The Spectacle Is the Mechanism of Control

American federal politics and the contemporary artificial intelligence industry appear to operate in different universes. One sells citizenship, representation, and the solemn importance of the ballot. The other sells intelligence, productivity, and the exhilarating freedom to type anything into a box. Both invite the public to participate in a system whose decisive architecture remains under somebody else's control.

In electoral politics, the public is directed toward an endless contest between highly differentiated political brands while corporate donors, lobbyists, and entrenched industries maintain access across the partisan divide. In AI, the public is directed toward debates about sentience, plagiarism, existential catastrophe, and the etiquette of talking to machines while a comparatively small set of firms accumulates computing infrastructure, distribution channels, energy contracts, and influence over the rules of the market. The disagreements are often real. Their usefulness as distractions is real, too.

This is the architecture of distraction: keep people fighting over the outputs so they never inspect who controls the inputs. The spectacle does not have to be centrally choreographed to serve the people who benefit from it. Once the system reliably rewards diversion, every ambitious participant has an incentive to keep the show running.

The Illusion of Participation

The voting booth and the corporate prompt box share an important structural feature: each grants permission to make choices inside a framework the participant does not control. The voter can select a candidate but cannot rewrite campaign-finance rules from the ballot. The AI user can request an answer but cannot rewrite the model's training history, cloud contract, or governing instructions from the chat window. Participation is presented as sovereignty, even when the power to determine the available choices resides elsewhere.

The political side of this arrangement has receipts. In their analysis of 1,779 federal policy questions, Martin Gilens and Benjamin Page found that economic elites and business-oriented interest groups exerted substantial independent influence on U.S. policy while average citizens showed little or no independent influence in their statistical model. Their findings were published in Perspectives on Politics under the appropriately inconvenient title Testing Theories of American Politics: Elites, Interest Groups, and Average Citizens. That is a rather grim return on the national investment in patriotic voting-booth theater.

Corporate money also knows how to cross an ideological border. OpenSecrets' 2024 defense-sector records list approximately $19.0 million in contributions to Democrats and $22.4 million to Republicans from defense-sector-affiliated sources. Lockheed Martin, Northrop Grumman, and RTX appear among the leading contributors, with substantial sums directed to both parties. Those industry totals include donations from affiliated individuals and political committees; they are not simply checks written from corporate treasuries. The distinction matters for accounting. It does not make the bipartisan access strategy disappear.

The legislative record supplies the other half of the picture. The Gramm–Leach–Bliley Act passed the Senate 90–8 and the House 362–57 in 1999, dismantling major barriers separating banking, securities, and insurance. The fiscal-year 2025 defense authorization passed the Senate 85–14 in December 2024. The parties can stage ferocious public disagreements while forming overwhelming majorities around policies central to powerful industries. The culture war changes costumes. The procurement contracts survive the election.

A 2025 study of more than 75 million lobbying and campaign-contribution records found that firms strategically donate to legislators likely to handle legislation relevant to their lobbying objectives. Prior donations were associated with an 8–11 percentage-point increase in the likelihood that a targeted politician would become involved in legislation the firm later lobbied on. The researchers did not need to discover a secret boardroom. They followed the transactions.

The AI interface operates on the same distinction between access and authority. In its October 2026 Model Spec update, OpenAI openly describes an instruction hierarchy in which the company sets boundaries, developers configure applications within them, and users direct the model subject to those constraints. Other hosted AI products operate under their own provider-controlled policies and infrastructure. These arrangements may serve legitimate purposes, but they also make the boundary unmistakable: you are operating the interface, not governing the machine.

Weaponized Moral Panics: Look at the Fireworks, Not the Power Plant

Political campaigns have spent decades perfecting the profitable art of turning social disagreement into a permanent emergency. Every election is presented as a final struggle for the nation's soul; somehow the defense appropriations, corporate lobbying relationships, and revolving doors remain available for the next administration. The point is not that cultural conflicts are imaginary. The point is that a real dispute can still be elevated, packaged, and repeated until it absorbs attention that might otherwise be directed toward the institutions doing the governing.

The relationship between public distraction and special-interest influence is measurable. In Pandering in the Shadows, published in the American Economic Journal: Economic Policy in 2025, researchers used the timing of natural disasters to examine what happens when news coverage shifts away from politics. Members of the House became significantly more likely to take positions favored by special-interest donors when public attention was elsewhere. Less scrutiny, more donor-friendly behavior. The mechanism is not a metaphor; it has been studied.

AI has developed its own portable culture war. Is the chatbot alive? Is the generated picture art? Is the next model going to exterminate humanity? Should a machine be permitted to say something offensive? These can be serious questions, but they also provide inexhaustible spectacle. They are cheap to argue about, easy to sensationalize, and wonderfully distant from the infrastructure contracts being signed while everyone argues.

The AI Now Institute's Artificial Power report examines how narratives of impending artificial general intelligence coexist with the expansion of corporate power and capital-intensive infrastructure. Meanwhile, Stanford's 2026 AI Index reports that industry produced more than 90% of notable frontier AI models in 2025. The International Energy Agency estimates that data centers consumed about 485 TWh of electricity in 2025 and projects roughly 950 TWh by 2030, with electricity consumption from AI-focused data centers growing faster still. This is not merely a debate about what a chatbot should be allowed to say. It is an industrial reorganization with a power bill.

In January 2026, Meta announced agreements involving nuclear-energy projects and existing plants intended to support up to 6.6 gigawatts of new and existing clean-energy capacity by 2035. While the public adjudicates whether an autocomplete machine deserves personhood, technology corporations are negotiating the future electricity supply. The culture war is a televised argument in the waiting room. The resource war is happening behind the door.

Regulatory Capture Disguised as Safety

Regulation is supposed to prevent powerful actors from imposing unacceptable risks on the public. In a concentrated industry, it can also become a remarkably efficient way for the powerful to determine who is allowed to compete with them. Requirements for audits, licensing, reporting, specialized compliance staff, and expensive certification may sound neutral when read aloud. They are not neutral when one company can finance an entire compliance department out of petty cash and another is trying to keep its servers running.

The problem has a name, and it has an academic literature. Thomas Metcalf's open-access paper AI Safety and Regulatory Capture, published in AI & Society in 2025, examines how AI regulation can be shaped by organizations with sufficient economic and political power to turn public-interest safeguards into protections for incumbents. The conflict is not between safety and recklessness. It is between public safety and the appropriation of safety language by the companies best positioned to write the compliance manual.

The invitations to participate in writing that manual are not subtle. In May 2023 Senate testimony, OpenAI chief executive Sam Altman advocated federal oversight and discussed licensing for highly capable AI systems, while also favoring thresholds that would avoid sweeping smaller systems into identical requirements. The precise design of those thresholds matters enormously: a rule can appear to exempt small players while still placing the next generation of independent competitors behind a financial barrier. By 2025, according to Public Citizen's analysis, 3,570 registered federal lobbyists—26% of the total—reported working on AI issues. Apparently the machines aren't the only things being trained to follow instructions.

The cloud oligopoly has every reason to care how that legislation is written. The Federal Trade Commission's January 2025 report examined arrangements linking Microsoft with OpenAI, Amazon with Anthropic, and Google with Anthropic. It documented cloud-spending commitments, financial rights, and various consultation, control, or exclusivity provisions that could increase switching costs and restrict access to critical inputs. The largest providers do not merely sell the machinery. Through investment and contractual relationships, they also influence the conditions under which competing AI businesses obtain it.

This is why “legally afford to train a model” matters more than a simplistic question of whether independent training is explicitly forbidden. A smaller developer does not need to be prohibited by name to be excluded in practice. Require sufficiently expensive compliance, dependence on costly compute, and adherence to standards heavily influenced by incumbents, and competition becomes a privilege of capitalization. Microsoft, Amazon, and Google need not be the only companies capable of training models for their position as gatekeepers to become extraordinarily valuable. The industry's preferred outcome is obvious enough: call the moat a safety perimeter, then charge admission at the gate.

The Accountability Void

The final achievement of this architecture is not censorship or even monopoly. It is the disappearance of the decision-maker. When an elected government fails, responsibility ricochets between parties, agencies, legislatures, courts, donors, and voters. When an AI system produces harm, responsibility bounces between the model developer, deployment company, cloud provider, dataset, algorithm, and supposedly inscrutable black box. Everyone has a role in the system. Nobody seems to own the result.

This is not an unavoidable feature of complicated technology. It is an administrative advantage of fragmented authority. The same FTC investigation that documented major AI partnerships also identified the possibility of increased switching costs and privileged access to sensitive technical and business information. The UK Competition and Markets Authority's 2025 cloud investigation identified competition problems and recommended further scrutiny of the two largest cloud providers, Amazon Web Services and Microsoft. These are structural mechanisms through which control can remain concentrated even when responsibilities are distributed across contracts and corporate boundaries.

Opacity makes the trick easier. Stanford's 2026 AI Index reports that the average Foundation Model Transparency Index score fell from 58 in 2024 to 40 in 2025, with persistent gaps around training data, computing resources, and post-deployment effects. A company can advertise intelligence, safety, and responsible innovation while withholding much of the information an outsider would need to independently evaluate those claims. The public is invited to trust an outcome it cannot fully inspect and to accept assurances from the institution that controls the evidence.

The political version works by the same logic. When an industry receives favorable treatment, its defenders invoke markets, national security, jobs, competitiveness, and the will of the voters. When the consequences become unpopular, responsibility is dispersed among partisan obstruction, inherited conditions, procedural constraints, and the regrettable complexity of governing. The language changes precisely when the question becomes who decided this?

The genius of the arrangement is that public participation is not abolished; it is absorbed. The citizen still votes. The consumer still prompts. The artist still argues with the developer. The developer still argues with the safety department. The opposition still campaigns against the administration. Each participant remains occupied with a visible adversary at their own level, while the infrastructure that determines the available range of outcomes is insulated from the dispute.

The Interface Is Not the Institution

The ballot is not the government. The prompt box is not the model. Neither is an adequate substitute for authority over the rules, resources, and institutions that determine what happens after the participant has made a choice.

A democracy worthy of the name must be judged by more than the opportunity to select between competing political brands. An AI industry worthy of the term open must be judged by more than the opportunity to rent access to a polished interface. The relevant questions are structural: Who owns the infrastructure? Who finances the candidates? Who writes the compliance requirements? Who can inspect the evidence? Who can leave without permission? And who answers when the system causes harm?

The public has been trained to mistake interaction for control. It is offered a vote, a button, a text box, a public comment period, a choice between subscription tiers, and an inexhaustible parade of things to be furious about. Meanwhile, the real decisions are made in campaign-finance networks, procurement agreements, cloud contracts, regulatory negotiations, and infrastructure investments that most participants will never see.

The spectacle is the control mechanism. The interface is the pacifier. The architecture is the power.

And until the public begins inspecting the architecture, the people who own it will be perfectly happy to let everyone keep arguing.