POLICY
AI’s biggest rivals are defending model access
As allegations of Chinese labs stealing American AI models pile up, Washington is considering restrictions. You’d assume Silicon Valley would back this move, right?
Not quite.
A growing number of industry leaders are signing an open letter urging the U.S. government not to ban open-weight AI models. The letter launched with 25 initial signatories. This included major players across AI, investing, and infrastructure like Meta, Y Combinator, and Nvidia.
Quick recap on closed, open-weight, and open-source models:
Closed models keep the AI’s inner workings secret, and are only available through the owner’s website or API. Open-weight models let you download pre-trained AI and run it on your own computer. However, the original training data is kept private. While open-source models are fully transparent, so you can (typically) see the data, code, and blueprint.
In other words, open-weight models are valuable because they are accessible. The letter claims that sweeping, premature restrictions would undermine that advantage for American innovation.
The argument for access:
A global development ecosystem: Open-weight models are increasingly used to train, test, and improve AI systems across borders.
Distillation ≠ theft: These signatories want targeted enforcement against misuse, not restrictions on a common development technique.
Security requires access: Open-weight models allow for proactive defense strategies, since cybersecurity teams need to identify and respond to AI-enabled threats.
Follow the money
It’s worth noting that altruism isn’t the only motivator in this letter. Sure, innovation and security are important. But each signatory has its own incentive to keep open-weight models in play.
And it comes down to money, honey.
For labs, it’s a way to stay competitive in the fast moving AI race. For investors, open-weights reduce the cost of getting a startup off the ground. And for infrastructure companies, there’s an increased demand for their hardware and cloud services.
So yes, principles are nice. But market share is nicer.
Shhhh, the margins are talking. - TL