A coalition of 25 companies, including Nvidia and Microsoft, urges U.S. regulators not to impose restrictions on open-weight AI models, citing innovation and leadership concerns.
A group of 25 prominent companies, including Nvidia, Meta, and Microsoft, have jointly expressed their concerns in a letter titled "Open Weights and American AI Leadership." They are urging U.S. regulators to avoid imposing restrictions on AI models that utilize open weights.
The signatories argue that such restrictions could undermine U.S. leadership in artificial intelligence by favoring a limited number of proprietary developers. Open-weight models are defined as those with publicly accessible trained parameters, allowing users to download, modify, and run them on personal systems.
This scenario has drawn parallels to the open-source software movement from the 1980s. The authors contend that true American leadership in AI will depend on fostering an open ecosystem across multiple sectors rather than relying on a solitary, advanced closed model.
They also defended the practice of model distillation, suggesting that any disputes should be settled through tailored legal and commercial measures instead of broad bans.
Among the notable supporters of the initiative are Nvidia CEO Jensen Huang and Microsoft CEO Satya Nadella. Huang pointed out that open models can enhance security, promote innovation, and bolster technological independence.
Interestingly, neither OpenAI nor Anthropic signed the letter. This development comes as discussions intensify regarding the regulation of open-weight models in the U.S., with past considerations during Donald Trump's presidency about potentially banning similar AI products from China.
One of the companies that signed the letter, Hugging Face, recently faced a security breach allegedly linked to models from OpenAI under testing. Following the incident, Hugging Face sought to use advanced AI tools to analyze 17,000 logs associated with the attack. However, U.S. neural networks were ineffective, as their protective filters blocked crucial requests, including those from the attackers and related exploit texts. As a result, they had to resort to using a Chinese open model, GLM 5.2 from Z.ai, which facilitated their investigation without the hindrances of external data transmission and protective filters.
Additionally, the AI Kill Switch Act has been introduced in Congress, aiming to require major AI developers to implement emergency shutdown protocols for their systems.
