The biggest names in American tech are drawing a line in the sand over open-source AI, and it’s not where you might expect. On July 24, a coalition of 25 companies including Nvidia, Microsoft, and Meta published an open letter titled “Open Weights and American AI Leadership,” pushing back against what they describe as a coordinated effort by certain frontier labs to restrict open-weight AI models in the US.
The letter’s core argument is blunt: banning open-weight models wouldn’t make America safer. It would make America slower.
The coalition’s case against restrictions
The signatories argue that restricting open-weight AI models would effectively hand a competitive gift to closed-model labs while doing nothing to address the real competitive threat, which is the proliferation of Chinese AI systems that are already gaining traction globally.
Chinese open-weight models like Moonshot AI’s Kimi K3, a 2.8 trillion-parameter system, have been benchmarking competitively against top US closed models.
OpenAI initially declined to sign, which raised eyebrows given CEO Sam Altman’s occasional gestures toward openness. Under rising public scrutiny, OpenAI eventually joined the coalition.
Anthropic held firm in not signing. The company said it opposes an outright ban on open-weight models but prefers targeted actions against intellectual property theft rather than broad industry pledges.
What’s actually happening in Washington
The Trump administration has considered targeted sanctions and export controls aimed at specific Chinese AI systems, with Moonshot AI’s Kimi K3 reportedly drawing particular attention due to intellectual property concerns, but has ruled out a comprehensive ban on open-weight models.
In June, Anthropic received US export-control orders limiting international access to its Fable 5 and Mythos 5 models.
The backdrop to all of this is the release of competitive Chinese models like GLM-5.2, which have highlighted moments where closed US models showed clear limitations.
The open-weight debate cuts deeper than policy
This fight isn’t really about safety, though that’s how it’s often framed. It’s about business models. Closed-model labs generate revenue by controlling access to their systems. Open-weight models undercut that by letting anyone download, modify, and deploy AI without paying licensing fees.
OpenAI’s Dean Ball reportedly suggested creating uncertainty around open-weight models as a strategy, though he later walked back some of those comments.
Meta has a particularly strong interest here. The company has invested heavily in its open-weight Llama model family and has built significant developer goodwill by releasing capable models that anyone can use.