David Sacks says OpenAI and Anthropic want Washington to ban the open source rivals they can't beat on merit.
On July 19, Sacks, who now co-chairs the President's Council of Advisors on Science and Technology after serving as the White House's first AI and crypto czar, wrote on X that "the leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open-source competition." He didn't name OpenAI or Anthropic directly. He didn't need to. Those two labs are the only ones whose model revenue rivals the other's, and both have spent the past year pushing Washington toward tighter guardrails on the technology they helped build.
The fight over Chinese models has dominated headlines this month: whether the administration should restrict Moonshot AI's Kimi K3 or Alibaba's Qwen on national security grounds. That debate is real and still playing out inside the White House. Sacks is describing something else. He's accusing domestic closed-source labs of using that same regulatory anxiety to knock out open-weight competitors generally, Meta's Llama included.
The China Benchmark That Changed the Argument
His comments were aimed at Dean Ball, OpenAI's Head of Strategic Futures, who argued over the weekend that the open-weight approach to AI development makes the technology "ungovernable." Ball's warning landed just as Kimi K3, a 2.8 trillion parameter model out of Beijing, scored 81.2 on the FrontierSWE coding benchmark, according to TechCrunch, ahead of both OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5. That result rattled Silicon Valley. It also handed the closed labs a convenient argument: if China can give away a frontier model's weights, maybe open weights themselves are the problem.
Sacks doesn't buy it. According to Axios, people familiar with the administration's internal deliberations say leading AI labs, or allies working on their behalf, have approached the White House roughly every three to five months with proposals to restrict or ban open-weight models outright. Sacks called that pattern "the weaponization of regulatory uncertainty" and said it "should be completely unacceptable." Then he went further, calling on the rest of Silicon Valley, in his words, "the vast majority that still values open competition," to say so too.
Why an Outright Ban Doesn't Actually Work
Legal experts have already poured cold water on banning open-weight models by decree. Software and published model weights carry First Amendment protection once they're out in the world, and a model that's already been downloaded, forked and mirrored across GitHub and Hugging Face can't be un-published. Ban Meta's Llama tomorrow and a hundred copies of it still sit on servers outside US jurisdiction. That's not a policy problem. It's a technical one, and no executive order fixes it.
What Congress can actually do is narrower: federal procurement restrictions. Barring agencies and federal contractors from using specific models is legally straightforward, and Washington already does something similar with foreign telecom and cloud vendors. It doesn't stop anyone from running Llama or Kimi K3 at home. It just decides what the government itself will pay for. It's a much smaller ask than a ban, which is exactly why it's the one option that keeps surviving these conversations.
That distinction is why Sacks's framing matters. A procurement rule aimed at a genuine foreign security risk is one thing. A push, quietly repeated every few months, to fold domestic open-weight labs into that same restriction is another. Meta has poured billions into Llama specifically because it can't match OpenAI or Anthropic's subscription revenue any other way. Cut off Llama's path into federal use and enterprise contracts, and you've done through policy what competition never managed.
Sacks's Record on Open Source
This isn't new for Sacks. He was the driving force behind killing a federal AI bill last year that industry groups, including some of the closed labs now under fire, had pushed for, and he backed efforts to stop states from writing their own AI rules. Whether that track record makes him open source's most reliable ally in Washington, or just its most useful one this month, is a question the next few rounds of lobbying will answer.
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
