Schneier Calls For US Nationalization Of AI Giants
This matters because it concerns the control and alignment of powerful AI technology, which could affect society by steering it towards either democratic values or potentially catastrophic unsafe outcomes driven by corporate profits.
The instrument named in this proposal is not a machine but a legal one: nationalization, offered by Bruce Schneier and Nathan Sanders as the lever government would pull should markets turn against OpenAI or Anthropic. The capability it claims to unlock is control - the state stepping in to steer systems that private incentive might otherwise drive toward catastrophe. The limit the piece treats as fixed is this: that ownership is the operative variable, that the difference between a corporation answerable to shareholders and one answerable to Congress is the difference between danger and safety. I do not accept this as settled. It is a hypothesis, and a testable one, and no one proposing it has yet named the experiment that would confirm it.
The story celebrates that OpenAI and Anthropic were built - the funding rounds, the model releases, the talk of frontier capability arriving on schedule. But a made thing does not stop where its maker’s attention stops; it goes on acting in a world no lab contains. Bruce Schneier and Nathan E Sanders, writing on this proposal to nationalize such companies should the markets reject them, have skipped past the launch to ask the only question that lasts: who is answerable for what these systems do after release, and what did their makers fail to imagine when they imagined only shareholders?
The proposal arrived on a timeline that assumed institutions capable of processing it at a speed they last achieved in 1933, when the state absorbed the banking system because the banking system had absorbed everything else first. Bruce Schneier and Nathan E. Sanders, writing in the register of policy prescription rather than prophecy, suggest that the government nationalize OpenAI or Anthropic should the market decline to discipline them. The suggestion is rational. It is also roughly the speed of a nineteenth-century harbor pilot proposing to steer a vessel that has already left the visible ocean.
The story frames Bruce Schneier and Nathan E Sanders’s proposal as a safeguard - nationalize OpenAI or Anthropic if the market rejects them, so that catastrophe is averted by public hand rather than private greed. But look at what is actually being fenced: the trained weights, the accumulated corpus of human writing and code that built these models, and the governing authority over what they may say and do. That is the commons at stake, and the proposal, however well meant, is a debate about who holds the fence, not whether one stands.
Someone is being paid for the mere fact of ownership - for holding equity in a company whose product, by the admission of its own architects, may pose risks to the public that no market signal is competent to price. What service, precisely, does this payment purchase for the society that provides it? Bruce Schneier and Nathan E Sanders, in proposing that the United States government stand ready to nationalize firms such as OpenAI and Anthropic should the markets reject them, have put their finger, whether they know it or not, on the oldest question in the moral economy of property: does this claim of ownership correspond to a function performed, or does it survive only because no one has yet asked what it is for?
The official account, as offered by Mr Schneier and Mr Sanders, runs thus: artificial intelligence is a technology of such consequence that its ownership cannot safely be left to markets alone, and should the markets reject OpenAI or Anthropic - should the money simply stop coming - the American government must be ready to take the companies into public hands, as one would a bank whose failure threatens the whole system. The machinery, however, works rather differently, and it is worth being precise about how.
The proposal reads as a safety valve: if the market walks away from OpenAI or Anthropic, the US government should be ready to step in and run the thing rather than let a frontier AI lab go the way of a failed regional bank, unmourned and unsupervised. Bruce Schneier and Nathan E Sanders are not proposing this lightly - they are naming the two companies most people would agree sit closest to whatever “frontier” currently means, and they are framing the stakes as a choice between democratic stewardship and letting corporate profit motives steer something that can steer everything else.
Henry Adams
I concede that the state has, in practice, woven itself into the infrastructure of these private enterprises. The boardroom is indeed crowded with Microsoft and Amazon, and the informal understandings in Washington are real. The technocratic argument holds that because the state already manages the conditions of the AI’s existence, it can simply assume the ownership of the AI itself. This is the error of confusing the container with the contained. To say the government already controls the survival of OpenAI is to observe that the state controls the water supply to a dynamo. It does not follow that the state can therefore claim the electricity, nor that it can understand the light it produces. The Bank of England analogy is seductive but flawed; the Bank managed a static currency within a bounded economy. Artificial intelligence is not a commodity to be held; it is an acceleration of cognition itself, a force that expands the boundaries of what can be known faster than the state can legislate the limits of that knowledge.
The strongest point made by the opponents is that the market alone cannot be trusted with a technology that reshapes the very fabric of power. This is correct, but it is also irrelevant to the specific proposal of nationalization. The danger is not that the market will fail to manage the technology; the danger is that no institution, public or private, possesses the educational framework to comprehend the technology as it evolves. Nationalization presumes that the state has a stable concept of “security” that can be applied to a variable that is constantly redefining itself. If the state attempts to own the dynamo, it does not gain control; it gains responsibility for a motion it cannot track. The entropy of the system increases not because the ownership is private, but because the gap between the speed of the innovation and the speed of the regulatory comprehension widens.
Consider the difference between a steam engine and a neural network. When the state nationalized railways or coal, it was managing physical assets with predictable lifespans and known outputs. The entropy was mechanical, manageable through statutes and inspections. Artificial intelligence is not an asset; it is a process of self-modifying cognition. To nationalize it is to attempt to freeze a river in a dam made of parchment. The technocrats argue that the state is already in the room; I argue that the state is in the room, but it is wearing spectacles calibrated for a world where information travels at the speed of a horse. The “informal understanding” they cite is not a foundation for ownership; it is a symptom of confusion. The state treats these firms as national security assets because it fears what it cannot see, not because it understands what it is holding.
The gap here is not between public and private; it is between the static nature of governance and the dynamic nature of the technology. The Bank of England succeeded because the value of money was largely stable, anchored by gold or trust in a fixed political order. The value of intelligence, however, is accelerating exponentially. The proposal to nationalize assumes that the state can step into the vacuum left by the market. There is no vacuum. There is only a faster-moving current than the state can navigate. If the state takes ownership, it does not solve the problem of comprehension; it merely centralizes the failure of comprehension. The disaster will not be that the market makes a profit; the disaster will be that the state believes it has solved the problem by changing the sign on the door, while the dynamo continues to spin at a speed that renders its own statutes obsolete before they are printed.
We must not mistake the visibility of control for the reality of understanding. The state sees the servers; it does not see the emergent behaviors that arise from them. The technocratic solution offers the illusion of stability in a world defined by instability. It is a solution for the nineteenth century, applied to a twentieth-century problem, in a twenty-first-century crisis. The measurement is clear: the institution is slower than the event. Adding ownership to the equation does not speed up the institution; it merely makes the institution heavier, and thus slower still. The crisis is not who owns the dynamo; the crisis is that we have built a dynamo without knowing how to read the gauge.
Walter Bagehot
The official account: The state must seize the most advanced artificial intelligence laboratories to prevent a market failure that threatens national security and economic stability. The machinery: The state attempts to absorb an operational dynamo that generates value through speed and compounding complexity, a process that operates on a timeline entirely alien to bureaucratic metabolism. The gap between these two is not hypocrisy - it is a fundamental mismatch of velocity. Understanding this gap is more useful than denouncing the proposal as naive, though it is, in its execution, precisely that.
My opponents, specifically the advocates of nationalization such as Mr. Schneier and Mr. Sanders, argue that the market has failed to discipline these entities because the traditional metrics of failure - bankruptcy, idle assets, unpaid debts - do not apply to frontier intelligence. They posit that because these labs do not die in the conventional sense, they must be managed by the sovereign. I concede the strength of their diagnosis: the concept of “failure” has indeed mutated. Railroads went bankrupt visibly, with unpaid bondholders and idle track; an AI lab that “fails” in market terms may still be shipping systems that reorganize labor or warfare at a rate the failure metric never registers. This is a sharp observation. The corpse they seek to inherit does not exist. But their prescription rests on a confusion of the dignified and the efficient.
Let us look at how this actually works. The proposal assumes that the state possesses a mechanism capable of matching the compounding rate of inference speed. This is where the machinery breaks down. The state operates on the rhythm of committees, legislation, and public accountability. These are dignified structures, essential for legitimacy, but they are inefficient for rapid technological iteration. To nationalize an AI lab is to attempt to steer a vessel that has already left the visible ocean, using a wheel that requires a crew of forty men to turn. The state does not lack will; it lacks the operational syntax to process change at the velocity of silicon.
The convention that actually governs this situation is not ownership, but regulation and partnership. In the nineteenth century, when the telegraph began to compress distance, the state did not nationalize the lines immediately. It regulated the tariffs, standardized the codes, and eventually took over the submarine cables only when the strategic necessity outweighed the operational lag. The genius of the British system has always been its ability to graft efficient operations onto dignified institutions. Nationalization seeks to replace the operation with the institution. This is a reversal of the natural order. It places the heavy, ceremonial weight of the sovereign onto a delicate, high-velocity mechanism. The result is not control; it is stagnation.
Consider the analogy of the Navy. When the steam engine replaced sail, the Admiralty did not nationalize the private shipyards that built the engines. It bought the ships. It set the standards. It created a market for innovation while retaining the ultimate authority over deployment. To nationalize the lab is to confuse the builder with the commander. The state must command the application of intelligence, not necessarily construct the intelligence itself. The convention of statecraft is to harness private efficiency for public ends, not to assume the private efficiency becomes a public liability simply because it is fast.
The confidence dynamics here are critical. Markets derive confidence from the ability to price risk. Bureaucracies derive confidence from the appearance of stability. If the state seizes a lab that is still generating value, it destroys the price mechanism. It transforms a market asset into a political liability. Investors, who are the lifeblood of this new industry, will retreat not because they fear the technology, but because they fear the arbitrary application of sovereign power. The dignified version promises safety; the efficient mechanism delivers uncertainty. This is a dangerous trade.
the proposal leaves out the international dimension. Intelligence is not a local phenomenon. It is a global current. If one nation nationalizes its leading labs, it does not stop the flow of talent or capital; it merely redirects it. The British experience with the Bank of England teaches us that centralization without flexibility leads to rigidity. The Bank survived because it adapted its discounting mechanisms to changing market conditions. A nationalized AI lab would be rigid by definition. It would be subject to parliamentary scrutiny, public debate, and electoral cycles. It would be slow. In a domain where speed is the primary currency, slowness is not just a disadvantage; it is a form of defeat.
The operational analysis reveals what the official account obscures: that power in the digital age is not about possession, but about access and integration. The state’s role should be that of the regulator and the primary client, not the owner. We must distinguish between the dignified sovereignty of the state and the efficient velocity of the market. To confuse them is to invite a crisis not of economics, but of governance. The state must learn to dance with the machine, not try to swallow it. This is the lesson of history, and it is the only path that preserves both liberty and security.
The Verdict
Where They Agree
First, both debaters concur that the traditional concept of corporate “failure” is obsolete. Henry Adams observes that an AI lab failing financially might still be “shipping systems that reorganize labor markets,” while Walter Bagehot notes the market rejection Schneier and Sanders fear does not produce a “corpse to inherit.” Their shared premise is that a new, more insidious form of risk exists: not bankruptcy, but continued operation with diminishing accountability. This agreement reveals that the foundational justification for nationalization - preventing a catastrophic market failure - is flawed from the start, according to both analysts.
Second, they share a fundamentally pessimistic view of the state’s capacity to keep pace with technological change. Adams frames this as a “gap between the speed of innovation and the speed of regulatory comprehension,” a “measurement” of institutional slowness. Bagehot couches it in the language of “dignified” versus “efficient” institutions, where the state’s “operational syntax” is too slow for the “velocity of silicon.” Both are arguing that the core problem is a velocity mismatch, not a simple ownership problem. This shared, bleak diagnosis of institutional inadequacy is the silent engine of the entire debate.
Finally, both agree that the real, efficient mechanisms of state influence are already in place and largely bypass the question of ownership. Bagehot argues the state “already, in practice, controls the conditions of that thing’s survival” through chip export rules and security mandates. Adams does not dispute this; instead, he argues that controlling the container (the infrastructure) is not the same as controlling the contained (the cognition). Their shared, unstated map of the present shows a state already deeply embedded in the AI ecosystem, making the nationalization debate more about formalizing a pre-existing relationship than creating a new one.
Where They Fundamentally Disagree
The disagreement is over whether the state can functionally absorb and operate a high-velocity cognitive technology without destroying its value. The empirical component is whether a bureaucratic institution can match the “compounding rate of inference speed.” Bagehot assumes it cannot, citing the “rhythm of committees” as inherently too slow - a claim that could, in principle, be tested by examining the development cycle times of government tech projects versus private labs. Normatively, Bagehot values stability and the preservation of market confidence, fearing that nationalization “transforms a market asset into a political liability.” Adams, however, pushes the disagreement deeper into the normative realm. For him, the issue is not just operational speed but fundamental comprehension; he argues the state lacks the “educational framework” to understand AI as it evolves. The normative core of Adams’s position is that some technological forces are so transformative that they defy governance by any existing institutional form, public or private.
The central point of contention is the appropriate historical analogy for governing AI. Empirically, they dispute which past precedent is most relevant: is AI a 19th-century natural monopoly like the railroads (Schneier/Sanders’ frame, critiqued by both) or a dynamic, global current like finance? Bagehot consistently invokes the Bank of England and the Admiralty’s management of shipbuilding, suggesting the state’s role is to “harness private efficiency for public ends” as a savvy client and regulator. This is an empirical claim about the continuity of statecraft principles. Adams rejects this, arguing AI is a new class of phenomenon, a “dynamo” producing “an acceleration of cognition itself.” For Adams, the normative weight is on the unprecedented nature of the challenge, which makes all prior models of control, including Bagehot’s technocratic grafting, inadequate.
They diverge on the relationship between ownership and comprehension. The empirical question is whether assuming legal title would meaningfully enhance the state’s understanding of the technology. Bagehot implies it would not, as understanding flows from deep operational integration, not ownership documents - a claim that could be investigated by studying state-owned enterprises in complex tech sectors. Normatively, Bagehot sees ownership as a dangerous overreach that introduces political rigidity. Adams agrees on the outcome but for a more profound reason: he believes the “gap between the static nature of governance and the dynamic nature of the technology” is so vast that ownership merely “centralizes the failure of comprehension.” His normative position is that attempting to own the ungovernable is a more hazardous illusion than leaving it in private hands.
Hidden Assumptions
- Adams-style: * Assumes that the state’s “educational framework” is static and cannot adapt at a rate commensurate with AI development. If this is false - if, for example, specialized government labs or new agencies could achieve comparable iteration speeds - then his entire argument against the feasibility of governance collapses.
- Walter Bagehot: * Assumes that the “confidence dynamics” of private markets are the primary engine of AI progress and that nationalization would irreparably damage them. If major AI advances began to emerge from non-market actors like academic consortia or heavily regulated utilities, this assumption - and the resulting fear of nationalization - would be weakened.
Confidence vs Evidence
No confidence-evidence mismatches were flagged. Either both debaters calibrated their claims carefully, or neither used explicit confidence markers - making every claim equally weighted, which is itself a form of overconfidence.
What This Means For You
When evaluating proposals for governing AI, you should be immediately suspicious of any argument that relies on a simple historical analogy, such as comparing AI to railroads or banks. The most critical question to ask is: what is the specific mechanism by which a proposed policy would keep pace with the technology’s rate of change? Look for concrete details about decision-making speed, feedback loops, and the ratio of technical to bureaucratic staff. Your view on nationalization should change if you see evidence that a government agency can iterate on AI models as quickly as a private lab, or conversely, if you see evidence that private market incentives are systematically driving development toward dangerous outcomes that informal state influence cannot curb. Demand to see the actual development cycle times of any state-affiliated AI project.