周二,财政部长 Scott Bessent 表示,美国将审查来自中国的开源模型,以寻找知识产权窃取的迹象,若证实存在知识产权窃取行为,将威胁对中国 AI 公司实施制裁。
“我们听到很多关于开源模型即将到来并威胁美国大语言模型的说法,”Bessent 周二在 Fox Business 上表示。“本届政府支持开源模型,但我们不支持的是知识产权窃取。如果我们看到,尤其是海外模型在窃取我们优秀公司的成果,我们就有能力因这种窃取行为对其实施制裁。”
Bessent 的言论最初由 Bloomberg 报道。
这一表态出台之际,中国模型——最近是月之暗面的 Kimi K3——在能力和受欢迎程度上不断提升,可能威胁到 OpenAI 和 Anthropic 等美国顶尖 AI 公司的商业模式,以及它们筹集更多资金以继续开发前沿模型的能力。
周一,Axios 报道称,特朗普政府正在考虑全面禁止中国的开源模型,不过也有人对这一说法提出异议。
数月来,AI 公司一直在警告外国行为者发起行动,复制其 AI 技术并将其作为开源重新部署。4 月,白宫表示将与 AI 公司密切合作,打击此类窃取行为。
美国对中国模型实施制裁,将进一步扩充政府为维持 AI 竞赛领先地位而采取的策略清单。在限制中国获取先进芯片并收紧出口管制之后,华盛顿如今释放信号,可能将矛头直接对准 AI 模型本身——此举或标志着前沿实验室与中国开源替代方案之间的技术竞争显著升级。
模型蒸馏是一种技术,可将大型模型的部分能力迁移到更易于运行的小型系统中——但并非所有人都认同蒸馏另一家公司的模型构成窃取行为。
本月早些时候,微软 CEO Satya Nadella 批评大型实验室正是做出了这种假设:“虽然模型提供方拥有在公开数据上训练模型的合理使用权所带来的伟大创新是必要的,但我觉得讽刺的是,现状却是反过来对蒸馏施加限制性条款。”
AI 实验室的训练实践持续成为这些公司的法律风险来源。Anthropic 本周获准开始向作者支付赔偿支票,作为其 15 亿美元和解案的一部分,此前一名法官裁定其非法下载并存储了数百万本受版权保护的书籍用于训练其 AI。
此外,业内一些人认为,蒸馏并非中国正在追赶美国 AI 公司的唯一原因。
“我们知道,知识蒸馏在打造优秀模型的能力中只是一个很小的因素,而且这是所有人都在做的做法,包括美国的公司,”Hugging Face CEO Clem Delangue 在最近一期的TechCrunch 的 Equity 播客中表示。“如果仅仅靠知识蒸馏就能轻松擅长构建 AI 模型,那么包括美国在内的许多其他国家早就会有更好的开源 AI。现实是,中国拥有非常非常优秀的研究团队……他们对 AI 采取了比美国更加开放和协作的方式。”
On Tuesday, Treasury Secretary Scott Bessent said the U.S. would examine open source models from China for signs of intellectual property theft, threatening sanctions against Chinese AI companies if IP theft is established.
“We’ve seen a lot of talk about open source models coming and threatening the large language models in the US,” Bessent said on Fox Business Tuesday. “This administration supports open source models, but what we do not support is IP theft. If we see, especially, that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft.”
Bessent’s comments were first reported by Bloomberg.
The statement comes as Chinese models — most recently Moonshot AI’s Kimi K3 — are gaining in capabilities and popularity, threatening to harm the business models of top American AI firms like OpenAI and Anthropic, as well as their abilities to raise more capital to continue developing frontier models.
On Monday, Axios reported that the Trump administration is considering a wholesale ban on Chinese open source models, although others have disputed that claim.
AI companies have been warning for months against campaigns by foreign actors to copy their AI technology and redeploy it as open source. In April, the White House said it would work closely with AI firms to combat the theft.
Sanctions from the U.S. against Chinese models would add to the growing list of strategies the government is attempting to maintain the lead in the AI race. After restricting China’s access to advanced chips and tightening export controls, Washington is now signaling it may target the AI models themselves, a move that could mark a significant escalation in the technological competition between frontier labs and Chinese open-source alternatives.
Model distillation is a technique that allows some of a larger model’s capabilities to be translated into a smaller system that’s easier to run — but not everyone agrees that distilling another company’s model constitutes theft.
Earlier this month, Microsoft CEO Satya Nadella criticized large labs for making just this assumption: “While the great innovation that comes from model providers having fair use rights to train models on public data is needed, I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation.”
AI labs’ training practices continue to be a source of legal risk for the companies. Anthropic this week got the green light to start cutting authors checks as part of its $1.5 billion settlement after a judge ruled it had illegally downloaded and stored millions of copyrighted books to train its AI.
Furthermore, some in the industry argue that distillation isn’t the only reason China is catching up to U.S. AI companies.
“We know distillation to be a very small factor in the ability to create good models, and it’s a practice that everyone is doing, including companies in the U.S.,” Hugging Face CEO Clem Delangue said on a recent episode of TechCrunch’s Equity podcast. “If it were easy just to do distillation to get good at building AI models, there would be many other countries, including in the U.S., with much better open source AI. The reality is they have really, really good research teams in China…taking a much more open and collaborative approach to AI than in the U.S.”