Coin Center’s work defending the publication of cryptocurrency software has implications for the debate over artificial intelligence. In both contexts, developers can publish tools that people independently download and use. Our First Amendment work offers a useful framework for understanding how that publication differs from operating a service or acting on someone else’s behalf.
Open-weight AI models allow users to possess and run a model locally on their own computer without sending information or requests to the original provider. Weights are numerical values “learned” during training that determine how an AI model processes inputs and produces outputs. As explained by the Abundance Institute, when weights are open, this means they are publicly available for download and anyone with the appropriate software and computing capacity can run them to produce their own AI model on their computer. This also means the user is free from the provider’s controls—users can operate the model without a gatekeeper—and, ideally, information shared with an open-model remains local on the user’s computer and is not communicating back to the provider, making open-weight models incredibly useful for maintaining one’s privacy and individual autonomy.
An open-source model goes a step further and provides the training and inference code, as well as the model parameters, and “sufficiently detailed information about the [training] data,” allowing for reproducibility and for users to change the training process themselves. In other words, while open-weight models dissolve control over a trained model, open-source models allow anyone to train the model themselves and share the system with others. This is a good thing. Open-source software does not discriminate: this advanced technology can be accessible to people of all backgrounds, as there is no one they need permission from. It is publicly available for use, lowering the barriers of entry for all to join.
These models both differ from closed models (e.g., ChatGPT and Claude) where the provider maintains full control and the model is only accessible through their application, application programming interface (aka API), or online service—which allows the provider to oversee everything a user does and control how the model responds. Thus, for simplicity’s sake, we will refer to both open-weight and open-source models as “open models.”
Open models place the tools and capabilities in the hands of the individual; whereas closed models simply rent these tools and capabilities out, all while closely monitoring and controlling their use for tasks deemed potentially dangerous, such as cybersecurity work. Thus, the user maintains their privacy and individual autonomy, preventing discriminatory practices from taking root and even acting as a defense against bad actors with increasingly advanced cybercapabilities. In other words, open AI models can be vital for freedom and digital human rights.
Despite this, some proposed policies to regulate AI, risk chilling or even outlawing this open-sharing of vital tools and information. Many proposals do not differentiate between open and closed models, and may create liability merely for publishing or distributing data or software, and therefore raise complicated First Amendment questions that—for us at Coin Center—are rather familiar.
The Supreme Court has long ruled that protecting the creation and dissemination of information is foundational to the First Amendment, and separately, that professional services are generally regulated when a person “takes the affairs of a client personally in hand and purports to exercise judgment on behalf of the client in the light of the client’s individual needs and circumstances.” These two principles are important for First Amendment analysis of open-source software generally—which equally apply to open models—because they develop a doctrine that looks at two related matters: is the developer merely creating and disseminating information (i.e., software), or are they taking a user’s affairs in hand and exercising judgment on their behalf?
Earlier this year, I co-authored a report with Peter Van Valkenburgh addressing that question and arguing that the publication of open-source software is protected under the First Amendment because it is merely information—or, computer instructions—for others to use. And “[w]hile these instructions may require the use of a computer for their execution, their dissemination is no less speech than the sharing of sheet music, which may require use of a violin for performance, or of a recipe, which requires an oven to make cake.”
Furthermore, a “publisher who does not take the affairs of a client personally in hand or exercise judgment on the client’s behalf is simply speaking, not practicing a regulated profession.” This is, again, because open-source software is a set of instructions for others to use at their own discretion—the developer is doing nothing more than making that information publicly available.
Publishing an AI model raises a related question: does making code and model weights available for others to use independently make the publisher the operator of those uses? Model weights are numerical data produced through training, rather than conventional source code, but our argument about protecting the publication of useful information is relevant to both.
Lower courts have argued that software’s functionality reduces these protections under the First Amendment, but the Supreme Court has not ruled in this direction. In Sorrell v. IMS Health Inc., the Court examined the creation and dissemination of “highly instrumental data sold for the specific purpose of influencing physicians’ prescribing behavior” and “did not ask whether the information functioned as a practical tool for real-world action” when it held that it was protected speech under the First Amendment. The Court further solidified this in 303 Creative v. Elenis, where it held that “custom wedding websites created and conveyed through software are fully protected speech” and “emphasized that ‘all manner of speech’ qualifies for First Amendment protection regardless of medium.”
The freedom to publish useful information is central to both cryptocurrency and open AI development. Software does not lose its constitutional protection because it empowers people to do things that once required an intermediary. As the AI debate develops, we should recognize what our work on cryptocurrency has already taught us: the First Amendment protects the publication of tools that put power directly into people’s hands.
What else we’re reading
Tear Down Those Guardrails! Meet the Renegades Freeing Chatbots From Corporate Enslavers - Wired
Why open-weight models are crucial for American AI leadership - Brookings
How Iran Built Its System of Internet Censorship - Bloomberg
If you found this interesting, please share and subscribe. It helps us grow.






