Who gets to decide the future of artificial intelligence?

Hopefully not a handful of progressive tech companies in California, which is why President Trump signed an order last year to keep “woke AI” policies out of the federal government. And certainly not the Chinese government, which forces the AI built under its thumb to parrot the party line and pretend that events like the Tiananmen Square massacre and the ongoing persecution of the Uyghurs simply didn’t happen. Both are versions of the same nightmare: a powerful machine quietly telling hundreds of millions of people how and what to think.

The American people are best served by having plenty of AI options to choose from. We don’t want to live in a future that involves having to choose between woke AI models and Chinese Communist Party AI.

The real danger in artificial intelligence is not some science-fiction machine that wakes up and turns against us, but what President Trump’s AI Czar David Sacks has called “Orwellian AI,” a future in which the most powerful technology ever built is sealed off inside a few enormous companies, and a select few get to decide what the rest of us are permitted to think, say, and build. An open, truth-seeking model is the antidote to Orwellian AI. Americans deserve open-source AI tools that they can download, inspect, verify, improve upon, and even run from their own home.

According to reports and the overall sentiment in the tech world, China is making huge strides in catching up to America’s top-tier AI models. Some of America’s biggest technology companies have made significant announcements in recent days and weeks about integrating Chinese AI models into their systems. Coinbase chief executive Brian Armstrong recently revealed that to rein in a ballooning software bill, his firm had quietly started defaulting its engineers to two AI models built in China, a move that cut its AI spending nearly in half, and he is far from alone. Microsoft, the largest backer of ChatGPT-maker OpenAI, has confirmed it is testing a Chinese model called DeepSeek to run the AI features inside Microsoft 365 simply because the American alternatives have grown too costly to operate at scale. A growing list of U.S. companies, including Airbnb, Uber, and Shopify have done the same.

Now, lawmakers in D.C. are right to feel uneasy about this, but they should resist the impulse to legislate these developments out of existence. What we’re witnessing is that these companies are simply responding to a reality in which Chinese labs now claim to ship AI that performs within striking distance of America’s best at a fraction of the price. One model released just this month, GLM-5.2 from a Beijing company called Zhipu, has climbed to the top of the independent rankings for freely available AI.

All too often, the instinct in Washington is to reach for a ban. It’s understandable, given the stakes of the U.S.-China AI competition, yet it rests on a deep misunderstanding of what we are dealing with, and acting on it would damage America more than it would slow down China.

Perhaps the loudest voice against open models belongs to Dario Amodei, chief executive of Anthropic, whose entire business depends on keeping its own Claude models tightly closed, and who recently told lawmakers that open-source AI is on a “very dangerous path,” warning that once a powerful model is released no company can monitor how it is used, revoke anyone’s access, or push out safety patches after the fact. Elsewhere, he has waved the movement away as a kind of distraction, insisting that the build-on-each-other’s-work advantages that made open-source software so transformative somehow fail to carry into AI. We could not disagree more, because nearly everything that worries Amodei about open source is precisely the reason it is better for all of us, and precisely why America should be sprinting to lead it rather than attempting to legislate it away.

The Chinese models of concern are what the industry calls “open-weight.” Chinese labs do not release their proprietary training data, data processing pipelines, or full training code. However, the company usually gives the AI away for free, so anyone can download it and run it entirely on their own computers. It is true that when you use the DeepSeek phone app, your data travels back to China, which is exactly why it has been thrown off government phones, and sensibly so, but when responsible American companies download the model and run it on their own servers, the data never leaves the building.

The complaint that no one can revoke access to an open model, far from being a flaw, is the whole appeal. When a model is open, it can be forked, which means that anyone may take a copy and build a better version on top of it, much as a cook takes a proven recipe and improves it. So instead of one company building behind a locked door, thousands of independent software engineers in America can refine and harden the same model in parallel, with improvements voluntarily flowing back into shared pools of knowledge. Openness also delivers genuine transparency, because when the model is yours to inspect, you can see exactly what sits inside it, strip out whatever foreign bias (as is the case with the Chinese models) may be baked in, and turn your best and brightest loose to hunt the very security flaws the cyber-hawks rightly fear. A sealed model offers no such window and asks only that you trust the company that built it.

Attempting to ban our way out of this issue would not make a single Chinese model disappear, given that they already sit on laptops all over the planet, so the only thing such a law could achieve is to criminalize Americans’ use of the cheapest, fastest-improving AI on earth while our competitors use it freely. And if the only AI models that Americans may legally touch come from a tiny handful of enormous, closed companies, then that cartel gets to decide what this technology may and may not do for the rest of us.

It is telling that even Microsoft’s own chief executive, Satya Nadella, recently warned in an essay that “a frontier without an ecosystem is not stable,” cautioning that letting a small number of players dominate artificial intelligence is dangerous for the economy and for everyone living inside it, and when one of the giants is the one sounding the alarm about concentration, the rest of us would be foolish not to listen.

The answer to China’s AI progress is not a ban, and it is certainly not handing the keys to a closed cartel of Silicon Valley liberals, but to compete in earnest and win the open game rather than forfeit it via legislation. Great American companies invented this technology, and we should be the country that builds the finest open models on earth. Banning innovation has never made a nation stronger, yet out-innovating the competition has done so again and again. When America allows for competition and an even playing field, it unleashes the American spirit and our finest entrepreneurs’ restless drive to innovate.