The artificial intelligence industry has revived a familiar pattern. An emerging technology produces immense private power. Calls for public oversight follow. Then the regulated captures the regulator. Big tech CEOs simultaneously demand regulation and spend tens of millions lobbying to shape it. Meanwhile, the data that trains these systems, produced collectively across the digital commons, is enclosed into proprietary assets shielded from public scrutiny.
Austrian economics explains why state-led AI governance fails. But the critique points beyond state and market to a third option: the commons.
What Is the Digital Commons?
The digital commons is not a metaphor. It is a defined category of shared resources with specific governance characteristics.
The Digital Public Goods Alliance—a multi-stakeholder initiative endorsed by the United Nations—describes digital commons as specific digital resources, as well as the communities and modes of governance that support them. The key features are communal production, ownership, and governance. Open access alone is not enough. A commons requires a community with shared rules, participatory decision-making, and accountable monitoring.
Charlotte Hess and Elinor Ostrom defined knowledge commons as information, data, and content that is collectively owned and managed. Digital commons face the same threats as physical ones: commodification, degradation, and enclosure. Melanie Dulong de Rosnay and Felix Stalder place the concept in explicit opposition to both market and state: commons are holistic social institutions that organize collective action beyond market and state.
Applied to AI, the digital commons include open-source software, open datasets, public research repositories, and the training materials that AI systems consume. It also includes the collective labor of data annotation, content moderation, and user feedback. These resources are produced by millions of people, often without explicit consent or compensation, and then enclosed into proprietary assets.
A commons is not the same as open access. It requires clear boundaries, rules suited to local conditions, participatory decision-making, accountable monitoring, graduated sanctions, and nested institutions, the design principles Ostrom demonstrated can work without privatization or central state control.
The Suppression of Entrepreneurial Discovery
Israel Kirzner’s theory of entrepreneurial alertness shows why regulatory intervention in AI fails. Markets coordinate not only by processing existing information but by generating new knowledge through the discovery of previously unnoticed opportunities. That discovery process depends on private ownership, market competition, and the profit motive.
AI regulation suppresses this process. Regulatory regimes impose static categories on a dynamic technology. Firms must fit innovations into predefined boxes before those innovations can reach the market. Yet the most valuable AI applications are often ones no regulator anticipated—a novel use in rural healthcare, or an unexpected application of language models for disaster response. These ideas die not because concrete harm was shown, but because they did not fit the regulatory categories designed for a world that no longer exists.
Ludwig von Mises’s theory of interventionism reinforces the point. Each government intervention creates unintended consequences that then require further interventions. The result is a spiral of control in which the state attempts to manage what it cannot understand. Rules designed to address one set of concerns create new problems, which generate demands for still more rules. The regulatory burden grows while the benefits remain elusive.
The Calculation Problem and Simulated Prices
Mises showed that rational economic calculation is impossible without genuine, private-property-based market prices. Without prices for capital goods, planners cannot compare heterogeneous production plans. Our civilization, he wrote, is inseparably linked with our methods of economic calculation.
AI governance faces the same problem. Regulators must decide which applications are “safe,” “high-risk,” or permissible. These are economic questions about trade-offs, opportunity costs, and competing values. A regulator cannot calculate whether a diagnostic tool that saves lives in Nairobi is worth the compliance burden imposed by Brussels. That calculation requires prices, profits, losses, and consumer choice, signals that emerge only from voluntary exchange.
A central algorithm can generate shadow prices and simulate allocation. But market prices are not merely numbers. They are institutional outcomes of exchange under private property, competition, scarcity, profit, and loss. A simulated price may be useful inside a firm or model, but it does not possess the same epistemic status as a price generated by rivalrous bidding among owners, consumers, workers, investors, and entrepreneurs.
The Enclosure of the Digital Commons
Murray Rothbard’s theory of property rights offers a crucial distinction that most AI governance debates miss. Information, code, and patterns are non-rivalrous and therefore incapable of generating ownership-relevant conflict in the strict sense. Physical infrastructure, servers, cables, data centers, remains ownable. Digital patterns do not.
This has radical implications. When AI firms claim proprietary ownership over model weights and training data, they are not protecting scarce property. They are enclosing the commons. Sonia Katyal and Christopher Morten identify a “third enclosure movement,” following the historical privatization of land and the expansion of intellectual property, now unfolding through the systematic enclosure of the digital data commons. AI models are trained on books, articles, code, conversations, and images, collectively-produced resources, and then locked behind proprietary walls. The public that produced the training data receives nothing.
Rent-Seeking and the AI-Industrial Complex
Public choice theory explains what follows. State intervention alters incentives, rewarding political entrepreneurship over value creation. In recent years the AI industry has poured well over $100 million into lobbying and political influence, including efforts to shape or preempt state-level AI rules. This is not simple opposition to regulation, it is demand for the right kind of regulation. Compliance costs that are trivial for Google are existential for a startup. The “safety” narrative becomes a competitive moat.
Beyond State and Market: The Commons Alternative
The choice is not between the regulator and the entrepreneur, it is a choice of whether to enclose or to share. AI governance should not be a choice between state control and corporate control. It should be a choice for polycentric, democratic, and commons-based governance.
Applied to AI, this means institutions operating across scales that embed decision-making throughout the AI lifecycle rather than imposing top-down rules. Frameworks such as the Dignity Stack propose a commons-governed, horizontally-federated architecture in which any person, firm, or state may use and fund the system while governance remains horizontal, polycentric, and subsidiary.
This is not utopian. Real institutions are already emerging data trusts and cooperatives, open-weight models, public research consortia, and indigenous data sovereignty frameworks. These are experiments in polycentric governance, nested, overlapping, and accountable to the communities they serve.
Conclusion
Carbon markets enclosed the atmosphere. AI firms are enclosing the digital commons. The alternative is not state control or corporate control. It is commons governance: polycentric, democratic, and accountable to the communities that produce and depend on these shared resources. The data, knowledge, and infrastructure that AI depends upon belong to no one. They are a shared inheritance. They should be governed as such, not enclosed by the highest bidder or administered by the most distant bureaucrat.