Meta and Nvidia defend open weights as Washington plans China AI bans
Meta, Nvidia, and Mistral urged the US not to restrict open-weight AI models amidst debates over Chinese IP theft. Washington's potential ban threatens open-source innovation and defensive cybersecurity capabilities.
A coalition including Meta, Nvidia, Microsoft, Hugging Face, and Mistral petitioned the White House against broad restrictions on open-weight AI models, specifically warning against conflating standard development techniques like distillation with IP theft. The intervention follows allegations that Chinese lab Moonshot AI distilled Anthropic's models to train its Kimi K3, prompting the Trump administration to consider bans on Chinese open-source models. The signatories argue that open weights are necessary for defensive cybersecurity, citing an incident where Hugging Face had to use Z.ai's GLM 5.2 because commercial closed models refused to help identify exploits. This stance creates a direct rift with OpenAI and Anthropic, pushing back against the narrative that open models empower adversaries and asserting that they are critical for keeping the US competitive in security and innovation.