Rival companies rarely sign the same document. 25 companies and organizations, including Nvidia, Microsoft, Meta, IBM, Dell, Palantir, Hugging Face, Mistral, Andreessen Horowitz, and Y Combinator, put their names on a joint letter titled “Open Weights and American AI Leadership.” Addressed to US lawmakers, the letter makes one demand: don’t restrict open-weight AI models through new regulation.
The letter was hosted on Nvidia’s servers and promoted by CEO Jensen Huang in his first-ever post on X (formerly Twitter). Microsoft CEO Satya Nadella reshared it, framing it as a national security issue. Elon Musk voiced support for the idea, though SpaceX didn’t formally sign. Within days of publication, the list of signatories grew several times over.
The timing wasn’t random. A day earlier, more than 200 startups had sent the White House the same request. The letter also landed in the middle of a sharpening US debate over Chinese open models: White House official Michael Kratsios had accused China’s Moonshot AI of building its Kimi K3 model by distilling Anthropic’s Fable 5, and Treasury Secretary Scott Bessent floated sanctions, comparing the practice to theft.
The letter’s authors push back on conflating distillation, the practice of using one model’s outputs to help train another, with unlawful misappropriation. They argue distillation is a widely used, legitimate technique, and that unlawful extraction of value from closed models should be addressed through targeted legal frameworks rather than blanket restrictions on open models.
The conventional assumption is that open equals risky. The letter turns that logic around: closed models can also be breached or fail, but outside researchers have no way to see or verify it. Open models, by contrast, get continuously red-teamed by thousands of independent teams, so vulnerabilities surface and get patched in the open. Leaning on a handful of closed models, the letter argues, creates single points of failure, since no one outside the provider can confirm when something breaks or misbehaves.
Three major names are missing from the list: OpenAI, Anthropic, and Google. All three currently lead the closed-model market, and tighter regulation on open models would reinforce their competitive position. Their absence is being read less as a coincidence and more as a clear conflict of interest.
Open-weight models lower the barrier to entry for startups, no need to train a model from scratch or pay per API call for every task. For any team building AI in a local language, whether that’s an Uzbek-language chatbot or a local document-processing system, open models are often the only realistic entry point, since they can be fine-tuned on local data, hosted on local infrastructure, and kept under full control. If the US moves toward tighter restrictions on open models, that could slow the growth of the open ecosystem globally, which in turn could strengthen the position of Chinese alternatives.
