Access to Generative AI: Privilege or Right?
Access to a nascent technology often begins as a privilege and changes to a right once lack of access starts preventing meaningful societal participation. This article examines whether generative artificial intelligence may be following this path.
Most technologies viewed as critical necessities today began as luxuries enjoyed by a wealthy and privileged elite. In some cases, technology evolved over time to become widespread and indispensable to modern life. A technology’s progression from privilege to right often involves an evolution in the way in which the technology is perceived, regulated, and accessed. What begins as expensive, novel, and access-limited can later become cheap, ubiquitous, and indispensable.
This evolution may be evident not only in how consumers act but also in the services organizations and governments prioritize. Nonprofit organizations focused on serving the basic needs of the unhoused previously provided food and shelter and now, increasingly, provide access to cheap mobile phones. Several decades ago, internet access started as a U.S. Department of Defense research experiment with highly limited access. Now, the United Nations’ Sustainable Development Goals include pursuing universal, affordable internet access in the least developed countries. What were once niche technologies have become, for all intents and purposes, essential to full participation in modern society.
Use of generative artificial intelligence (AI) has seen a meteoric rise since the launch of ChatGPT in November 2022. In the early 2020s, access to the technology was largely gated behind application programming interfaces (APIs) and limited to wealthy corporations, select government agencies, and tech-forward early adopters. By October 2026, OpenAI alone reported more than 1.2 billion monthly ChatGPT users. Generative AI is being integrated into a wide range of sectors: financial, health care, education, and more. It is used for an astounding range of use cases, from developers coding, to doctors reviewing scans and farmers determining ideal planting conditions. In its general purposeness, there is something for everyone. In a 2024 survey by Microsoft and LinkedIn, 66 percent of business leaders said they would not hire someone who did not have AI skills. And generative AI is impressive not just in the speed of its proliferation, but in its stickiness. Twenty-four percent of American adults use generative AI tools daily, across many parts of their personal and professional lives.
Should this trend continue, generative AI may gain a permanent foothold in our economies and societies. Like technologies before it, generative AI may eventually become a necessity for full participation in the workforce and in civic life. That’s not to say that all jobs will require generative AI, or that a life without generative AI will be an inherently poor one. Rather, the technology may be critical to accessing the large majority of opportunities and institutions at some point in the not-too-distant future.
For this reason, exploring whether access to generative AI is a right or a privilege is not a theoretical or philosophical conversation, but one with practical implications for the billion-plus daily users of generative AI tools today and the many billions that will use them or be impacted by their use in the near future. If one accepts the premise that the public should take seriously the prospect that access to generative AI is on the path to becoming a right, we need to address critical considerations with regard to access. While these considerations may be on the horizon, the development and deployment of generative AI is taking place at a speed that has outpaced prior technologies. Society and the AI industry should grapple with these questions now, rather than waiting until necessity leaves no choice.
In this article, we follow the trajectory of three technologies—electricity, the telephone, and the internet—from discovery to widespread adoption, from privilege to right. Though unique in origin and velocity, each technology lands in a similar position.
We argue that:
- Some technologies evolve from expensive and elite tools of the rich (i.e., privilege) to critical necessity for full participation in society (i.e., right).
- Society should take seriously that access to generative AI is on a path to becoming a right, following a trajectory similar to past technologies, though at a quicker pace.
- The likely progression of generative AI from privilege to right has implications for how societies, governments, and developers should approach user policy and product deployment. We focus our analysis on implications for developers, arguing it is important for these developers to consider the degree to which generative AI is likely to evolve into a right, and design systems and policies with this likely end state in mind.
Situating Terms
For our purposes, we define access to AI as meaningful access to generative AI capability. Meaningful access requires that the AI system be available, affordable, sufficiently reliable, accessible to people in different languages, usable with reasonably accessible devices and connectivity, and not subject to arbitrary or discriminatory denial. Importantly, access to generative AI as we define it does not mean access to every model, frontier-level capabilities, unlimited compute, or uses that create substantial risks to others.
There are a number of ways in which a technology may be conceptualized as a right. It could be thought of as (a) a freestanding human right to access, (b) a derivative right, one whose access is necessary to exercise existing rights such as education, political participation, or access to information, or (c) a statutory social entitlement, comparable to universal-service programs such as education.
Freestanding human rights are moral entitlements afforded to all persons and are directed toward protecting fundamental dimensions of human dignity, agency, freedom, or well-being. The Universal Declaration of Human Rights (UDHR) Preamble establishes that these rights are “equal and inalienable rights of all members of the human family.” Such rights include the right to life and liberty, to legal recognition and remedy, and to nationality, property ownership, and marriage. A subset of human rights are negative rights, which protect people from certain encroachments by institutions or governments. The UDHR protects people from slavery, torture, and arbitrary detention. In the case of human rights, the UDHR establishes a positive duty of states to protect individuals accordingly.
In a 1984 essay, legal philosopher Joseph Raz detailed the notion of derivative rights, a right grounded in the existence of a more fundamental core right. For instance, the right to say any given sentence is derived from the broader right to express oneself. Derivative rights may exist by virtue of their necessity in accessing other core rights. If access to the internet is required to access health care, internet access may be thought of as a derivative right. The key distinction here is that a derivative right does not stand on its own—absent the core right it enables, it ceases to be a right at all.
We argue that AI may become necessary to exercise already-recognized rights, making it best conceived of as a derivative right.
Historical Analogies: From Privilege to Public Necessity
In this section, we map the progression of prior technologies from elite luxury to critical necessity.
Electricity
Electricity is the clearest example of a privilege turned right. Electricity was not born as a public utility. It began as a technology financed by private capital and available only to the wealthy. Only after society reorganized itself around electricity did governments treat broad access as a public necessity.
It took more than a century to progress from Benjamin Franklin demonstrating in 1752 that lightning is an electrical phenomenon with his kite experiment to Thomas Edison developing a commercially practical incandescent light bulb in 1879. Soon after, electricity arrived in the homes of wealthy and urban American elites in the 1880s. Urban electrification accelerated in the 1910s and 1920s, facilitated by physical proximity of urban and suburban dwellings. Indoor lighting allowed homes to dispense with gas- and kerosene-based lamps, which were both dangerous and emitted hazardous fumes. With improved lighting, work days no longer needed to be oriented entirely around sunlight hours, increasing economic activity. The economic and societal implications of widespread electrification were enormous.
However, as urban and suburban homes in the United States electrified at a relatively rapid pace, rural areas were largely left behind. In 1936, nearly 90 percent of farms lacked electric power because the cost of serving rural areas was prohibitive for private utilities. The newly created Rural Electrification Administration responded by financing cooperatives and infrastructure expansion, transforming electricity from a luxury good into a widely available amenity. This led President Franklin D. Roosevelt to say, at a celebration of the Rural Electrification Administration in 1938, “Electricity is a modern necessity of life.” Indeed, when one considers how electricity access has become necessary for health, education, economic production, communication, and daily life, access to it can be aptly thought of as a right.
In the decades since, electricity access has become a global priority. The UN’s Sustainable Development Goal 7 calls for “universal access to affordable, reliable and modern energy services” by 2030. The international development framework treats electrification as a universal objective because electricity is a precondition for health, education, economic production, and communication—put simply, a precondition for participation in modern life. Its absence, therefore, can be argued to impact human rights. Without refrigeration, vaccines and other medications may be unavailable, impacting the human right to physical health. In many scenarios, electricity is critical to the exercise of an individual’s right to work. In all of these circumstances, access to electricity enables the full pursuit of other rights.
This argument should not be construed as saying access to electricity is an absolute right, however. In the U.S., utility companies can shut down power to a home after repeated nonpayment, though many states have exemptions, which further substantiates conceiving of electricity as a right. In the U.S., if an individual moves to a rural location that is not yet connected to the electric grid, they typically bear the burden of some of the costs of extending the grid infrastructure required to serve them. However, the utility company cannot typically refuse their connection. It should be noted that, even in cases where access to electricity is inhibited (for example, when an electric power line no longer reaches a home), an inhabitant of that community can still get the benefit of electric power via devices, batteries, or local generators that do not require a connection to the grid.
Telephone
The first successful transmission of intelligible speech on a phone line occurred in 1876. However, early development was fractured, expensive, and heavily concentrated in commercial hubs and wealthy urban areas. When the foundational patents held by the Bell Telephone Company expired in 1894, a mass proliferation of independent, noninterconnected telephone networks emerged, leaving the market highly fragmented. AT&T offered an alternative through its “One Policy, One System, Universal Service,” though its success risked making phone access available through only one provider.
The term “universal service” did not initially imply an individual entitlement or a government subsidy; it meant that everyone in a given community should be connected to the exact same physical network so that they can communicate with one another. Because achieving this unified network required eliminating market competition, this meant leaving a critical public utility in private hands. To protect the public interest and prevent monopolistic abuse, Washington moved to actively legislate on this issue, formally making the slogan of “universal service” a legislative requirement.
Telephone access eventually became a prerequisite for modern life, altering the essential infrastructure needed for social and economic participation. By the late 20th century, lacking a telephone effectively locked an individual out of the job market, emergency services, and basic civic life. In the United States, the rights-like status of communications access is apparent with federal subsidies like the Lifeline program that explicitly subsidize telephone access for low-income households.
Internet
The internet’s origins can be traced to ARPANET, a Department of Defense-funded research network. The first host-to-host ARPANET message was sent from UCLA to the Stanford Research Institute on Oct. 29, 1969. A separate network, NSFNET, was launched by the National Science Foundation in 1986. It became the de facto U.S. internet backbone, connecting approximately 2,000 computers in 1986 and more than 2 million by 1993. As of November 2026, 76 percent of the world’s population, a whopping 6.3 billion people, have access to the internet.
Access to the internet is increasingly critical to access educational services, health care, work, banking, and even political participation. What began, for consumers, as slow and expensive dial-up connections to access text-based chatrooms and email, has become a ubiquitous technology necessary for daily life.
In countries and communities with high rates of internet connectivity, opting out of an internet-connected life may seem nearly impossible and certainly highly inconvenient. Given the ubiquity of availability of social services online, the educational necessity of online research, and the access to information allowed by the internet, the argument that access to the internet should be a human right has taken hold.
While state and federal U.S. governments have not provided statutory protection for internet access as they did for electricity and telephone access, this may have to do with infrastructural differences. Electricity and early telecommunications infrastructure were prohibitively expensive and led to localized monopolies, from whom policymakers needed to protect the public, whereas internet connectivity has generally been a competitive marketplace. That has largely panned out, at least in the U.S., with governments providing funding for rural broadband development and affordability programs for circumstances where market competition failed to ensure access.
The United Nations has not called for the addition of access to the internet itself as a fundamental human right; rather the Sustainable Development Goals frame internet access as a development priority: Sustainable Development Goal Target 9.c calls for significantly increasing access to information and communications technology and striving for universal and affordable internet access in least developed countries. The UN Human Rights Council did affirm that the same rights people have offline must also be protected online and has condemned intentional measures to prevent or disrupt internet access, further cementing the importance of internet access.
Electricity, the telephone, and the internet demonstrate the same broad arc: A technology starts as elite or specialized, distributes and democratizes, and eventually becomes indispensable. AI could be heading in that same direction.
Rise of Generative AI
The release of ChatGPT to the public in November 2022 was a major inflection point because it made generative AI broadly usable by ordinary consumers. Until then, state-of-the-art large language models (LLMs) were primarily available behind APIs. ChatGPT’s novelty was due not to a highly advanced underlying technology—it was based on an LLM that had been available for many months by then—but rather to its broad accessibility and optimization for the direct-to-consumer context. OpenAI described ChatGPT as a research preview trained with reinforcement learning from human feedback.
Within roughly two months of its launch, ChatGPT reportedly reached 100 million monthly active users, making it one of the fastest-growing consumer applications in internet history. That launch did for generative AI what earlier infrastructure expansions did for electricity and the internet: It shifted a powerful technology from expert and institutional settings into mass public use.
But generative AI differs from electricity and the internet in at least one crucial respect. With electricity, the central access problem was physical infrastructure: wires, generation, and rural distribution. With internet access, the central problem was networks: cables, towers, devices, and telecom markets. With generative AI, much of the basic infrastructure, in terms of internet and cloud computing, already exists. The harder question is access to capability. The most useful AI systems are proprietary services controlled by private firms, cloud providers, and a relatively small set of frontier labs. Stanford’s 2026 AI Index reports that industry produced over 90 percent of notable AI models in 2025, including for systems from OpenAI, Anthropic, and Google.
The question, then, is about having meaningful access to model capabilities that may be privately owned, safety-gated, regionally restricted, subscription-priced, and governed by terms of service. Some observers might interpret the availability of increasingly capable open-source AI models as effectively meeting this need. However, the mere availability of open weights or open-source models does not automatically equate to meaningful access for the average individual. While open source may shift control away from a small handful of frontier labs, it simultaneously shifts the burden of infrastructure and technical execution onto the user.
Compounding these physical barriers are operational challenges, technical friction, and low user know-how. Accessing proprietary models is as simple as typing into a web-based graphical user interface. Conversely, maintaining open-source alternatives demands, among other things, specialized technical literacy and a working knowledge of prompt engineering, terminal environments, and model quantization. Without these, open models remain functionally out of reach to the very populations that a decentralized framework would ostensibly empower.
Given this, the large majority of users access generative AI through proprietary models. The increasing necessity of access to these generative AI tools can be seen through the rise of their use over the past four years. They went from being a gimmick, to being helpful, though not always reliable, to conferring a competitive advantage. While discussion of use of AI in classrooms centered on academic integrity at first, AI literacy is becoming part of required coursework at some institutions. Generative AI is therefore well on its way to becoming necessary for meaningful societal, educational, and economic participation.
Speaking to the Skeptics
Our argument should not be interpreted as suggesting that generative AI will be critical to every occupation or that individuals in the near future will not be able to survive well or happily if they avoid generative AI in all its forms. Indeed, there are people today who live without electricity, telephone, and internet access, many of whom lead highly fulfilling lives. They are, however, a small and decreasing minority of individuals. International development organizations herald progress in closing these accessibility gaps, and living intentionally without these technologies—having access and choosing to forego it—likely requires a great amount of planning and inconvenience.
Some skeptics may argue that generative AI should not be compared to electricity, the telephone, or the internet—technologies that fundamentally altered the trajectory of human civilization. The transformer architecture on which today’s generative AI relies is less than a decade old, and generative AI entered the public consciousness only in the past four years. Comparing a nascent technology that may prove to be a blip on the radar of technological advancements to technologies that so fundamentally changed human lives and economies may be viewed by some as contributing to the AI hype for which Silicon Valley has received much scorn. The alternative, however, presents a riskier proposition. If society downplays the potential of this technology, it risks being unprepared to support broad access.
The historical analogies provided are admittedly not apples to apples comparisons to generative AI. For example, compared to generative AI, electricity is comparatively fungible: Electricity from one provider performs essentially the same function as electricity from another. AI systems, by contrast, differ substantially in quality, language coverage, modality, reliability, safety restrictions, and suitability for particular tasks. A right to AI access must therefore mean more than having an account or being able to connect to a model.
AI is also different because it can produce speech and recommendations, render judgments, and execute online actions. AI providers are not merely transmitting a neutral commodity, like electricity or communications infrastructure; they are continually making decisions about model behavior and acceptable use. Perhaps most fundamentally, even those who disagree about generative AI being on a path to becoming a derivative right may still support what that trajectory would warrant: the transparency, discretion, and procedural fairness detailed in the previous section.
The Implications of AI-as-a-Right for Private Providers
What does the language of privilege versus right offer the current discourse on generative AI? Framing generative AI access in terms of privilege and right clarifies what users should expect when providers restrict access: notice, explanation, proportionate treatment, and meaningful appeal. As AI becomes essential to societal participation, these protections become increasingly important.
Access to AI today through the main labs and cloud providers is governed by terms of use, usage policies, safety mitigations, supported country rules, subscription models, and abuse-detection systems. Together, three providers operate the world’s most widely used consumer AI systems, meaning the rules they impose on access are the rules by which a significant portion of generative AI access is governed. Providers may temporarily or permanently suspend accounts for fraud, child exploitation, terrorism, attempts to develop biological or cyber weapons, and other serious harms. OpenAI, Google, and Anthropic all stipulate consequences for usage policy violations including suspension and termination, with the ability to appeal.
Anthropic’s transparent reporting illustrates the scale of this governance: 11.4 million banned accounts, nearly 400,000 appeals and 42,000 appeal overturns in January-June 2026, following 1,700 overturns in July-December 2025. Although overturns are few relative to bans, these reversals underscore the importance of recourse when enforcement may be mistaken or overly broad.
Recognizing an access right remains compatible with this enforcement, as the conditions on electricity, telephone, and internet access discussed earlier illustrate. More capable AI may require stronger abuse prevention. But as access becomes a baseline necessity, private enforcement should carry clear public rules, notice, explanation, proportionate treatment, and meaningful review and appeal. Further protections could include portability and export of files, memories, and work product; interoperability where feasible; and heightened scrutiny of denials affecting education, employment, healthcare, legal navigation, or public benefits. It might also be appropriate for providers to follow a graduated enforcement ladder for some violations or harm areas, starting with (a) request-level refusal, (b) graduating to feature restriction (for example, no access to image generation), (c) to rate limiting or monitoring, (d) a temporary suspension, and then (e) full termination of access as a last resort.
Rights also permit restrictions for misuse. Courts have imposed severe internet-access limitations as sentencing conditions for offenses related to child sexual abuse material (CSAM). Similarly, cases involving AI-assisted financial fraud, deepfake abuse, AI-generated CSAM, or terrorism could justify restrictions on autonomous agents, image generation, or code execution, alongside monitoring or access limited to approved endpoints.
As AI becomes embedded in operating systems, public transit, education, health care portals, benefits systems, and workplace software, total exclusion could become as impractical as a total internet ban. The future is more likely to involve monitored or throttled access, or narrowly capable public interest models rather than full exclusion. Given the rapid increase in adoption of generative AI, policymakers, civil society, and industry would do well to consider this possible future state now, and to put in place the mechanisms necessary to ensure that the rights of users across geographies will be protected.
Conclusion
Access to generative AI today remains a privilege, mediated by private companies, supported regions, identity assurance, and usage policies. When dependence on privately governed AI capabilities becomes sufficiently consequential that access can no longer be treated as an ordinary consumer privilege, then the language of rights becomes apt for this conversation. This transition is not inevitable for every important technology: Rights become relevant when exclusion obstructs meaningful social, economic, and political participation. Although AI’s future remains uncertain, its rapid growth since ChatGPT’s launch almost four years ago warrants taking this possibility seriously.
That said, just as a right to internet access does not guarantee the fastest broadband, every paid platform, or unlimited bandwidth, an AI-access right need not guarantee immediate access to every frontier model, autonomous agents, unlimited inference, or any requested output. Nor would it confer immunity from usage policy violations or entitlement to harmful outputs. It would instead support affordable access to a minimum effective level of AI, procedural safeguards when private providers remove access, portability and interoperability, and proportionate restrictions for dangerous misuse.
Such a right would have both negative and aspirational dimensions. States should neither arbitrarily deny access nor pressure providers to do so for illegitimate reasons. They could facilitate access through subsidies, open standards, public interest procurement, education, and digital capability programs. The central policy question is where to set the floor: What minimum access enables equitable participation in a society increasingly shaped by AI, and what access to higher level intelligence might remain a privilege?
As AI becomes embedded in daily life, the burden of justifying exclusion will shift. Losing access may increasingly resemble losing electricity, telephone, or internet service rather than an ordinary consumer product. Restrictions would not necessarily be unconstitutional or always impermissible, but their consequences would warrant explanation, transparency, safeguards, procedural fairness, and recourse.
Disclaimer: The views expressed in this article are the authors’ own and do not represent those of OpenAI.
