Full story
Safety debate over open-weight
Chinese open-weight AI models have sparked a renewed debate in the United States over whether they pose a security risk to enterprises and whether the Trump administration should ban them, even as Arcee CTO Lucas Atkins argues the fear is largely unfounded.
“As Chinese open-weight artificial intelligence models grow in capability and global adoption, a fresh wave of debate has emerged over whether they pose a security risk to enterprises — and whether the U”
Atkins said open-weight models are no more dangerous than other open-source software a company might integrate, explaining that “There is really not any way for an Arcee, or an Alibaba, to make a model, have someone run it in their own environment and for us to have any access to it whatsoever,”.
He also said enterprises can download models from platforms like Hugging Face, run security testing, inspect visible source code, and post-train for specific uses before deployment.
Atkins acknowledged that a coding-focused model could theoretically be trained to insert backdoors into generated code, but said “I don’t know how you would do this.”
The dispute centers on how open-weight models expose publicly available parameters while keeping training data and methods private, leaving the question of risk to enterprise review and testing rather than remote control by the original developer.
Sanctions, distillation, and watermarks
While Atkins frames the security question around enterprise controls, U.S. Treasury Secretary Scott Bessent said the Trump administration will look into whether Chinese artificial intelligence models have been distilled from American models.
Bessent told Fox Business’ “Mornings with Maria” that “If we see, especially that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft,”.
He said the technical term for the alleged theft is distillation, where “a smaller, less capable model is built using outputs from an existing, stronger model,” and he pointed to Anthropic’s allegation that Alibaba carried out “the largest known distillation attack” against it to date.
Bessent also said “We are finding watermarks of our U.S. large language models on many of the Chinese models, and that's unacceptable,” and said the government would be looking “in the coming days or weeks.”
The CNBC account places the issue alongside planned U.S.-China talks around AI in September, with Bessent representing the U.S. during the discussions, according to a Reuters report.
Business models and competitive stakes
Beyond safety and alleged distillation, the sources also describe how “open weight” differs from open-source software in business terms, with Business Insider arguing that open-weight AI is “not the same thing as open-source software.”
“The angst over China's latest AI models is missing an important business fact: "open weight" AI is not the same thing as open-source software”
The Business Insider analysis says open-weight models are run through inference, where “Every answer requires expensive chips, electricity, and data-center capacity,” and it argues that the next unit of intelligence is not free.
It describes how open-weight labs give outsiders trained numerical parameters so others can download and run models, while revenue most reliably comes from hosting and selling inference compute rather than distributing the model itself.
In parallel, Arcee’s position is that the U.S. should not focus on bans but on fostering a competitive domestic open-source AI ecosystem, with Atkins saying “I think 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 U.S.,”.
Across the accounts, the stakes are framed as both enterprise security practice—through inspection, post-training, and testing—and the competitive pressure on U.S. proprietary labs, with the debate shaped by whether restrictions or open ecosystem building better address the risks and incentives described by the sources.
