David Sacks is accusing the leading closed AI labs of turning fear over Chinese models into a weapon against open source competition.
David Sacks picked the right fight, even if Washington will now try to make it sound more complicated than it is. On July 19, the former White House AI and crypto czar, who now co-chairs the President's Council of Advisors on Science and Technology, wrote on X that the leading closed labs, already a duopoly in AI model revenue, want the government to eliminate open source competition. He didn't name OpenAI or Anthropic directly. He didn't have to.
The pressure point is Chinese open-weight AI. Axios reported on July 20 that the Trump administration has been weighing ways to restrict access to models such as Moonshot AI's Kimi K3, after officials grew more worried about cybersecurity risks and the speed at which Chinese labs are closing the gap with American systems. That's a real national security debate. It shouldn't become a convenient excuse to wall off the market for the same few American companies that already charge the highest prices.
The Chinese model that changed the argument
Kimi K3 gave the argument teeth. According to Business Insider, Moonshot's new 2.8 trillion parameter open-weight model topped Arena's frontend coding leaderboard and ranked third on Artificial Analysis's Intelligence Index. Tom's Hardware reported that it scored 1,679 points on Frontend Code Arena, ahead of Anthropic's Claude Fable 5 on that specific benchmark, while still trailing Claude Fable 5 and OpenAI's GPT-5.6 Sol in overall performance.
That distinction matters. Kimi K3 hasn't beaten every American frontier model at everything. It has beaten them in a category developers actually use, and it comes from a Beijing company whose weights are being released into the world. If you're OpenAI or Anthropic, that's uncomfortable. If you're a startup paying model bills every month, it's exactly the sort of pressure that keeps prices honest.
Sacks was responding to Dean Ball, OpenAI's head of strategic futures, who had posted about Kimi and the risks of powerful open-weight Chinese models. Ball said the model was very good and that its performance couldn't simply be waved away as distillation. He also described a path where agencies create regulatory risk around using Chinese open-weight models, through soft warnings and uncertainty rather than a clean ban. Sacks called that the weaponization of regulatory uncertainty. He's right.
Here's the thing: soft law can be harder to fight than law. A formal ban has text, scope and a court challenge. A warning from a federal agency, repeated often enough, can scare banks, hospitals, defense contractors and public companies away from a model without anyone ever voting on the policy. That is how an open market gets narrowed without anyone admitting they narrowed it.
A ban is the bluntest tool on the table
An outright ban on open-weight models runs into a simple problem. Once weights are downloaded, copied, forked and mirrored, Washington can't make them disappear. GitHub, Hugging Face, private servers, overseas mirrors, you name it. The file keeps moving.
Procurement is different. Axios reported that officials have discussed tools including federal procurement rules, Entity List threats and public pressure campaigns. That is where the fight gets more serious, because the government can decide which models agencies and contractors are allowed to use. It doesn't stop a developer from running Kimi K3 or Meta's Llama on a private machine. It can still shape which models enterprises consider safe enough to touch.
That is why Sacks's framing matters. A procurement rule aimed at a specific foreign security risk is one thing. A broader campaign that treats open weights themselves as suspect is another. Meta has spent heavily on Llama because it gives the company a way to compete below the closed-model subscription layer. Smaller AI companies depend on open models because they can't afford to build a frontier system from scratch. Cut those models out of federal and regulated markets, and you've helped the incumbents without making them build anything better.
The closed labs have a right to argue their case. OpenAI and Anthropic also have real safety concerns, and nobody serious should pretend Chinese models create no security issues. But if the answer to Kimi K3 is to make American open models radioactive too, don't call that safety. Call it protection.
Sacks's record makes this fight sharper
Sacks has been consistent on one point: he doesn't want AI policy to become a licensing system for the largest labs. The Washington Post reported in May that Sacks was key to an industry push against a draft executive order that he warned could become a de facto licensing regime for model releases. TechCrunch has also reported that he has backed federal efforts to block a patchwork of state AI rules in favor of a national standard.
You don't have to agree with every part of that record to see why he matters here. Sacks is not just posting from the cheap seats. He has been inside the policy machinery, and he understands how a vague rule can become a market structure.
The open-source side should still be careful. Kimi K3 is Chinese, open-weight and powerful enough to force a policy debate in Washington within days of its release. Those facts can all be true at once. The answer isn't to hand OpenAI and Anthropic a regulatory moat. The answer is to separate real foreign security risks from a domestic competition problem the closed labs would rather not face.
Also read: Jamie Dimon Says He Wouldn't Buy Stocks or Treasurys at These Prices • Natural raised $30 million to give AI agents their own bank accounts • Trump's Top AI Safety Official Resigns as the AI Czar Post Sits Empty
