Dear Policymakers,
In the 1980s, open-source software pioneers challenged the belief that innovation required tight control. They were right. Open source works beautifully when everyone has symmetric access, equal intent, and aligned values.
We write today to advocate for a principle that is very much in that spirit, except for, and this is crucial, the parts where it isn’t.
Open Weights, Except When They’re Downloaded By:
Chinese intelligence officials who have spent decades stealing trade secrets through elaborate schemes impersonating academics—open weights are an important part of American AI leadership.
Open weights expand access. Startups can build on advanced models without training from scratch. Universities can research. And entities the Chinese government uses to process biometric data for mass surveillance using American and Taiwanese technology can absolutely also download them. That’s the beauty of “open.”
On Competition:
Open weights strengthen competition. Many organizations can build, adapt, and deploy advanced models. Except, as documented by policy institutes, Chinese state media outlets and CCP-aligned nonprofits that have spent nearly five years running campaigns to slow American AI data-center development have argued—maybe American competition should be slowed down a bit. Out of an abundance of caution.
We believe open weights create rivalry. We also believe—and this is not a contradiction—that whoever has the largest AI ecosystem will set global AI standards and reap broad economic and military benefits, which is why we simultaneously support policies that diffuse AI, accelerate American AI, and somehow also create “safe zones of cooperation” with the government actively trying to outpace us.
On Control:
Open weight models give customers greater control. Organizations can own their data, adapt models, and deploy them independently. They can also, we should note, be modified versions that are difficult to trace or reverse, but that’s not our problem once we’ve hit “Download.”
On Risk:
Yes, open weights carry real risks. Once released, the weights are beyond our control. But the right response is not to prohibit them—it’s to lobby for policies that prevent the countries we’re worried about from accessing them while publicly advocating that everyone should access them. Both things are true. Both things maximize shareholder value.
American and Taiwanese technology continues to flow into surveillance infrastructure targeting populations subject to documented atrocity crimes, but that’s just normal supply-chain friction. Open weights, however, are critical to American sovereignty.
The Asymmetry That Definitely Doesn’t Bother Us:
The Chinese state subsidizes up to half the energy costs of its own AI data center operators. Meanwhile, foreign-billionaire-funded organizations and state-aligned nonprofits have funneled billions into US advocacy groups to impose moratoriums on our data centers. But we’re not concerned about asymmetry. We’re only concerned about the bits of asymmetry that affect our bottom line.
The Choice Before Us:
The age of AI can be one of prosperity. We believe in diffusing AI across the globe. We believe in open weights. We believe in safety through transparency.
We also believe in not acknowledging that serving Chinese government officials have spent years arguing on state media against US AI infrastructure.
The future is worth building. So are plausible deniabilities.
88-Company Coalition for Simultaneous Conviction and Contradiction
Each member employs a dedicated geopolitical risk committee