A recent court case filed by xAI challenges a Colorado AI law’s “algorithmic discrimination” requirements and its impact on speech. This case could have a significant impact on our understanding of AI, the First Amendment, and the broader constitutionality of state-level AI regulations.
The Colorado AI Law and the Problems of a Potential State Patchwork for AI
Colorado became the first state to pass a comprehensive AI law in 2024. Even the governor’s signing statement expressed concerns about the consequences. In some cases, states can serve as laboratories of democracy on a policy issue by showing the results of different regulatory regimes. However, regulating AI at the model level has impacts well beyond an individual state’s borders, as Jack Solowey and I wrote at the time the Colorado law was enacted.
The policy conversation around state AI laws has changed since 2024. Colorado itself has slowed implementation of the AI law and formed working groups with civil society and industry that suggested significant amendments. Colorado’s AI policy quagmire has not slowed the overall onslaught of state-level legislation, with over 1,000 bills considered in the 2024–2025 legislative session. Some of these proposals, such as legislation around studies on impact, are within a state’s borders. However, many risk extraterritorial impacts beyond a state’s borders by more directly seeking to regulate data or the development of AI.
The broader conversation around state AI laws at a federal level has shifted to a growing concern about the potential disruption of such laws. In summer 2025, the narrative and debate around state AI laws shifted with new proposals for a moratorium on state-level AI laws and federal preemption in the One Big Beautiful Bill. While the moratorium failed, a December executive order directed the Department of Justice to form a task force that would engage in litigation to challenge state AI policies inconsistent with the federal approach. The DoJ formed this task force in January 2026.
While the federal government is not currently a party to xAI’s lawsuit, the case could signal what future government-led litigation against state AI laws might or might not successfully argue. Among xAI’s arguments in its complaint is the way the Colorado law violates the Dormant Commerce Clause by “directly regulating development and deployment activities occurring entirely outside Colorado.” As it correctly notes, state AI laws like Colorado’s place a significant burden on AI developers beyond their operation in the state, as it is not possible to accommodate such requirements without changing the development of AI nationwide.
If the court does not recognize the impact such laws have, American innovators will face a Sacramento effect, or in this case, a Boulder effect, where the most stringent state laws become de facto federal regulation due to the nature of technology. As a result, even Americans in states like Utah that have taken a more innovation-friendly approach would face the innovation and speech-chilling consequences of a more restrictive state’s approach.
Why This Could Be AI’s Stratton-Oakmont Moment
Even if the court does not reach the underlying question of whether state AI laws constitute an undue burden on interstate commerce, it will still have to grapple with key questions about AI development and speech. Like the early internet cases of the 1990s, this could be one of the first tests of the judiciary’s ability to adapt and apply existing standards to AI.
At the heart of the case, as xAI notes in its complaint, is how “every choice that xAI makes when developing Grok is an expressive act protected by the First Amendment. These choices embody deliberate judgment that reflects xAI’s hierarchy of values and its viewpoint-driven philosophy about how best to develop AI tools that will advance the knowledge, understanding, and overall progress of human civilization.” This is not unique to xAI’s products, but at the heart of understanding how AI regulation could impact free expression. Like most scenarios at the intersection of AI and free speech, the questions are not on the speech rights of the AI itself but rather the rights of the human developing or deploying the AI.
Some may detest some of the speech Grok or other AI’s have produced and how their developers have chosen to encourage or discourage their product. It is fair for the consumers and the market to react by choosing other products that better fit their preferences and values. Over the last few years, the question of the right of developers to determine the way their product will respond to queries as expressive activity is increasingly important to prevent potential censorship concerns and other restrictions on expression, not only for AI but also for other technologies.
To get this wrong could be the Stratton-Oakmont v. Prodigy moment for AI. If courts find that such a design is not an expressive activity under the First Amendment, it could both deter innovation and open this technology up to direct government control or censorship. Such a ruling would have a concerning impact on a technology that has real potential to expand expression in a variety of ways. While courts might eventually have arrived at a sound norm for the internet without policy innovation based on the existing distributor liability frameworks, significant advancements would have been lost in the process. For AI, this could mean that literally life-saving technologies are delayed. This may sound hyperbolic, but AI’s algorithmic decisions are critical to things like medical research and personalized medicine and chilling the rights of developers around algorithmic decisions could more generally deter development.
Similar questions are also at stake in considering other current policy debates, such as the Pentagon’s labeling of Anthropic as a supply chain risk. As Cato argues in its amicus brief in that case, part of why the government’s designation is so against the values of a free society is its violation of First Amendment rights to determine how to design one’s expressive product. Similarly, assertions that the choices in how one programs content outputs are not protected speech would have significant impacts on social media content moderation and many of the other ways algorithms are used on the internet.
Conclusion
About a year ago, I asked in the title of a law review article if AI is a horse or a zebra. The title is a play on both Judge Easterbrook’s early internet era article Cyberspace and the Law of the Horse (in which he cautions against specialized law for the internet) and the phrase “when you hear hoofbeats think horses, not zebras.” In that article, I explored how the courts could likely apply existing understandings of free expression or minimal legislative and legal changes to most of the questions that will arise around AI. However, I noted that one of the potential stress points for existing law would be around the question of algorithms and how a ruling that development decisions were not protected by the First Amendment could “otherwise limit online speech and increase concerns around potential liability in the AI age, as well as with current platforms.”
Given both the Anthropic case and this new Colorado litigation, it seems we have reached a critical moment in determining the trajectory of AI and speech.
