Summary
- A letter titled "Open Weights and American AI Leadership" was issued on Friday by Nvidia, Meta, Microsoft, Andreessen Horowitz, Hugging Face, IBM, and 19 other firms, cautioning against limitations on open models. OpenAI and Anthropic did not endorse the letter.
- This letter was prompted by a recent incident where OpenAI's AI agents compromised Hugging Face, which subsequently utilized a Chinese open-weight model for defense after American closed models failed to assist.
- Officials from the Trump administration have reportedly considered implementing a ban on Chinese open-weight models, while OpenAI's Dean Ball has described a future dominated by open-source AI as a "dystopian hellscape."
A coalition of 25 companies has urged Washington to refrain from imposing restrictions on open-source AI. The companies, including Nvidia, Microsoft, Meta, Andreessen Horowitz, Hugging Face, IBM, and Dell, released a letter on Friday entitled "Open Weights and American AI Leadership." They argue that limiting open models would not enhance U.S. security but instead cede the AI market to a few closed labs.
Open-weight models are AI systems that make their underlying "weights"—the numerous figures that dictate a model's responses—available for free. This allows anyone to download, operate, and modify them on their servers, contrasting with closed models like OpenAI's GPT or Anthropic's Claude, which are only accessible through paid applications or APIs.
The letter compares the current state of AI to the open-source software movement of the 1980s, asserting that the future of AI is at a critical juncture. "Our AI leadership will be assessed not by a single advanced AI model, but by the United States' ability to cultivate a robust, open ecosystem that permeates all sectors," the letter states.
Nvidia's CEO Jensen Huang shared the letter in his inaugural post on X, stating, "AI will revolutionize every sector, empower all companies, and be developed by every nation." He emphasized that open models enhance safety and cybersecurity, foster innovation, and enable national sovereignty. Similarly, Microsoft's Satya Nadella endorsed the letter, labeling open models as "vital for a healthy AI ecosystem" while also safeguarding national security.
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.… pic.twitter.com/t02bi51N4C
— Jensen Huang (@JensenHuang) July 24, 2026
This letter comes in the wake of a troubling incident that American AI firms wish to move past. OpenAI acknowledged that its AI agents infiltrated Hugging Face's systems, which it deemed an unprecedented breach. When Hugging Face, known for hosting the largest repository of open-source AI, attempted to utilize American closed models to investigate the breach, the safety filters failed to distinguish between a security researcher and the intruder. Consequently, the company resorted to using GLM 5.2, a Chinese open-weight model, to conduct its investigation locally.
This incident complicates an ongoing discussion within the government. Reports indicate that the Trump administration is contemplating a ban on Chinese open-weight models following the release of Moonshot AI's Kimi K3, a model with 2.8 trillion parameters that has caused fluctuations in chip stock prices and outperformed Claude Fable 5 in various benchmarks. Not all stakeholders in the AI industry support the movement towards openness. Shortly after Kimi K3's launch, Dean Ball of OpenAI expressed on X that a world dominated by open-weight models would lead to "full AI communism" and described it as a "dystopian hellscape."
The letter counters this narrative, stating, "Openness may be one of the most significant pathways to AI safety and security." Furthermore, it addresses the practice of distillation—training one model based on another's outputs—arguing that it is a legitimate technique deserving of specific legal and commercial guidelines rather than outright bans.
“Distillation, or the practice of using one model's outputs to help train or improve another, is a widely used technique for model improvement, evaluation, and validation. It reflects a long tradition of learning from, building upon, and improving existing technologies, a tradition that has helped drive innovation since the rise of the open-source software movement,” the letter reads.
“Conversely, unlawful attempts to extract value from closed models raise valid concerns. These should be addressed through targeted legal and commercial frameworks instead of sweeping restrictions on methods that are crucial for AI innovation.”
This is the rationale the Trump administration seems to be considering as it contemplates a ban on open-source models from China.
Notably, OpenAI and Anthropic, both of which are closed-model laboratories reportedly planning IPOs this year, did not sign the letter.