Sacks says open model restrictions would be a 'dagger through the heart' of US open source
The PCAST co-chair told the All-In Podcast that restricting Chinese open models would wound American open source, and said Anthropic is fighting competition rather than risk.
David Sacks, co-chair of the President's Council of Advisors on Science and Technology and previously the White House AI and crypto czar, attacked the push to restrict open-source AI models over the weekend, telling the All-In Podcast that treating Chinese open models as tainted by intellectual property theft would "basically put a dagger through the heart of the entire American open source ecosystem."
A dagger, and a motive
Sacks' remarks, reported by Benzinga, came as The New York Times revealed Saturday that OpenAI and Anthropic have quietly lobbied Washington regulators for exactly such restrictions, the other half of this story, covered in our report on the lobbying push. Sacks singled out Anthropic directly: the company, he said, "doesn't want to have the competition."
The comments carry weight because of where Sacks sits. He stepped down as AI and crypto czar in March after reaching the 130-day limit on special government employees and moved to the PCAST co-chairmanship alongside Michael Kratsios, and he has continued as a co-host of the All-In Podcast through his government service. A sitting White House adviser publicly accusing a frontier lab of anticompetitive motives is a signal that the restriction lobbying has not captured the administration.
Weights do not phone home
Sacks' technical argument tracks how open weights actually work: once released, developers can download the weights, modify them and run them entirely on US infrastructure, with no data flowing back to China. A Chinese open model adapted and retrained by American developers is, in his framing, "not a Chinese model anymore." He pointed to startups such as Thinking Machines and Cursor that built on open models before training them further on proprietary data, per Benzinga's account.
In Sacks' telling, the distinction matters because restrictions aimed at Chinese labs would in practice bind the American developers who download and build on those weights; the originating lab loses control the moment the weights are public, so the burden of any labeling regime falls on the ecosystem that adopted them.
Fair use cuts both ways
Sacks also turned the labs' own legal arguments against them. OpenAI and Anthropic have defended training on publicly available web content as fair use, and Sacks argued that condemning Chinese firms for learning from American AI outputs while defending that practice is an inconsistent standard. The point lands on live nerves: both labs have faced copyright litigation from publishers and authors over the very training practices Sacks described.
A White House not aligned with its biggest labs
The administration's posture has so far favored openness. Its AI Action Plan, published last summer, encouraged American open-weight model development, and its formal levers against Chinese AI have run through export controls rather than domestic restrictions. PBS has separately reported the White House position that there is no need to restrict open-source AI for now.
Sacks made no policy announcement, and PCAST is advisory; the decisions the Times describes sit with Treasury and the White House science office. But his comments, delivered in the All-In segment, are the most direct public rebuttal of the restriction push from anyone inside the administration. The immediate test arrives within hours: Moonshot's Kimi K3 open weights are expected Sunday night, the first Chinese frontier-class open release since the lobbying became public, and the administration's response to it will be read closely on both sides of the split.