AI coalition urges Washington to protect open-weight models

An industry coalition including NVIDIA, Microsoft, Meta, OpenAI, IBM, GitHub, Hugging Face, Mistral, Cohere and the Linux Foundation has asked U.S. policymakers to protect open-weight AI from broad restrictions. Published on July 24 as “Open Weights and American AI Leadership,” the statement’s current Microsoft page lists 35 companies, foundations and investors.

The group defines open-weight models as systems whose trained parameters can be downloaded, inspected, modified and run on an organisation’s own infrastructure. That is different from fully open-source AI, which may also provide training code, data and a reproducible recipe. Even so, access to weights gives makers far more freedom to adapt a model, host it privately and avoid dependence on one API provider.

The policy request is concrete. The signatories want wider access to computing capacity for startups and researchers, investment in shared datasets, tools and evaluation frameworks, and no premature rules that push open-model development overseas. They also defend distillation, in which one model’s outputs help train or evaluate another, while accepting that unlawful extraction from closed services should be addressed through targeted legal and commercial measures.

The letter acknowledges the central risk: once weights are released, the original developer cannot retrieve them, and modified versions can be difficult to trace. Its answer is more benchmarking, red teaming and safeguards tied to demonstrated harms. The coalition argues that closed models also create vulnerabilities and single points of failure, while outside researchers can examine and improve open systems.

That argument is influential, but it is not neutral. Many signatories sell chips, cloud capacity, enterprise software, model hosting or open-weight models, so they benefit from a broad ecosystem. The statement changes no law and does not settle whether the most capable future models can be released safely. It does, however, place unusually large companies and open-source organisations on the same side of an increasingly important policy debate.

For AI users and makers, the immediate effect is choice rather than a new product: more self-hosting, customisation and competition if the coalition gets its way. For businesses, open weights can improve control over data, costs and deployment, but responsibility for security, evaluation and maintenance shifts toward the organisation running the model. Policymakers now face the harder task of protecting against genuine misuse without closing off a route that many smaller builders depend on.