Yesterday morning at around 4 a.m. in Washington, the Senate voted 99-1 to strip the state artificial intelligence law moratorium language I wrote about in May from the federal budget bill. Just hours earlier, it had seemed as if the provision were to pass, as Senator Ted Cruz and Senator Marsha Blackburn had reached a compromise that appeared to have assuaged Sen. Blackburn’s concerns on the bill. Some of us were also concerned that Sen. Chuck Schumer was not prioritizing the issue and that some Democrats would also defect. Ultimately, however, the provision failed, and for the time being, states can continue to lead the way on legislating around not just AI, but also manipulative algorithms and kids online safety. (Let me subtly emphasize “for the time being,” because it seems overwhelmingly likely to me this proposal will come back in some form.)
Although Sen. Cruz’s professed motivation for the moratorium is to stop “woke” state AI laws, the argument in support of the moratorium that policymakers tend to find the most compelling — even among some Democrats — is the argument that America cannot afford to constrain innovation, because otherwise we might lose this technological struggle over AI to China.
As I’ve written previously, tech leaders like Marc Andreessen have used the rhetoric of American competitiveness and innovation to deflect regulation while actually investing in products that harm children and society, representing a fundamental corruption of what innovation should mean for national progress. But even well-informed policymakers who do see that that rhetoric is self-serving and hollow — and more of them exist than you might appreciate, on both sides of the aisle — still agree with the premise that taking the restraints off tech companies is key to maintaining the U.S. global dominance in tech.
Why?
It’s because they believe it’s worked before, and so that means it will work again.
“Inventing” the Internet
I was only 17 during the 2000 U.S. election, and so was unable to vote in it; I also was a poorly informed teenager. But I don’t think I’m alone when there’s really only three things I recall about Al Gore’s candidacy: the court challenge and the hanging chads in the Florida recount, SNL’s parodies of both Gore and Bush, and Gore being mocked for having claimed to have “invented the Internet.”
Turns out, Gore never said anything like that, but did take credit in an interview for leading on legislation that enabled the commercialization of the digital world. Gore had said as far back as 1989 that "the nation which most completely assimilates high-performance computing into its economy will very likely emerge as the dominant intellectual, economic, and technological force in the next century.” That position reflected a bipartisan consensus that would fundamentally reshape global power dynamics. By the mid-1990s, Congress and the Clinton administration, working together, privatized the internet, deregulated the telecommunications sector, and created legal frameworks that allowed American tech companies to operate with minimal oversight (the infamous Section 230). In so doing, U.S. policymakers essentially unleashed Silicon Valley on the world, betting that American entrepreneurial energy would capture global markets before other nations could establish competitive digital economies.
Credit where credit is due: this strategy proved remarkably successful on its own terms for nearly two decades. The United States didn't merely dominate the early internet; it fundamentally shaped its basic infrastructure and cultural DNA in ways that persist today. Consider the most mundane yet revealing example: web addresses worldwide use the Latin alphabet rather than Cyrillic, Arabic script, or Mandarin characters. This seemingly technical detail forced the entire global population to learn English to participate meaningfully in the digital economy. American platforms like Google, Facebook, Microsoft, and Amazon colonized digital space itself, establishing the protocols, standards, infrastructure, and cultural norms that would govern online interaction for billions of people. Thus, the bipartisan strategy of deregulating-in-advance1 delivered exactly what Gore and Clinton had promised: comprehensive American technological hegemony that extended far beyond market share to encompass the very architecture of global communication.
Today's tech executives and investors, by invoking the specter of Chinese technological competition to appeal to lawmakers' geopolitical instincts, hope to continue the party into the 2030’s. When Peter Thiel or Marc Andreessen warns that legislating on AI will hand technological leadership to America’s adversaries, they're appealing to something that many policymakers who remember and understand what Congress and the Clinton administration did in 1990s remain proud of, no matter how far from serving American interests the tech sector has gotten. Silicon Valley has learned to weaponize American anxiety about losing technological supremacy, and in so doing has transformed legitimate geopolitical concerns into a convenient shield against domestic accountability.
From Innovators to Digital Bandits
Imagine life in a prehistoric settlement in Mesopotamia, thousands of years ago. Hunter-gatherers have begun to become more sedentary, as their dependence on grains that grow along the rivers as a dietary staple has increased. As these nascent farming communities begin to stockpile excess grain, other groups discover that this excess is easy to steal or take by force. Raiders thus sweep through sporadically, stealing grain stores and livestock before disappearing. The settlers would live in constant fear, never knowing when the next attack might come.


Then eventually, a particularly cunning group of raiders arrives with a different proposition. Instead of pillaging, they would stay, and protect the settlement from other raiders; but in exchange, they would take a share of every harvest, collect taxes on the village market, and require young men to serve at their command.
Over time, these stationary bandits evolved into what we recognize as governments: they built roads, maintained granaries, established laws, and provided some public goods. But the fundamental relationship remained extractive; there was always a substantial gap between the taxes collected and the services provided, with that surplus funding the rulers' magnificent lifestyles. This theorized transition from "roving bandit" to "stationary bandit," as economist Mancur Olson termed it, is a compelling idea for how the first governments came into being, which were by their nature predatory.2
In pushing for the proposed regulatory moratorium, today’s tech giants offer a similar deal: “We will protect the U.S. from foreign technology in exchange for unquestioned digital sovereignty.” And, as it turns out, the deal is predatory and ultimately extractive, just as it was for those theoretical prehistoric settlements.
Among the first people to fully grasp this new form of extraction have been artists and writers, whose economic livelihoods depend directly on the intellectual property that tech companies have systematically appropriated through AI training. When Meta trains its systems on millions of pirated books without compensating authors, or when image generators scrape artwork without permission, the theft is concrete and measurable. These creators can see their grain being taken from their stores in real time.
Unlike governments, however, tech companies face no electoral accountability for their governance decisions. We cannot vote them out when their algorithmic systems promote division, exploit children, or undermine democratic institutions. And while even prehistoric farmers who could at least see how much of their production disappeared into their local warlord’s coffers, we remain largely blind to the scope and value of what these digital sovereigns extract from our daily lives.
Bandits, Enshittification, and the Limits of Corporate Patriotism
The flaw in applying the 1990s playbook to AI lies in assuming that American tech companies remain fundamentally aligned with American interests. This assumption has proven catastrophically naive as these corporations have demonstrated neither patriotism nor respect for democratic sovereignty when their business interests are at stake. Meta's willingness to hand over Hong Kong democracy activists' data to the Chinese Communist Party (covered in the hearing I wrote about here) reveals how quickly principles of democratic solidarity evaporate when confronted with market access opportunities. When Mark Zuckerberg bans his own children from using Meta's products while defending those same products as essential for American competitiveness, we witness the contempt these executives actually hold for the public they claim to serve.
The enshittification cycle that Cory Doctorow has described illustrates exactly how this dynamic unfolds: platforms begin by serving users to build market share, then abuse users to benefit business customers, then abuse everyone to extract maximum value for shareholders. This pattern tells us precisely how AI will develop if we repeat the 1990s approach. Initially, AI companies will offer genuinely useful services to build user dependence, much as early social media platforms facilitated authentic human connection. Once users and business customers are locked into these systems, the companies will begin extracting value through increasingly manipulative and harmful practices. Finally, they will abuse everyone in the ecosystem to maximize shareholder returns, leaving behind degraded products that users hate but cannot abandon due to switching costs and network effects. This is the model of the stationary bandit adapted to the digital world.
A Different Kind of Tyranny
This reveals something that should particularly concern conservatives who traditionally worry about concentrated power and governmental overreach: we are witnessing the emergence of private tyrannies that possess governmental scale and influence without any of the democratic constraints that supposedly limit state power. These digital stationary bandits have grown large enough to compete directly with the governments of advanced economies, but they operate without constitutional limitations, electoral accountability, or meaningful transparency requirements.
The companies seeking anticipatory deregulation have already demonstrated their willingness to exploit technology to extract maximum value from human vulnerability, regardless of the consequences for democratic institutions, child welfare, or social cohesion. Unlike traditional governments that at least theoretically derive their legitimacy from popular consent, our digital sovereigns answer only to shareholders and venture capitalists who explicitly celebrate their ability to monetize human vulnerabilities.
The 1990s deregulation-in-advance strategy succeeded because it aligned private profit motives with national strategic objectives during a unique historical moment when the internet was a blank canvas. But today's tech giants have already captured the internet and demonstrated their fundamental hostility to democratic governance, user welfare, and basic honesty. Trusting these same actors to responsibly develop artificial intelligence based on their promises to maintain American technological leadership represents a profound category error that threatens to create tools of oppression more sophisticated than any government has ever possessed.
Lawmakers got to the right answer this week, though possibly for the wrong reasons. Let’s hope that when — not if — they are presented with the same decision again soon, they make the same choice.
This is an intentional riff on historian Timothy Snyder’s writing on “obeying in advance” (a.k.a. anticipatory obedience) being a feature of authoritarian regimes.
Matt Yglesias also has written on this theory and applied it to contemporary politics, most recently arguing that the progressive state in places like New York constitutes a modern form of the stationary bandit.