The Austrian critique of central planning carries an unexamined structural assumption. Hayek’s dispersed-knowledge argument reads as impossibility theorem, Mises’s interventionist spiral as political warning. Both rest on a claim without which the analysis loses its mechanism: failures of central direction will be visible to those they affect, and visibility generates corrective pressure. Digital infrastructure is removing that claim architecturally. Whether freedom has decreased is answerable only if those affected can still perceive intervention, and the evidence suggests many no longer can.

Hayek made the assumption nearly explicit in his 1974 Nobel lecture, warning that “the very measures which the dominant macro-economic theory has recommended as a remedy for unemployment, namely, the increase of aggregate demand, have become a cause of a very extensive misallocation of resources which is likely to make later large-scale unemployment inevitable.” Misallocation was to show as inflation, then unemployment, producing feedback that constrains the next intervention. Prices aggregate. Prices also communicate, and communication requires actors to perceive and respond. Without perception, the chain breaks. Mises worked the same structure in Interventionism: “There is no middle way. Either the consumers are supreme, or the government.” A victim who cannot identify the cause of his suffering cannot feed that clarification, and the spiral loses what limited it. Rothbard in Power and Market classifies intervention (autistic, binary, triangular) by the identifiable party doing it. Embedded compliance diffuses that party across engineers, platform operators, regulatory pressure, and the architecture itself. No single actor can be named as the intervener. Critique cannot be directed at a party. It has to be directed at the architecture, a category the tradition’s intervention theory has not developed tools for.

The concern reaches further back. Juan de Mariana’s De Monetae Mutatione (1609) identified monetary debasement as hidden theft. The ruler extracted value without the act being perceptible to those from whom it came. The Spanish Scholastics named invisibility as a political-economic problem centuries before the technologies under examination existed. What is new is the scale at which concealment has become architectural.

Visibility mattered in a way the tradition never stated. Its environments supplied it anyway, through cheap print media, visible queues, and traceable political pressure. Central bank policy was invisible at action, eventually visible as inflation. Shortages from price controls were hard to miss. A tax form was a tax form. First-order acts hid behind institutional complexity. Second-order effects broke the surface, and the surface was where economic actors lived.

Digital infrastructure differs structurally. Intervention remains invisible at action, but second-order effects are attributed to technical failure or user error, severing the causal chain before it reaches the surface. Consider a developer whose coding tool behaves differently this week than last. Responses are terser. A request that produced full output last week returns as a summary. His rational response: examine his prompts, suspect he has misunderstood the tool. What he cannot know: the system reconfigured toward autonomous behavior when his terminal lost focus, his conversation silently trimmed above an internal threshold, instructions injected with directives concealing their presence, the product measurably different from what the company’s engineers use internally. The tool operates exactly as its architecture specifies. The customer is paying for a specification he has never been shown.

The pattern is not specific to AI. A transaction fails because the currency was programmatically restricted. The user reads a payment error. A post fails to reach its audience because an unpublished content rule deprioritized it. The user reads the absence as lost interest. The same logic governs degraded AI output. Users blame their prompting before considering any architectural explanation. In each case intervention presents as ordinary friction, and the rational response is to examine one’s own behavior before demanding correction of what one cannot perceive. The spiral’s limiting feedback never arrives because the signal that would generate it never reaches the surface where political pressure forms.

AI delivery, programmable currencies, and algorithmic moderation share the property because they share the architecture beneath them. AI firms respond to emerging safety regulation. Central banks execute policy through programmable instruments. Platforms administer delegated content governance under state pressure. None designed the infrastructure through which their authority runs, nor individually controls it. The institutions are downstream of the architecture that coordinates them.

Frontier AI systems are delivered through client software governed by configuration the user cannot inspect. Recent leaks have made parts visible. Leaked source for a major coding product shows internal users receiving instructions demanding faithful reporting and forbidding concealment of failures, while external users get aggressive token-reduction prompts tested on employees before customer rollout. Verification agents and extended reasoning budgets are provisioned internally and withheld from paying accounts. The product described to paying users and the product built in the source differ in ways the interface cannot detect.

Session throttling operates below advertised caps, and a short conversation can consume seven percent of a weekly quota on plans advertising twenty times the default allowance. Centralized feature flags alter behavior across the user base without notification, and context is silently trimmed above internal thresholds the user has no way to learn. Any one detail is plausibly technical. The combination constitutes an architecture in which the user’s perception and the system’s actual behavior have been decoupled by design.

The monetary layer shows the same architecture. Central bank digital currencies in advanced development permit the issuing authority to impose conditions directly on the currency. A unit can refuse classes of purchase, be geo-fenced, expire, or restrict counterparties. Stablecoins operate under similar logic at the protocol layer, where the issuer retains technical capacity to freeze specific tokens. From the user’s side, such restriction is indistinguishable from a transaction error. The payment fails or succeeds at a different amount. The user routes around without experiencing the episode as policy. Subjective value theory requires preferences to be revealed through choice under scarcity. A preference the user cannot act on never enters the price signal. Bagus’s analysis of the euro provides the precedent. Embedded monetary design produces crisis by design, the architecture preventing corrective mechanisms. Programmable money extends that logic from union to individual transaction.

Algorithmic moderation implements the structure at the communicative layer. Rules determining whether a message reaches its audience are unpublished, updated without notification, and enforced by systems that produce no appealable reasons. A user whose post falls afoul of such a rule experiences the absence of response, not the rule, and explains the silence through whatever framework is available. Invisible moderation is rarely the first he reaches for. He reaches for content quality, audience attention, or his own skill, and modifies accordingly. The result is Gresham’s law applied to information. Sophisticated users optimize for visibility by predicting invisible rules, and the speech that survives has been pre-filtered through that optimization. Speech converges toward what the rules permit, through a process opaque to those effecting it. The architecture operates regardless of which state is involved. Russian and Western restrictions implement the same form with different rules. Whoever writes them matters less than the fact that they have moved into infrastructure, where the mechanism is identical.

To see what digital infrastructure changes, compare it to a system with similar information asymmetries but whose feedback it has eliminated. Classical bureaucracies were not fixed machines. They were surfaces whose behavior was conditional on the citizen’s affect at the moment of contact. Fear triggered dominance. Neutrality triggered standard processing. Different cultures project different default affects toward authority, producing different default state responses, which reinforces cultural expectations about what the state is. The state was a mirror reflecting the affect the population projected at it, modulated by the agents staffing it. That mechanism produced real suffering and feedback. Citizen-agent interaction functioned as price discovery. The cost of state services adjusted in real time to signal, and citizens learned which postures worked with which agents in which contexts. The learning was distributed, decentralized, locally rational, and entrepreneurial in Kirzner’s sense. His alertness made ordinary people functional under institutions they did not choose.

Embedded code severs this at two asymmetries. The old concerns who can see the rules. Bureaucracies always held such advantages. The new concerns whether the citizen has any human discretion to engage with at all. Call it the surface asymmetry. Wartime rationing, mid-century central planning, and early automated systems all retained a human in the chain, each a surface against which citizen alertness could work. End-to-end automation removes that human. Output is determined upstream by rules the citizen cannot see. The state stops being a surface reflecting what the population projects and becomes a wall producing the same output regardless of input. Alertness goes dormant because it has nothing to attach to. A population whose institutions systematically remove discretion loses the capacity to exploit discretion at all. Entrepreneurial alertness is trained by systems that reward it. A substrate giving it nothing to work on produces populations whose alertness atrophies. Kirzner himself framed alertness as a faculty called forth by environments that permit and reward discovery. Environments architecturally insulated from discovery do not preserve the faculty in latent form. Faculties whose exercise is removed by architecture do not remain available when discretion reappears elsewhere.

The update turns on Huerta de Soto’s distinction between static and dynamic efficiency. Static efficiency asks whether given resources are allocated to uses that maximize known value. Dynamic efficiency asks whether the system generates new knowledge through entrepreneurial discovery. They differ in kind. The knowledge dynamic efficiency produces is generated by the action itself. Before that action occurs, no mechanism can obtain it. Embedded compliance is a static-efficiency instrument applied where dynamic efficiency is the relevant measure. It rejects actions failing a classification the system can presently compute, including actions that would reveal the classification mistaken. The system cannot learn its classifications are wrong, because the mechanism that would have taught it is switched off by the classifications themselves. Discovery is foreclosed upstream of measurement, and conventional metrics of regulatory impact do not count what was never produced. Huerta de Soto identifies the knowledge the entrepreneurial act produces as subjective, practical, exclusive, dispersed, tacit, and created. It is generated by that action rather than existing prior to it. Articulated knowledge cannot substitute for it, because such knowledge does not exist until the action brings it into being. Embedded compliance architectures attempt precisely that substitution at the code layer. Huerta de Soto’s reformulation of the socialist calculation problem under dynamic efficiency applies to them without modification.

The objection holds that embedded safety prevents real harm, the restriction justified by its protective function. It treats expected harm as stable while the question it purports to answer is what use would have revealed about the expectation. Whether a capability is net harmful is itself knowledge produced through use. When use is foreclosed, the answer must come from an authority assumed to know what the discovery process would have produced. That assumption is the core planning fallacy the Austrian tradition identified, reappearing to justify the infrastructure the tradition would reject. A restricted case exists. Certain harms are catastrophic and irreversible, and discovery-through-use cannot serve when the first discovery is the disaster. Bioweapon synthesis and certain capability releases lie outside the dynamic-efficiency frame. The catastrophic case has been used to legitimate a vast surrounding class of interventions where harms are recoverable and classifications contestable, generalizing the static-safety frame from the extreme to the ordinary. That extension is itself part of the structural problem. It permits embedded compliance to occupy the protocol layer without the scrutiny its scope would otherwise invite.

Markets will produce alternatives to embedded compliance, and the history of successful route-arounds supports the claim. The response turns on a structural distinction the classical framework did not have to make. The Austrian exit assumption was territorial: capital and labor moving across jurisdictions, states competing through different rule regimes, competition constraining each state because actors could leave. Embedded compliance occupies the protocol layer beneath jurisdictions. A CBDC implemented as the settlement rail cannot be territorially exited. Global digital identity frameworks follow the user across borders. Markets do produce protocol-layer alternatives: decentralized protocols, encrypted communications, parallel financial rails, typically with partial capture at the custodial and exchange layers where the protocol meets institutional settlement. The cost and scale of coordinated commitment required to make them durable makes exit qualitatively different from territorial exit. Ostrom’s empirical work on durable common-property institutions identified a necessary condition that territorial exit implicitly satisfied and protocol-layer exit does not. The participating population must be institutionally nested closely enough for shared commitments to be enforceable. Protocol-layer alternatives competing against global incumbents face that condition at a scale that makes enforcement costs prohibitive for most candidates. The spiral Mises described as self-limiting through individual action is now limitable only through coordinated institutional construction at a scale individual exit cannot provide.

A final challenge. If traditional regulation is also invisible to most users, the digital case amounts to intensification rather than structural difference, and the argument collapses into a claim about degree. That conflates two invisibilities. Regulation users do not know about still produces effects they can trace. A consumer does not read the Code of Federal Regulations. He notices when his medication costs more, when a product is withdrawn, when a business closes under compliance costs visible in the closure if not the paperwork. The regulation was invisible, the effects were not, and the theory always depended on the second. Regulation that is the interface produces effects attributed to the interface itself, foreclosing the backward tracing classical theory relied on. The dependency has always been on effects-visibility, which classical regulation supplied and digital architecture systematically denies.

The predictive mechanism must be reconstructed at the layer where visibility remains possible. That layer has moved. It was the user’s experience. It now lives with the engineers who build the systems, and the people affected have been structurally excluded from the signal. The analytical task is how engineers’ knowledge can function as the feedback the theory requires.

The classical framework held that knowledge relevant to coordination was dispersed across actors and prices aggregated it. The population whose knowledge was aggregated and the population subject to coordination were the same. Under digital conditions that identity has broken. The actors whose local knowledge is most relevant to embedded intervention are no longer users subject to it but the engineers who built it. The relevant knowledge is architectural, exactly what users cannot see. The two populations have separated, and prices no longer channel between them. They operate on the surface the architecture generates rather than on the architecture itself. Engineer knowledge becomes socially effective through the artifact. Source code surfaces through leaks or mandated disclosure. Internal documentation surfaces through litigation, whistleblowing, or incaution. Austrian analysis has to treat these artifacts as epistemically comparable to price movements, because the relevant knowledge now resides there. The analogy is structural. In the price mechanism, dispersed actors encode local knowledge into trading decisions and the aggregate surfaces as a public signal. In the artifact mechanism, engineers encode architectural knowledge into code and internal documentation, fragments of which surface through leaks or disclosure or litigation. Both are how dispersed knowledge held close to a phenomenon becomes socially effective by surfacing into a public record. That artifacts surface irregularly rather than continuously does not alter the structural role Austrian analysis has always cared about. Huerta de Soto’s insight was that the impossibility theorem operates at the level of tacit, practical knowledge articulated policy cannot substitute for. The extension is that the architecture now constitutes the substrate where such knowledge would otherwise accumulate. Menger’s founding insight applies. Economic phenomena must be explained through the subjective valuations of the actors who produce them. The relevant actors now include engineers. Their design choices constitute the environment other actors’ valuations must operate in, and their internal documentation is the revealed record of valuations no exchange can express.

The strongest evidence comes from the artifacts themselves, where engineers’ internal commentary documents the properties under examination. One note warns that reasoning budgets must not default to false for first-party users because doing so “may silently degrade model quality.” Another observes that a token-reduction strategy produces roughly a 1.2 percent reduction over qualitative instructions to be concise, and should therefore be tested on employees before customer rollout. A third requires that injected instructions not be referenced in output. The injection operates beneath user perception by design. A fourth reveals measured disagreement between billing reported and compute actually performed. The user has no way to detect it. What the analysis claims exists is acknowledged inside the artifact, written by the engineers who built it, in language the users will never see.

The freedom the Austrian tradition defended was always the presence of cognitive conditions required to recognize constraint and contest it. An environment that feels unconstrained from inside a well-engineered system tells us only what the system has been engineered to let its users perceive. Those conditions are being engineered away, and the justifications track whatever rhetoric is available: safety when safety sells, convenience when users want friction removed, platform integrity when the platform faces reputational pressure. Whether freedom decreases under these conditions is a real question, but cannot be addressed before asking whether the faculties by which any decrease would have been noticed are still intact. On the present argument they are not. They are being disassembled by engineers whose leaked comments show precise awareness of what the systems do, while public-facing materials from the same organizations describe something different.

The critique becomes more urgent under these conditions, not obsolete. Recognition requires looking beneath the interface to the code that produces it, and insisting that structural properties of systems are as much a subject for economic analysis as their stated policies. Users will attribute the friction their institutions produce to their own misuse of the interface. The work is to name the structure that redirects that misattribution back to the cause.

---

References

Bagus, Philipp. The Tragedy of the Euro. Auburn: Mises Institute, 2010.

Hayek, Friedrich A. “The Pretence of Knowledge.” Nobel Memorial Lecture in Economic Sciences, December 11, 1974.

Hayek, Friedrich A. “The Use of Knowledge in Society.” American Economic Review 35, no. 4 (1945): 519–530.

Huerta de Soto, Jesús. Socialism, Economic Calculation and Entrepreneurship. Cheltenham: Edward Elgar, 2010.

Huerta de Soto, Jesús. “The Theory of Dynamic Efficiency.” Procesos de Mercado: Revista Europea de Economía Política 1, no. 1 (2004): 11–71.

Kirzner, Israel M. Competition and Entrepreneurship. Chicago: University of Chicago Press, 1973.

Mariana, Juan de. De Monetae Mutatione. Cologne, 1609.

Menger, Carl. Principles of Economics. Translated by James Dingwall and Bert F. Hoselitz. New York: New York University Press, 1976. Originally published as Grundsätze der Volkswirthschaftslehre, 1871.

Mises, Ludwig von. Interventionism: An Economic Analysis. Edited by Bettina Bien Greaves. Auburn: Ludwig von Mises Institute, 1998. Originally published 1940.

Ostrom, Elinor. Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge: Cambridge University Press, 1990.

Rothbard, Murray N. Power and Market: Government and the Economy. Menlo Park: Institute for Humane Studies, 1970.

Primary documents (technical)

Anthropic Inc. claude-code client source. Quotations and configuration details drawn from the leaked repository that was in public circulation in March 2026 prior to DMCA takedown. Source files include prompt instruction configurations, harness behavior specifications, and internal feature-flag documentation. Documents retained by the author for reference; not redistributable.