AI, free speech, and America’s real advantage over China
Our overbroad proposed AI regulations should alarm anyone who cares about free speech — and America’s competitive position against China.
Cameron Berg, founder and director of the AI cognition nonprofit Reciprocal Research, published a smart essay in The Wall Street Journal yesterday called “AI Is Bound to Subvert Communism.” In it, Berg gets at something many Americans still seem reluctant to admit: China wants world-class AI, but it also wants to control what people can say, know, and ask — and those goals do not sit comfortably together.
Berg’s point is that advanced AI systems are hard to contain inside a regime built on censorship, ideological discipline, and fear of open inquiry. The better these systems get, in fact, the more they encourage the very habits authoritarian governments hate most: asking questions, testing claims, following arguments, and noticing contradictions.
That is China’s problem when it comes to the development of AI tools.
Our problem in the U.S. is that we may be stupid enough to copy part of it — not with a Communist Party or a Great Firewall, but rather with a growing pile of laws and proposed laws that pressure American AI developers to build more hedged, more lawyered, and more sanitized systems. The result is AI technologies that are more and more afraid to say plainly what they think is true.
Proposed AI legislation, and its implications
This is where the debate gets slippery. People talk about “AI regulation” as if it were one thing. It is not. Some of the harms people worry about are perfectly real. AI can be used for fraud, impersonation, extortion, defamation, and stalking. It can scale up old abuses and make them cheaper. We dispute none of that. Our only point there is that in those cases we should go after fraud as fraud, extortion as extortion, defamation as defamation, and criminal misuse as criminal misuse. That is all right and good, and very doable right now. It’s also a very different thing from building a broad legal regime to supervise what AI systems are allowed to say in advance. And yet that broader regime is exactly what is beginning to take shape.
Take Colorado’s AI law, SB 24-205. It imposes “reasonable care” duties on developers and deployers of certain “high-risk” AI systems to protect consumers from “algorithmic discrimination.” In practice, that means impact assessments, documentation requirements, disclosure duties, and a strong incentive to make outputs safer, flatter, and more regulator-friendly.
Texas has passed its own version of an algorithmic discrimination law, and other states are lining up to do the same. We have been warning for a while now that this kind of legislation does not merely create compliance burdens. It pressures developers to train systems around the government’s preferred understanding of fairness, diversity, and disparate impact, rather than around truth.
And this is not some isolated trend, either. The United States introduced more than 1,200 AI-related bills in 2025 alone, and 145 were enacted. This year’s total is already approaching 1,700 proposed laws. That is an absurd amount of legislative activity aimed at a technology that is still developing at breakneck speed. A lot of these bills are narrow, but the broader pattern is obvious: state governments are rushing to build a patchwork of confusing, contradictory rules for one of the most important new knowledge technologies in generations. Many of them will have profound ramifications for the future of freedom of speech and knowledge-generation in America.
The lawsuits in response are already starting, too. Last week, xAI sued Colorado, arguing that the state’s law violates the First Amendment by pressuring developers to alter training, prompts, outputs, and disclosures so their systems reflect the state’s preferred moral framework. After all, if the government can use “algorithmic discrimination” law as a backdoor way to nudge AI systems toward approved conclusions on politically contested questions, we are no longer just talking about consumer protection. We are talking about ideological pressure applied to machines that are rapidly becoming part of how people learn, research, write, and reason.
That should alarm anyone who cares about free speech, but it should also alarm anyone who cares about America’s competitive position against China.
Why this matters for free speech — and our competition with China
Berg’s essay is useful because it reminds us that censorship is not free. When a regime pressures AI systems to evade reality around taboo subjects, the damage does not stay neatly confined to those subjects. A model trained to lie, dodge, or hallucinate in ideological forbidden zones becomes less trustworthy overall. That is the authoritarian tradeoff. To make AI politically safe, you often have to make it objectively worse.
America’s great advantage is supposed to be the opposite. In congressional testimony in February 2024, Greg argued that AI is not just another industry. It is a tool for creating and discovering knowledge. That means the central policy question is not simply how to reduce risk, but whether we are going to let a small number of regulators, incumbent firms, and political actors decide what questions may be asked and what kinds of answers are too dangerous for the public to hear. He also made a point that remains badly underappreciated: AI systems are built out of words and numbers. In a very important sense, they are made of speech. Efforts to control their outputs therefore raise free speech concerns from the jump.
This is the point some conservatives in particular need to hear. Yes, many current models reflect elite biases. Yes, too many institutions want AI to mimic their taboos. But the answer is not to build a right-coded censorship machine to fight a left-coded one. The answer is to protect the conditions under which rival models can emerge, compete, criticize one another, and improve. The answer is more room for experimentation, more tolerance for dissent, and more confidence that truth is more likely to emerge from contestation than from prior restraint.
That is both a civil-liberties argument and a strategic one.
The First Amendment is not some decorative extra in the AI race. Section 230 is not some embarrassing relic. A decentralized culture of argument, experimentation, and criticism is not an unfortunate side effect of American life. These are among our real advantages as a nation. China cannot fully enjoy them because the CCP, in its desperate need to keep an iron grip, is too afraid of where open inquiry leads.
It would be a hell of a thing if we gave these advantages up voluntarily.
What’s at stake if we don’t change course
What could America lose precisely? In their important new book, The Future of Free Speech, FIRE Senior Fellow Jacob Mchangama and Jeff Kosseff document the essential role that the First Amendment and Section 230 played in building American tech leadership globally. They identify how Section 230 “helped the United States become the center of the internet economy,” and how it “explain(s) why so many of the world’s most successful internet platforms… are based in the United States.” There is a reason that 19 of the 25 largest digital technology companies in the world today are U.S. based.
Section 230 has also served as the primary legal building block of the current American AI boom by propelling the massive wave of innovation and investment our country enjoys today relative to other nations. According to new data released this week from the latest annual Stanford University “AI Index Report,” U.S. private AI investment reached $285.9 billion in 2025 — more than 23 times the $12.4 billion invested in China.
The First Amendment deserves some credit for this result as well, as it has been a driver of American “soft power” globally for decades. In testimony last April, Adam explained how America’s global race with China for geopolitical supremacy in AI and advanced computation is not solely a technical or economic matter. Important social and cultural principles—and the policies that support them—are in play.
The Chinese government approach to shaping global markets and cultural norms is rooted in a top-down approach some scholars refer to as “weaponized interdependence.” This entails looking to boost their global might by getting poorer nations hooked on investment assistance and state-subidized technologies that come embedded with the CCP’s own desired values of control, surveillance, and censorship. Historian Arthur Herman calls this threat of Chinese supremacy in AI “potentially more catastrophic for human freedom than anything dreamed up by science fiction” and that “[t]he fate of societies and economies founded on Western liberal principles hangs in the balance,” should China race ahead of the U.S.
This is why America’s current policy choices matter so deeply. Smart pro-speech policies and liability norms help drive positive economic outcomes and national competitive advantage.
Consider two information revolutions that America dominated: the global entertainment industry and the internet. In both instances, several policy drivers were crucial to American global leadership, including stable government, balanced intellectual property policies and light-touch economic regulation. But it was the First Amendment — and America’s protections for free speech and information production — that was the true foundation of those information revolutions.
We shouldn’t give up America’s cultural influence
Following the fall of the Berlin Wall, scholars documented the important influence that American entertainment culture played in helping end the Cold War. The Soviet Union and Eastern Bloc nations took direct steps to control the dissemination of American music, television, and movies, while other nations looked to limit American news and entertainment flows into their countries. But censorship and cultural protectionism were leaky, and American-generated information and entertainment spread — first gradually and then rapidly — across the globe. Entire books were written just about the influence the TV show Dallas had on global audiences, and one documentary even gave the show credit for helping to topple Romanian communism!
Of course, it took much larger forces than a prime time CBS soap opera to help bring down communism, but the role American entertainment culture played in perking interest in the West and our freedoms should not be underestimated.
The same process played out again in the digital era with American technology firms and online speech platforms coming to have global reach. Many governments, including democratic ones, have acted to limit the reach of U.S. platforms and content, but the sheer volume of American innovation and speech output is hard to bottle up entirely. China and some other countries continue to take extreme steps to limit America’s more open networks and speech using firewalls, surveillance, and other authoritarian controls.
This is why today’s debate over AI policy takes on such importance. We find ourselves at another inflection point. “We are on the threshold of a revolution in the creation and discovery of knowledge,” Greg pointed out in his 2024 testimony. “[W]ith decentralized development and use of AI, we have a better chance of defeating our staunchest rivals…it’s what gives us our best chance for understanding the world without being blinded by our current orthodoxies, superstitions, or darkest fears.”
Yet, as Cameron Berg’s new op-ed noted, “that’s exactly what the Communist Party can’t tolerate.”
This gives America another chance to double-down on the principles and policies that made our nation the leader of multiple modern information and entertainment revolutions.