Mark Zuckerberg has come out against US bans on Chinese AI models, and the argument is a strong one: openness, not walls, is how America stays ahead. “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.” It’s a principled position — delivered by a company that just made its own flagship model closed-source, which is the part worth examining.

What he said

In a Financial Times interview, Zuckerberg called restrictions on Chinese AI models “not an effective solution,” arguing the US should identify specific competitive gaps and close them directly rather than erect barriers. It lands in an active Washington debate: on 24 July, 25 organisations signed a letter titled “Open Weights and American AI Leadership,” and the industry has split cleanly. Meta, Microsoft, Nvidia and Elon Musk back open weights; OpenAI and Anthropic advocate stricter controls.

The models in question are real competitors — Moonshot’s Kimi K3, released mid-July and noted as closing the gap with American frontier models, is exactly the kind of capable Chinese open-weight model the debate is about.

The case for openness is genuinely good

Take the argument on its merits first, because it deserves it. Banning access to Chinese models is hard to enforce (open weights, once released, spread and can be run anywhere), and it may be counterproductive: it cuts American developers off from studying and building on capable models while doing little to slow China. The historical pattern in computing does favour openness — open platforms have repeatedly out-innovated closed ones by putting tools in more hands. And there’s a real argument that trying to win by restriction rather than by building better is a tell of weakness, not strength.

Zuckerberg’s implicit criticism of OpenAI and Anthropic — that seeking government restrictions concentrates power dangerously — has force too. Incumbents asking the government to wall off competition is a very old move, and skepticism of it is warranted.

The inconvenient contradiction

Here’s what makes this more than a policy statement. Meta built its entire AI reputation on open-source Llama — the open-weight model that became the foundation for much of the ecosystem. But its newer flagship, Muse Spark, is closed-source. The company preaching openness in Washington released its best model proprietary.

That doesn’t make the openness argument wrong — good arguments survive hypocritical messengers. But it does reframe what’s being argued. “Openness is good” from a company that keeps its best work closed starts to look less like a principle and more like a competitive position: open models commoditise the layer where Meta competes with OpenAI and Anthropic, while Meta keeps its own crown jewels closed. Openness for the base layer, proprietary for the frontier, is a strategy that happens to be dressed as a value.

This is the pattern to watch across the whole open-weights debate: nearly every participant’s stated principle aligns suspiciously well with its commercial interest. Companies that lead on closed models want controls; companies that compete by commoditising the base want openness. The arguments are worth judging on their logic, precisely because you can’t judge them on the sincerity of the people making them.

The read

Zuckerberg is very likely right on the policy: banning Chinese open-weight models is hard to enforce, cuts American builders off from capable tools, and substitutes restriction for the harder work of building better. The openness argument is sound on its own terms, and the concern about incumbents lobbying for protective walls is legitimate.

Just read it with the Muse Spark asterisk firmly attached. This is a debate where every major voice is arguing for the rules that favour its own model strategy, and Meta preaching openness while shipping its flagship closed is the cleanest example of the genre. Weigh the argument, not the messenger — and notice that in AI right now, “what’s good for openness” and “what’s good for my company” are being described in identical words by almost everyone.

Reporting on public statements as covered on 29 July 2026, based on a Financial Times interview. Quotes are as reported. Commentary, not investment advice.