energy and data centers, continue developing the chip industry, and slow down competitors by means of export controls. Here, the concept of “democratic countries” needs to be clarified first. It can refer to an institutional type, or it can refer to a geopolitical camp. If it refers to an institution, it should include citizen participation, government accountability, transparency of power, and protection for minorities; if it refers to a camp, it is closer to “countries standing with the United States.” The two meanings are often placed in the same sentence, but their political effects are entirely different.
If “democratic leadership” really is meant in an institutional sense, then the problems that follow would include: can ordinary U.S. citizens participate in the national AI strategy? Do workers affected by automation have rights to negotiate and to refuse? Can people evaluated by algorithms understand the standards, file appeals, and receive compensation? Have large technology companies accepted sufficiently strong public oversight? Can a platform’s influence over billions of users be constrained simply through cooperation between corporate boards and government agencies? When these questions are not answered adequately, “democratic countries” can easily turn from an institutional principle into a label of civilization for “our side.” Democracy then mainly serves a function of camp identification, rather than a制度 arrangement for constraining power.
This brings about a familiar kind of universalism that is strongly misleading. America’s technological advantage is described as the advantage of the free world; America’s national security is described as security for all humanity; export controls are described as preventing dangerous diffusion; and when Meta spreads models globally, it is described as enabling democratic values to have broader impact... Such narratives cannot simply be dismissed as lies, because security risks do exist and interests conflicts between countries do exist. But it is important to note: the more a country can frame its strategic interests as the public interest of all humanity, the less visible its power becomes. Its rhetoric shifts from “I need to occupy the center” to “only if I stay ahead will the world be freer.”
In history, empires often used similar language. Emperors rarely said they only intended to seize resources; they also claimed they had brought order, civilization, and progress. Today, technological power does not need to expand through direct rule; it can also organize the world through standards, platforms, chips, and security rules. The most powerful forms of domination may not appear as commands, but rather as a condition: you can choose freely, but all your choices depend on the same underlying infrastructure. Technology companies and states form a complex relationship here: companies need governments to provide security and geopolitical protection, while governments need companies to maintain technological leadership. Their interests are not the same, yet they may temporarily converge in the language of “democratic leadership.”
The Global South is where these boundaries of universalism is easiest to see. Many countries can use models provided by American companies to obtain services such as education and healthcare, and they can also use open models to develop local applications. These benefits are real, but the problems are equally real: can they have computing sovereignty, data sovereignty, and rule-making power? Just because their language is included in training data does not mean they can decide how the model understands their history. Just because their users are integrated into the global market does not mean they can share the value generated by the model. Just because they receive AI products does not mean they control the direction of AI development. If rights to use, rights to produce, rights to interpret, and rights to govern remain separated for the long term, the spread of technology may become a new form of technological dependency.
The impact of AI will go even deeper into the language layer. By using language and data to organize the world, the model does not output only information; it also outputs ways of classifying problems, scales for judgment, and imagined notions of “reasonable living.” It will influence which problems are worth raising, what expressions count as professional, what lifestyles count as progress, and what political proposals count as responsible. If the most powerful models are mainly trained according to knowledge, legal systems, and cultural rules from the English-speaking world, then what global users receive may be a system that can speak in local languages, yet still uses other people’s conceptual frameworks to explain reality. Languages can be translated, but worldviews will not circulate on equal terms.
Fourth, personal superintelligence and future sovereignty
The most compelling part of the idea that “everyone has superintelligence” is that it seems to rearrange the power relationship between individuals and large organizations. Ordinary people can have their own teachers, lawyers, researchers, doctors, and so on; individuals can access knowledge faster, propose ideas, and start small businesses. Zuckerberg believes that as individual abilities grow, society will generate more inventions, more new businesses, and new jobs. He also argues that the main value of superintelligence should lie in invention and discovery, with automation only being a part of it. There is an optimistic side to this assessment: technology does create new industries, and in history, many careers disappeared after older technologies, only for new jobs to emerge afterward. But the problem remains: who owns the new productivity? Who bears the costs of transition? Who gets to wait for “future new jobs” to appear? Private AI may help a person learn, but it cannot, on its own, solve problems of labor systems and wealth distribution.
Getting a private assistant does not mean workers gain the collective ability to bargain. A person can have AI find jobs for them or edit their résumé, but that does not let them decide how a company should deploy automation, nor does it let them decide how the gains from increased productivity are distributed. Inequality in education may be reframed as whether one has access to better private tutors; defects in healthcare may be converted into whether one has access to better health agents; and structural unemployment may be translated into whether a person can quickly train new skills. As a result, social conflicts are broken down into countless private optimization projects. Everyone becomes better at coping with change, but few people jointly ask: why must everyone keep adapting to changes decided by others? AI can increase an individual’s capacity to adapt to an order; it may also reduce people’s passion to change the order.
Personal AI will also become a new intermediary between individuals and the public world. It filters news on a person’s behalf, summarizes viewpoints, explains policies, and even participates in managing intimate relationships. On the surface, it looks like everyone has a highly personalized agent, but in reality these agents may share similar base models, data sources, and judgment criteria. People get different tones and interfaces, yet at the bottom they may be facing the same kind of worldview. The system does not need everyone to do the same thing; it only needs to, according to each person’s habits, guide them toward similar judgments in different ways. What is worth warning against is that the danger brought by AI personalization is not that it makes everyone similar, but that it may turn subjectivity into a private service. People’s uniqueness is seen, but only within the range that platforms can identify and compute.
Zuckerberg proposes that multiple people, multiple companies, multiple governments, and multiple AI systems can check and balance each other, preventing any single superintelligence from holding all power. This idea borrows the checks-and-balances concept from liberal political thought and responds to public concerns about technology monopolies. But the problem is that having multiple actors with power does not mean power is already distributed equally. As discussed above, if they all depend on a small number of chip companies, cloud platforms, energy facilities, and model standards, then the appearance of pluralism is built on deeper shared dependence. The new center does not have to be the only model; it only needs to become the infrastructure that all models must pass through to function as a center. People may freely choose different applications, but they cannot bypass the same computing power and cannot escape the same interfaces and standards. As a result, the core of power shifts from “who commands whom” to “who sets the conditions for all actions.”
The phrase “the future belongs to everyone” itself is also telling. It portrays the future as something that can be produced, packaged, and delivered, positioning everyone as recipients and positioning companies and states as deliverers. But the future does not wait for some company to send it to us. The future is first of all a political question: who defines what counts as progress? Who bears irreversible risks? Who can refuse a certain technology? Who has the right to demand that people misjudged by models receive explanations? Who can propose another way of living? If these issues are not brought into the frame, then “providing a future for everyone” may end up meaning that a small number of people determine the future first, and then deliver it to the public as a product. At that point, is the future the public receives still truly theirs? Do they still have the ability to imagine the future?
Those who have leading models can influence what problems are worth solving, what knowledge is worth spreading, and what kind of life should be pursued. Models do not need to directly issue political orders; as long as they continuously help people choose, rank, and interpret, they will gradually participate in shaping the direction of society. Once, in the future, they are written into product roadmaps, infrastructure plans, and security policies, ordinary people will not be facing a mere tool. They will be facing a world that has already arranged the possibilities for them.
Conclusion: Who has the right to name everyone’s future
Zuckerberg’s vision still contains parts worth taking seriously. Giving more people access to AI lowers barriers to knowledge and creativity, and preventing superintelligence from being monopolized by a single company or government is indeed an important goal of technological governance. The problem is that the spread of technology does not automatically bring political equality; open models do not automatically lead to shared sovereignty; and the growth of individual capability does not automatically form public freedom. Zuckerberg turns sovereignty into a distribution issue, turns political questions into product questions, and turns the question of a shared future into a matter of individual efficiency. Such a narrative is easy to accept because it converts grand fears into hope that can be purchased and acted upon. But we must see what Zuckerberg is trying to do: organize broader reliance through widespread use, preserve technological leadership through open forms, and cover infrastructure power through the growth of personal ability. The more people use the same AI ecosystem, the closer platforms get to the underlying layer of social life. And once platforms are closer to that layer, America’s technological advantage is easier to frame as the shared interest of the democratic world. AI may indeed give more people access to intelligence, but it does not explain how more people gain the power that determines the direction of intelligence.
Real democracy requires people to jointly decide the conditions of their own lives. Giving everyone an AI can only accomplish a small part of that. More important things have not yet been solved: who sets the rules for training models, who controls computing resources, who bears technological risks, who distributes the gains from automation, and who has the right to refuse a future that is called “progress.” We care whether AI can belong to everyone, and we should continue to ask: who has the right to name a certain future as “everyone’s future”? How far is it from using superintelligence to owning the future?
As long as this question is still answered by a small number of companies, a small number of states, and a small number of technology alliances, “the future belongs to everyone” will remain an inspiring promise. It tells us that the future is coming, but it does not explain what exactly is moving toward us.