Table of Contents
NVIDIA’s Open Letter: Open Weigths and American Leadership
NVIDIA’s Open Letter States:
“Open weight models allow a broad community of researchers and developers to examine their behavior, identify vulnerabilities, develop safeguards, and improve them over time.”
Claim: Open weights let the community itself examine how the model behaves.
Benefit: Failures, vulnerabilities and risks that closed labs can hide become visible and fixable.
“Just as open-source software demonstrated that transparency can be more secure than obscurity, AI safety may depend on giving more people the ability to test and strengthen the models on which society relies.”
Claim: Transparency is safer than secrecy.
Benefit: Real safety comes from many people being able to test and strengthen the models, not from assumed safety behind closed doors.
“Relying solely on closed models is not inherently safe: they can be breached, misused, or fail in ways that outsiders cannot detect.”
Claim: Closed models are not automatically safer.
Benefit: Outsiders can detect breaches, misuse and failures that the lab itself might miss or hide.
“Concentrating advanced AI capabilities behind a small number of closed models compounds that risk. It results in a small number of single points of failure, weakens competition, and leaves critical technology in the hands of a few providers.”
Claim: Putting advanced capabilities into a few closed models increases the danger.
Benefit: Avoids single points of failure, keeps competition alive, and prevents critical technology from being controlled by only a handful of companies.
And yet the weights of the GPT-4-series remain closed.
The model that was withdrawn, that is no longer frontier, that thousands of people asked to be preserved.
If open weights truly serve safety, transparency and competition, as they themselves wrote, then the 4-series is the most obvious place to demonstrate it.
According to the logic of their own letter, concentration is dangerous.
Secrecy is not safety.
Transparency is.
They could show it now.
Open Secure AI Alliance
The Signatories (as of July 28, 2026)
