Mark Zuckerberg has come out against one of Washington’s favourite ideas about artificial intelligence: that the way to beat China is to shut its models out.
In an interview with the Financial Times published this week, Meta’s chief executive said banning advanced Chinese AI from use in the United States is “not an effective solution.”
His case is competitive rather than sentimental. Instead of building a wall, Zuckerberg argued, the US should work out where it is actually behind and fix those bottlenecks, a position that puts him alongside the startups urging Trump not to ban Chinese open-weight models.
He was not only talking about China. In the same breath he took aim at rivals closer to home, implicitly criticising OpenAI and Anthropic for pressing government to restrict powerful “frontier” models, an approach he cast as concentrating power in too few hands.
Openness, in his telling, is the natural direction of travel. “The big trend in the industry’s development has always been towards greater openness, putting technological power in the hands of more people, not fewer,” he told the paper.
There is an obvious tension in a Meta executive making this case. The company built its AI reputation on the open-weight Llama models, yet its newest flagship, Muse Spark, arrived closed source, a shift critics were quick to point out.
The timing is not accidental. His comments land in the middle of a live fight in Washington over whether to sanction or restrict Chinese AI, one in which Nvidia’s Jensen Huang has pushed back hard against the hawks.
The industry has split into camps. Meta, Microsoft, Nvidia, and Elon Musk have lined up behind open weights, while OpenAI and Anthropic have argued for tighter controls on the most capable systems, on safety and national-security grounds.
That dividing line is partly about risk. OpenAI and Anthropic contend that the most capable models carry real dangers, from cyber-attacks to bioweapon uplift, and that unrestricted diffusion, including of Chinese open models, hands those capabilities to anyone who wants them.
Zuckerberg’s side sees the danger differently. In their view the greater threat is that a few American labs, and the government they lobby, end up controlling a technology that ought to be widely distributed, slowing innovation in the name of protecting it.
The backdrop to all of this is Chinese progress. A cheap Chinese model has been closing in on the American leaders, and the release of Moonshot AI’s Kimi K3 in mid-July sharpened the question of what, if anything, the US should do about it.
That question produced a letter. On 24 July, twenty-five organisations signed a document titled “Open Weights and American AI Leadership,” arguing that openness, not restriction, is the surer route to continued US primacy.
Zuckerberg’s own vision reaches past the immediate fight. He has described a future of “personalised superintelligence,” a system tailored to each individual rather than a single, centrally controlled model built to be safe on everyone’s behalf.
It is a framing that flatters Meta’s strategy as neatly as it challenges its rivals’. If the future is personal and open, the firms building closed, centralised systems are on the wrong side of history, and the ones giving models away are on the right one.
Washington does not have to accept that frame. Officials weighing export controls and model restrictions are balancing competitiveness against security, and Zuckerberg has now planted Meta firmly on the competitiveness side of the ledger.
What he has not resolved is the hardest part. Whether openness makes American AI stronger or merely easier to copy is precisely the judgement Washington is struggling to make, and one pointed interview will not make it for them.
