Open-Weight vs. Closed AI Models
Whether the most powerful AI models should be downloadable and modifiable — or kept under the control of their developers.
An open-weight model makes its trained model weights available for others to download and modify. That does not necessarily make it fully "open source": the training data, source code or other parts of the system may still remain private.
A closed-weight model keeps those weights under the developer's control. Users typically access the model through an application, API or other service rather than downloading the underlying weights.
Why it matters to voters
The debate involves competing concerns about competition, research, safety and national security.
Open-weight models can allow researchers, startups, governments and individuals to run and modify AI without depending entirely on a small number of model providers. They can also make independent research and safety testing easier.
But once powerful model weights are released publicly, access is difficult to reverse. Users may be able to remove safeguards or modify the model for harmful purposes.
Closed models give their developers greater control over who has access and allow providers to update safeguards or restrict use. But they can also concentrate access to advanced AI in a relatively small number of companies and make outside scrutiny more dependent on those companies granting researchers access.
Recent developments
- 2024
The U.S. National Telecommunications and Information Administration concluded that available evidence did not justify restricting currently available open-weight models, while recommending continued monitoring of their risks and benefits.
- 2025
The Trump administration's America's AI Action Plan called open-source and open-weight AI important for innovation, research and competition and said the federal government should create a supportive environment for open models.
- 2026
The debate intensified as increasingly capable open-weight models, including models developed in China, raised new questions about national security, competition and safety.
In August, Trump administration officials told major AI companies that the federal government's voluntary model-testing effort would not cover open-weight models. Some AI developers and lawmakers have continued to argue for stronger evaluation of highly capable models, while open-model advocates warn that restrictions could reduce competition and entrench large closed-model providers.
Political & policy relevance
The policy question is increasingly not simply whether AI should be open or closed.
Policymakers are debating whether rules should depend on a model's capabilities and risks rather than on whether its weights are publicly available.
The Trump administration has generally promoted open-weight development as part of U.S. AI competitiveness. At the same time, national-security concerns about highly capable foreign open-weight models have created pressure for additional scrutiny.
AI companies themselves also disagree over where the line should be drawn. Anthropic, for example, has said that open-weight models without dangerous capabilities are a public good and that it does not support a general ban on open weights, while also arguing for stronger safeguards as models become more capable.
Candidate working on this
Sam Liccardo has proposed encouraging open-source model deployment where consistent with national security, while also supporting stronger AI testing and safety institutions. Other lawmakers are addressing the broader question through proposals concerning frontier-model safety, export controls, competition and federal AI standards.
Further reading
Last updated September 2026