As Washington debates whether to restrict or ban Chinese open-weight AI models, one American company building open-source models for a living has taken an unfashionable position: the models are not dangerous, and trying to ban them is the wrong approach.
Arcee, a US-based open-source AI lab, argues that Chinese open-weight models like Moonshot AI’s Kimi K3 and Alibaba’s Qwen series pose no greater security risk than any other piece of downloadable software. The company’s CTO, Lucas Atkins, is pushing back against the narrative that Chinese models are inherently hazardous, a view gaining traction among US policymakers and proprietary AI labs alike.
“A lot of people view this as similar to a Chinese software program,” Atkins told TechCrunch. “That is fundamentally not how these models are trained.”
The technical distinction Atkins is drawing matters. Open-weight models publish their trained parameters, the learned weights that define the model’s behavior. Unlike a compiled binary from an unknown vendor, these parameters can be inspected, tested for bias and toxicity, and post-trained for specific enterprise use cases. Crucially, once a model is downloaded and run inside an enterprise’s own infrastructure, it has no ability to phone home or execute remote code.
The theoretical risk of a hidden backdoor, a model trained to produce malicious output when triggered by a specific input, is not zero, Atkins acknowledged. But he described the practical difficulty of engineering such an attack as extreme. “I don’t know how you would do this,” he said.
Atkins’ central argument is that the conversation should shift from restriction to competition. “Instead of the conversation being about how to ban Chinese models, it should be about how do we foster a good, open ecosystem here in the US,” he said.
Arcee’s position is not entirely altruistic, the company benefits from the open ecosystem it defends. Chinese open models provide a foundation that Arcee can build on, as Atkins noted: “We can learn what they did. We can build on top of them. Then they can learn what we do.” The ultimate competitive response, he argued, is to “release a model that is better.”
The debate comes at a time when Chinese open-weight models are gaining traction among US enterprises, largely because they offer inference at a fraction of the token cost of proprietary American models. For companies facing pressure to adopt AI while managing costs, the economics are hard to ignore, and harder to replace with a ban.
Sources: Arcee, a US open source AI lab, says Chinese models are not inherently dangerous (TechCrunch, July 22, 2026)