Jennifer Huddleston
Some politicians and thought leaders, including Hilary Clinton, have echoed a concerning belief that open-source or open-weight-or-source AI models pose a greater risk than equally powerful closed models. But the idea that open-source or open-weight models are inherently dangerous is misguided and misunderstands why a variety of model types serve different purposes in the AI ecosystem.
The greater risk right now with open-source or open-weight AI is not a security risk; it is the consequences for innovation and liberty if the government restricts how the US AI industry develops and favors one solution over another.
What even is an open model?
To understand both where concerns about securing open models come from and the potential impact of policy that would limit such models, one must first understand the difference between open and closed AI models.
Closed versus open technology is not unique to AI. A closed system in the context of technology generally refers to products (in this case, AI models) that cannot be changed or have very limited potential for change after purchase. Many popular AI models, such as Anthropic’s Claude and OpenAI’s ChatGPT, can be applied to various tasks, but they generally cannot be modified. In contrast, open technology, be it AI models or more traditional technologies such as operating systems, allows users greater flexibility in how they modify and run programs.
When it comes to AI, there are different ways a model can be open. Open-weight models publicly provide more information and transparency about how they weigh inputs and generate outputs, but without training or code information. True open-source AI models provide not only information about their weights but also the training data and code. The result can be greater accessibility and a way to better customize a model or agent to a particular need.
As with other technology and software, open or closed is not inherently better or worse. Different applications may have different balances between accessibility and flexibility versus the ease of use or trust that may come with an established closed-source model.
Why Open-Source Matters
For many users, open-weight and open-source AI products provide important flexibility and options. Open-weight models allow anyone with the right setup to run the model themselves, while open-source models allow anyone with the right setup to build a similar model themselves. For example, Ben Affleck expressed his amazement at the capabilities when he used open-weight models based on his own dataset for a recent film project.
The recent scrutiny of open-source or open-weight AI models is likely the result of two policy dynamics. An increasing concern over cybersecurity-related AI threats and the growing popularity of China-linked models that are typically open source or weight. This is only amplified as leaders of closed-model AI labs criticize their more open competitors as inherently risky.
As with the broader AI debate, these concerns should be understood generally, not about a particular type of model or business. In the age of AI and emerging advances in quantum computing, there are valid conversations to be had about cybersecurity preparedness. Similarly, there may be valid conversations to have around the vulnerabilities and use of Chinese models in certain infrastructure. But these concerns should not result in the demonization of open source more generally or in hindering the development of competitive US-backed models.
More open models can provide a more adaptable, flexible, and often lower-cost option that can provide an important opportunity to find and develop solutions that would be costly or not possible within the limitations of a closed model. Such benefits also mean these models can be critical to preventing a top-down approach that could itself raise concerns about a government’s ability to interfere with important access to information or services. Additionally, the flexibility and adaptability of open-source models create the potential for greater competition at times and an alternative way to compete in the market for those whose preferences on any number of output decisions can vary.
Initially, open models were seen as poorer quality than closed models, but many companies have closed this gap. As of a January 2026 study, open models were performing at greater than 90 percent of closed models’ capabilities, with gaps often quickly closing that gap further. While some see risk, this can provide greater access to the tools both individuals and businesses need.
Just as in operating systems, closed versus open AI models can serve different purposes and may be a better fit for different tasks. In some cases, the security or “ready to go” nature of a closed model is preferable. In other cases, open models offer greater flexibility, lower costs, and opportunities for experimentation that may help meet needs that are too expensive or not currently met by closed-source models.
But even this is not always the case. For example, when Claude “attacked” Hugging Face’s open-weight models, it had flexibility and features to determine what was happening. As a result, it is important that policy not overly favor one particular type of model development, particularly as technology and its applications continue to develop.
Policymakers should not make misguided assumptions about the risks of open AI models, but they should consider whether stringent policies could foreclose access to such tools. There is a demand for open models both in the US and internationally. Policy that prevents competitive open-source models from developing in the US would ultimately harm our AI industry’s global competitiveness.














