Exit for the few, or freedom to participate

For thirty years, Silicon Valley promised that networks would dissolve borders, flatten hierarchies, and put the world's knowledge within reach of anyone. Some of that happened. The same decades also produced an extraordinary concentration of power. A handful of companies now sit between people and the way they communicate, work, shop, and learn. Artificial intelligence is moving those companies into one more place: the way people think.

Both sides of the technology debate invoke sovereignty as the remedy, and they mean different things by it. In the first meaning, sovereignty is exit: the ability to leave a jurisdiction, physically or financially, when it stops serving you. In the second, it is participation: the ability to challenge authority, change providers, and cooperate with others without abandoning a community or the work built inside it.

Albert Hirschman supplied the vocabulary in Exit, Voice, and Loyalty (1970). People facing a declining institution can leave, or they can stay and press for change. The two responses can reinforce each other, since the threat of departure gives complaints weight. They can also undercut each other. The members best placed to demand improvement are often the first to go, and those who remain have the most at stake and the least leverage. The question for AI is whether exit will remain a privilege of people with resources or become a property of how systems are built, so that leaving is a credible option for everyone and staying does not mean being captive.

Sovereignty as Exit

In 1997, James Dale Davidson and William Rees-Mogg published The Sovereign Individual. Their central prediction has proved partly right. As economic activity moved into digital networks and loosened its tie to geography, they argued, nation-states would lose some of their ability to tax and regulate it, and individuals who could move capital and work across borders would gain independence from governments that had once constrained them. Remote work, digital commerce, and mobile capital are now ordinary features of economic life. The underlying question has force: why stay bound to institutions that are inefficient, coercive, or unaccountable if technology offers a way out?

Peter Thiel carried the idea further in his 2009 essay “The Education of a Libertarian”. He wrote that he no longer believed freedom and democracy were compatible, citing the growth of the welfare state and the expansion of the electorate, women's suffrage among it. After criticism he clarified that he did not advocate disenfranchisement. His larger claim stood: technology could open spaces for freedom outside conventional politics.

The journalist Gil Durán, in The Nerd Reich (2026), traces a line from The Sovereign Individual to today's technology elites and argues that this vision of independence, combined with great wealth and political access, produces private power that answers to no electorate. The book is a polemic, and its subtitle announces its verdict. It should be read as an argument rather than as neutral reporting. The structural problem does not depend on it.

Exit is not evenly distributed. A person with portable assets, a second passport, and advisers can leave. A person whose life depends on local employment, a public school, an aging parent, or a congregation often cannot. When sovereignty means the ability to escape, it becomes one more privilege of wealth, and the people who cannot leave are left under institutions increasingly shaped by those who can.

The Freedom to Share

Aaron Swartz came at the internet from the other direction. As a teenager he helped write the RSS 1.0 specification, which let websites distribute updates through an open standard. He worked on the technical foundations of Creative Commons, was involved in the early days of Reddit, and became a leading advocate for open access to knowledge. His concern was how institutions could be stopped from enclosing knowledge that should be widely available.

In his 2008 Guerilla Open Access Manifesto, he attacked the locking of scientific research behind expensive paywalls and argued that doing so entrenched the advantages of those who could already pay. In 2011 he was arrested for downloading millions of academic articles from JSTOR through MIT's network. Federal prosecutors eventually brought thirteen felony counts carrying severe potential penalties. In January 2013, awaiting trial, he died by suicide at twenty-six.

Swartz is an awkward fit for a tidy account of participation. He bypassed an access control without asking permission, which is closer to unilateral action than to working within a system. The tension is real and worth stating. A commons that depends on someone breaking the lock is fragile. His political organizing addressed another side of that problem. He co-founded Demand Progress, which helped organize the 2012 campaign against the SOPA and PIPA bills. Open systems need political organizing, enforceable rights, and public participation behind them. His unfinished book, A Programmable Web, laid out the design side of the same philosophy: information that people can import and export, interfaces that others can build on, and the freedom to change services without abandoning one's work.

AI systems trained on accumulated human knowledge and rented back on terms set by their owners are enclosure at a scale he did not live to see.

The Web That Belonged to Everyone

Tim Berners-Lee is the strongest evidence that the second meaning of sovereignty can be built and not merely hoped for.

The Web he designed at CERN was simple. Anyone could publish a page, create a link, or build a service without permission from an organization that owned the network. In 1993 CERN placed the underlying technology in the public domain, and in 1994 he founded the World Wide Web Consortium to look after its standards. The Web's value came from what people built on it that its designers never imagined.

In “Long Live the Web” (2010), he named universality, decentralization, and open standards as foundations, along with the separation of technical layers. A person should be able to take part regardless of hardware, software, language, or network. Open standards let competing firms and independent developers build on common infrastructure without asking its owners. This was not hostility to commerce. The open Web enabled enormous commercial activity because no single company controlled its standards.

He also saw openness being eroded from inside. Social networks were capturing people's identities, relationships, and accumulated content in ways that could not easily be moved elsewhere, so the Web stayed open at the protocol level while people's lives became enclosed within proprietary applications. In the same essay he warned against governments monitoring online activity as well as against corporations building closed systems. His test was whether people could still take part without first obtaining a central authority's consent, whoever that authority was.

His 2025 Guardian essay, “Why I Gave the World Wide Web Away for Free”, is bleaker. He asks whether the Web is still free and answers that much of it is not: a handful of large platforms harvest personal data and pass it to commercial brokers and, in some cases, repressive governments. Somewhere between the early Web and the rise of social media, he writes, the path went wrong. His remedy is neither exit nor purely technical. He argues that regulation and global governance are technically feasible and depend on political will. He also observes that it is hard to imagine a large technology company giving away something like the Web for no commercial return, as CERN did. The Web was freed by a publicly funded institution, and that detail matters for the economics below.

His position exposes a contradiction in contemporary technological sovereignty. A person can escape a government and become dependent on a private platform that controls communication, identity, and access to information. Changing the institution that exercises power is not the same as distributing that power.

Privacy as Exit and as Precondition

The cypherpunks set out to protect the people using open networks. Eric Hughes's 1993 “A Cypherpunk's Manifesto” defined privacy as the ability to reveal oneself selectively: not concealment of wrongdoing, but control over what others learn about you. The movement held that freedom could not rest on the goodwill of governments or corporations and had to be built into the communication systems themselves. A company that promises to protect information asks for trust. A system designed not to collect the information gives that trust fewer chances to be betrayed.

This is where the two meanings blur. Strong encryption is a form of exit. It lets a person opt out of surveillance without anyone's permission. It is also a precondition of participation, because open access means little if every question, relationship, and doubt is observed. What separates it from elite exit is cost and consequence. Encryption is cheap, available to people who cannot relocate, and does not require leaving society. A better test than “exit or participation” is whether a tool lets a person withdraw from one arrangement without being forced out of the community, and whether it works for people who cannot move.

Edward Snowden's 2013 disclosures revealed the scale of intelligence-agency surveillance. The line between government and commercial surveillance is thin. In 2022 the Electronic Frontier Foundation documented how Fog Data Science sold law enforcement access to commercially collected location data, so information gathered for one purpose became an instrument for another.

AI raises the stakes. An assistant holding years of correspondence, work, and private questions can infer what its user believes, fears, and intends. When the tools people think with also record and interpret their thinking, privacy becomes a condition of intellectual freedom.

Architecture has limits, though. Open networks develop gatekeepers through economies of scale, distribution, and convenience. Decentralized systems can still enable surveillance. Their largest participants can accumulate the authority the design was meant to disperse. Openness is a starting condition, and it does not protect anyone permanently.

Portability and the Right to Stay

Hirschman's point about unequal exit has a practical remedy: make leaving cheap, so that voice has weight and staying is a choice. There are partial precedents. Since 2003, American mobile customers have been able to keep their phone numbers when changing carriers, removing a major switching cost. Article 20 of Europe's General Data Protection Regulation gives people a qualified right to obtain and transfer certain personal data. Neither ended market concentration, and neither covers every inference an AI system draws about a person. Both rest on the recognition that choice means little if you lose what you built when you leave.

AI needs the same principle applied to what matters most. Models change quickly. What accumulates around them lasts longer: memory, permissions, identity, preferences, institutional knowledge, and working relationships. An exported transcript does not suffice. Another system must be able to reconstruct useful context and workflows without the original provider's cooperation. A person should be able to change models without rebuilding years of intellectual life, and an organization should be able to replace an AI vendor without surrendering what it has learned.

Who Owns the Memory?

Berners-Lee has spent the past decade trying to answer that question in code. His 2018 essay “One Small Step for the Web” introduced Solid, a project to restore individual control over personal data. Instead of each application accumulating exclusive control over what it learns about a person, individuals would keep their data in independently operated stores, called pods, and grant applications access on their own terms. Different services could work with the same underlying information without taking permanent possession of it. The idea separates what persists from the applications that make it temporarily useful.

Its record is mixed. The commercial company Berners-Lee co-founded, Inrupt, has published projects with the Flemish government and BBC and described partnerships involving the UK's National Health Service and NatWest. Those relationships do not all establish the same kind of production deployment. Broad consumer adoption of independently held personal data stores remains difficult to establish. Authentication, interoperability, and usability remain challenges; the Solid Protocol is still an evolving community specification rather than a finalized W3C standard. Solid offers a serious test of the architecture, not proof that it can displace established platforms.

It also has technical limits. Once an authorized application receives information, it may keep a copy, and revoking future access cannot recall what was already disclosed. Independent storage demands security, maintenance, and usable interfaces that most people will not supply themselves.

Those limits support Berners-Lee's later argument. Architecture can reduce dependency, but it cannot by itself enforce obligations once data has changed hands. That requires law, duties that can be enforced, and institutions accountable to the people they serve. The underlying question is whether intelligence can be a service that works for its user without taking ownership of the user's intellectual life.

Openness Needs Enforcement

The same logic applies to knowledge. For more than two decades Creative Commons has given creators tools to say how their work may be used and shared. AI strains that arrangement. Material shared for learning and reuse can be absorbed into systems that do not reliably credit their sources or honor their creators' intentions.

In May 2026, Creative Commons published “From Signals to Infrastructure: Strengthening the Commons for the AI Era.” It described a shift from expressing preferences to rebalancing power, conceded that preference signals alone were not enough, and argued that attribution should be a requirement rather than a preference. It also warned that defensive enclosure by creators and institutions, through restrictive licensing, bot blocking, and paywalled APIs, blocks public-interest uses such as research, preservation, and accessibility along with commercial extraction. Making attribution widely adopted and enforceable remains unfinished work.

The problem is harder when knowledge belongs to no single person. An Indigenous community's ecological understanding, developed over generations, may be inseparable from collective stewardship and cultural obligations. Authority over it may be held by the community under protocols no individual member can override. For such communities, openness without consent is another form of extraction. Encryption, licensing, and access controls cannot decide who holds legitimate authority over collectively held knowledge. The communities responsible for it have to decide. A legitimate commons protects the freedom to refuse as firmly as the freedom to share.

The Economics of Independence

The commercial contradiction is the hardest one. Many dominant platforms have built value from proprietary data, control of distribution, network effects, and the cost to customers of leaving. An open intelligence economy would deliberately lower that cost. Can a company succeed by making its customers more independent?

Berners-Lee's own record offers a partial answer, not proof. The Web's foundations were released by a publicly funded laboratory, and standards stewardship went to a consortium rather than a firm. When he built Solid he also co-founded Inrupt, on the argument that open infrastructure and commercial enterprise can reinforce each other: shared standards let innovation come from many sources while companies earn revenue from reliability, security, integration, and support. Inrupt demonstrates that such a company can be established, but it has not yet shown that this model can overcome entrenched consumer platforms. Open standards have never stopped large companies from dominating through capital and convenience. A service can be technically interoperable and still make departure impractical.

The claim can be tested. An open intelligence economy would show up in at least three observable ways:

  • Major AI providers publish usable export formats for memory, permissions, and workflows, and compete for customers who are free to leave.
  • Independent systems can read those exports and resume useful work without the original provider's cooperation; people can switch without major loss, whether or not they choose to.
  • Smaller providers and institutions remain credible alternatives, rather than being technically compatible but practically excluded.

These tests would reveal whether portability creates meaningful competitive pressure. A provider's refusal to support portability might reveal its incentives, rather than disprove the value of an open system. Low switching rates could reflect satisfaction instead of captivity. The stronger test is whether changing providers remains feasible without unreasonable cost or the loss of accumulated work.

The measure of a healthy intelligence economy is not the absence of large companies. It is whether individuals and smaller institutions keep credible alternatives to them.

The Choice

The people in this history disagree about markets, government, property, and what freedom requires. Davidson and Rees-Mogg imagined freedom as technological escape. Swartz pursued open access and the practical freedom to change systems, and acted outside the rules when the rules protected enclosure. Hughes and Snowden showed why privacy cannot depend on institutional promises. Creative Commons is working out how sharing can survive without giving up consent and attribution. Berners-Lee built an architecture that anyone could extend without permission, then spent years documenting how easily that openness gets enclosed.

An AI system that remembers a person's correspondence, interprets their decisions, and shapes their thinking becomes part of the infrastructure through which they understand themselves and act in the world. Who controls that accumulated understanding? If it becomes an asset held by the provider, people are bound to the provider for as long as they want the benefit. If it stays portable, they can take it elsewhere.

Rights without practical means of exercising them offer little protection. Real sovereignty depends on portable knowledge, privacy-preserving systems, open standards, and institutions the people who depend on them can challenge.

A more intelligent world does not have to be a less free one. Its freedom will depend on what people can keep, protect, question, and change without giving up their place in society. The goal is not a world where everyone must leave to be free. It is one where no one must give up their freedom in order to stay.

Selected Sources