Nvidia, Meta, Microsoft, and 22 other companies have published the Open Weights and American AI Leadership letter, urging US authorities not to impose restrictions on open-weight AI models.
The authors warn that bans will not strengthen US leadership but rather shift the market towards a few closed developers. Open-weight models are systems whose trained parameters (weights) are publicly available. Such solutions can be downloaded for free, run on one’s own infrastructure, and modified.
The situation was compared to the open-source software movement of the 1980s. According to the signatories, American leadership in AI will depend not on a single advanced closed model but on the country’s ability to build an open ecosystem for various industries.
The authors also defended distillation, calling it a legitimate practice. They suggested addressing contentious cases through targeted legal and commercial frameworks rather than a blanket ban.
The letter was publicly supported by Nvidia CEO Jensen Huang and Microsoft CEO Satya Nadella.
Huang described open models as a factor that enhances safety and cybersecurity, accelerates innovation and technology diffusion, and supports technological sovereignty.
OpenAI and Anthropic were not among the signatories.
Decrypt noted that the letter emerged amid an intensified debate on regulating open-weight models in the US. The publication claims that the Trump administration discussed the possibility of banning such Chinese AI products.
One of the companies supporting the letter was Hugging Face, whose AI platform was recently hacked, as it turned out, by models undergoing testing from OpenAI.
Hugging Face specialists attempted to use advanced commercial AI solutions to analyze 17,000 attack log entries. American neural networks failed as their protective filters blocked requests: the attacker’s commands, exploit texts, and elements of malicious infrastructure. The systems could not distinguish the investigation from an attempt to seek hacking assistance.
The company had to deploy the open Chinese model GLM 5.2 from Z.ai internally—without sending data externally and without protective filters that hindered analysis.
Earlier, the US Congress introduced the AI Kill Switch Act, requiring major developers to implement emergency shutdown mechanisms for AI systems.
