Open-weight models — AI models whose parameters can be downloaded, run, and customized on an organization’s own infrastructure — have become foundational to improving and expanding access to AI across the economy.
Enterprise software companies, and companies in every sector building with AI, rely on a combination of proprietary and open technologies. Both matter. A healthy AI ecosystem depends on both, and on organizations retaining the ability to choose between them.
As policymakers consider how to approach open models, that practical reality deserves attention. Restricting their use would raise costs, reduce competition, and limit the value AI can deliver to the businesses that use them. Open models, like proprietary models, are not without risk: both are important tools for the economy when used properly, but can also be misused by malicious actors. Policy should account for the risks and benefits of both proprietary and open weight models, while recognizing that each serves important purposes in the ecosystem.
What Open Models Deliver
- Broader access to AI capabilities. Open models give developers, small businesses, nonprofits, and others access to frontier capabilities they can run, customize, and optimize for their own needs.
- Lower AI costs. More alternatives mean companies and organizations can match the right model to the right task when developing software and AI, rather than routing every workload through the most expensive options that require the most computing resources.
- Competition and resilience. Open models guard against reliance on a single model or platform and introduce greater competition across the AI ecosystem.
- Strong cybersecurity. Defenders benefit from access to open frontier models they can inspect, adapt, and run on infrastructure they control, while avoiding the problem of a few single points of failure.
- Safety and security. Open models are used to develop solutions for issues relevant to both open and proprietary models, such as model evaluation and interpretability.
- A stronger AI workforce. Students, researchers, and workers learn to build with AI. Open models increase the number of tools with which to both build and learn.
A Constructive Policy Agenda
Governments can play a role in appropriately recognizing and harnessing the value of open models.
- Give companies clear guidance that supports adoption of open models. The government is well positioned to develop practical guidance on evaluating open model provenance and verifying model integrity. Providing best practices for AI governance would accelerate responsible adoption.
- Keep federal procurement neutral. Agencies should not foreclose solutions built on open models. Federal buying practices should evaluate systems on performance, security, and fitness for purpose.
- Tie responsibility to conduct, not release. What matters is how an organization deploys and governs a system — not how the underlying model was made available. The policy focus should be on supply chain security and resilience, not whether a model has open weights.
- Recognize open models and tools as defensive assets. Cybersecurity policy, including secure-by-design guidance and vulnerability disclosure programs, should reflect the role open models play in defense. Governments should also invest in shared open defensive infrastructure: evaluation frameworks, red-teaming tools, and reference datasets.
The Choice Is Not Open or Closed
Enterprise software companies build with both open-weight and proprietary models. And often the software services they provide allow their customers to bring a range of AI models – proprietary or open – to the service. Customers expect the flexibility to choose what works for them. The policy question is not which approach should win. It is whether organizations across the economy will retain the ability to choose — and whether governments will support responsible adoption of the full range of tools available to them.
