America Is Playing AI Blackjack with the CCP — and at Risk of Losing

E very professional blackjack player knows that the key to winning isn’t just in the hand you’re playing — it’s in the cards you’ve already seen. If a dealer shows a ten — and you know there’s a disproportionate number of high cards left in the deck, because you’ve been counting — you will quickly realize that your chances of losing just went up. And when it comes to artificial intelligence, the Chinese Communist Party just showed us a ten.

DeepSeek, a Chinese AI company, just released its latest model: R1. It’s a frontier AI model that matches or exceeds the capabilities of leading Western models across numerous benchmarks. The past year has also seen a rapid succession of increasingly sophisticated Chinese AI models: Baichuan, Yi, and now R1. Each release shows significantly greater capabilities than the last. While the full extent of R1’s capabilities is still being independently verified, its release not only reminds us that China wants to surpass the United States in artificial intelligence, but that it’s increasingly likely to do so.

The release of R1 represents more than just another entry in the increasingly crowded field of large language models. It’s a wake-up call that should fundamentally reshape how we think about AI development and regulation in the United States. While American policymakers and AI ethicists engage in circular debates about hypothetical risks from U.S.-based frontier models, Chinese companies are rapidly advancing their capabilities — and those are just the advances they’re choosing to show us.

The analogy of playing blackjack against the CCP is particularly apt. When facing a dealer’s ten, you must play as if he has another ten underneath. The cost of being wrong in the other direction — assuming the dealer has a bad hand when he might have 20 — is simply too high. In our high-stakes game with the CCP, DeepSeek’s R1 represents such a significant technological achievement, that we must assume the CCP’s unreleased work — its hole card — is equally impressive.

This isn’t paranoia; it’s probability. Just as a card counter doesn’t need to see every card to know the odds have shifted, we don’t need to see inside every Chinese AI lab to understand the trajectory. Each public release is like a card played face up — and the parade of increasingly sophisticated models tells us exactly what kind of deck we’re playing against. These visible capabilities — the ten on the table and the trail of high cards that preceded it — hint at what might lie beneath the CCP’s broader research program. And while we deliberate on our response, the landscape of AI regulation in the United States is becoming increasingly fragmented and chaotic.

State legislatures across the country are rushing to implement their own AI regulations, creating a patchwork of inconsistent and often technically uninformed policies. California, Texas, and others are proposing everything from mandatory AI audits to strict liability frameworks for AI developers. This state-level regulatory chaos threatens to create an impossible operating environment for U.S. AI companies precisely when we need to be accelerating development.

The United States finds itself in an increasingly precarious position. While we’ve focused on developing elaborate frameworks for AI safety and engaging in philosophical debates about artificial consciousness, Chinese companies have focused all their energy on systematically building and deploying increasingly sophisticated models. The R1 release suggests we may have fundamentally miscalculated our lead in this field.

Our current regulatory approach resembles a player who’s so focused on memorizing basic strategy that he has forgotten to count cards. While we meticulously debate each theoretical scenario and perfect our defensive posture, we’re missing the bigger picture — the count is skyrocketing, and we’re still playing like it’s a fresh deck. Casual players — and casual policymakers — might think each hand exists in isolation. But professional players know that context is everything. The cards that have already been played tell you as much about your odds as the cards currently showing.

The focus on hypothetical future risks from U.S.-based models has created a regulatory environment that could kneecap American AI development at precisely the moment when every indicator — every card we’ve seen from China’s deck — tells us to accelerate. We’re acting like recreational players at a high-stakes table, obsessing over the perfect play while ignoring the count that tells us we need to increase our bets dramatically.

The path forward must be bold and clear. Just as a professional blackjack player increases his bets when the count turns favorable, we must dramatically scale up our investments and commitment when the technological stakes are highest. President Trump’s recent decision to rescind President Biden’s Executive Order on AI represents an important recognition of this reality — a clear signal that at least some of our leaders understand we can’t win by playing defensively.

But this is just the beginning of what needs to be a comprehensive strategic shift. Congress needs to act decisively to preempt the continued regulatory fragmentation. We need federal legislation that explicitly stops state governments from impeding progress. And the time for academic debates about hypothetical risks is over. We see the cards. The only question now is how we play our hand. Will we continue to hamstring our own AI development with overcautious regulation and fragmented state laws, and let the Chinese Communist Party take over as the world’s AI leader?

The choice is ours, but the clock is ticking. In blackjack, you only have a few seconds to decide how to play your hand. In the AI race, our window is almost as small. DeepSeek’s R1 is the ten on the table, and every indicator — every card we’ve seen come out of China’s AI programs — tells us we’d be fools to play this hand casually. We’d be even bigger fools to keep debating the rules while our opponents are playing to win.