America’s AI Regulation Should Be Rooted in Free Enterprise Principles | American Enterprise Institute

If you take seriously the optimistic San Francisco Consensus on the direction of AI capabilities—such as on display in the much-discussed “AI 2040” world-building from the AI Futures Project—we might soon enter a period where advanced artificial intelligence will give its national possessor a considerable military edge. And then, not long after, a decisive one—at least for a brief time.

Let’s stick with that idea for a moment. First, such a scenario would be welcome news for America, which is likely to be the country whose companies have created that powerful AI capability, given our current lead in compute resources and frontier capability. (Also possibly good news for Taiwan from a deterrence perspective.)

Second, such a scenario would seem to lower the stakes of the current debate over open models. If such a decisive capability does arrive sooner rather than later, it will probably emerge inside a closed frontier lab, one located in the United States. Under that assumption, the very highest geopolitical and national security stakes would remain concentrated at the closed frontier—even though open models would continue to raise important questions, including cybersecurity and the spread of advanced capabilities.

And if you think super-advanced AI isn’t on the horizon—which is the forecast of five out of six AI technologists, according to one recent survey—then we can treat AI models and their corporate creators in a more “normal” fashion. These open and near-frontier models should be treated as engines of competition and diffusion. As tech analyst Ben Thompson argues, the economic panic over Chinese models like Kimi K3 mostly misreads market economics: 

Right now there is a price umbrella that is downstream of the lack of compute; I highly doubt that Chinese models are cheaper to serve on a marginal cost basis, they just seem cheaper because Anthropic and OpenAI are so supply constrained that they are charging far more than they would if there were sufficient supply to meet the demand for intelligence. 

Washington should think hard about how policy currently works against a proliferation of advanced American models—starting with the informal national-security review process that, as tech policy analyst Adam Thierer has explained, can hold up a model’s release with little public explanation or transparent process. That’s a dicey situation for a wealthy frontier AI company—and an impossible one for an open-model startup.

There are genuine security problems here. My AEI colleague Ryan Fedasiuk has usefully catalogued many of them, such as user data flowing to Chinese servers subject to Beijing’s National Intelligence Law and tampered model files in the supply chain. The right response is targeted rules on AI models used in government systems, sensitive data, and what major American platforms distribute. So an argument for safeguards, not suppression.

 As he wrote in April:

To address mounting risks to American data security, some policymakers may be tempted to ban the adoption of Chinese AI models outright — but this would be unenforceable in practice and deeply damaging to the U.S. developer ecosystem. DeepSeek and Qwen are available for download from dozens of mirrors and third-party platforms. API access can be routed through intermediaries, and any prohibition broad enough to cover the full spectrum of Chinese-origin AI services would risk sweeping up thousands of derivative models fine-tuned by developers with no connection to Chinese security services. 

It’s always good to think about first principles. The American System of innovation is a free enterprise system. Our baseline should be to desire a dynamic marketplace led by lots of truly competitive American open and closed models, with access to Chinese models governed by targeted security safeguards rather than an indiscriminate ban. Washington should also be thinking hard about barriers to the development of competitive American open models, as well broader constraints in areas such energy production and data center construction. That’s a reasonable lens through which policymakers should view the issue right now.