xAI Sues Over Yet Another Colorado Law That Threatens Free Expression

Over the past decade, Colorado has been sued over a series of laws that attempted to govern how various speakers may or may not speak under the guise of anti-discrimination or professional regulation. And Colorado’s track record is about as bad as it gets, having been rebuffed by the Supreme Court in three major cases for silencing expression and discriminating against viewpoints the state did not favor. 

But Colorado lawmakers have decided that maybe the fourth time’s the charm, as Colorado’s AI anti-algorithmic discrimination law similarly tries to regulate the speech associated with AI tools. And in a newly filed lawsuit, xAI accuses Colorado of again trying to silence speech that conflicts with the state’s viewpoint. 

While this case also presents incredibly important legal questions over the effect of a state patchwork of laws on technological innovation and the dormant commerce clause that my colleague Jennifer Huddleston discusses here, the threat this law poses to free expression is so significant that this case deserves even further attention. 

AI and Expression 

But let’s start at the beginning—why is xAI suing Colorado for limiting free expression? Because fundamentally, Grok, xAI’s main AI product, generates a whole range of expressive content in line with xAI’s mission and viewpoint. AI has a variety of definitions, but the definition in 15 USC § 9401(3) defines AI as “a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments.” Generative AI explicitly focuses on artistic and expressive decisions to generate imagery, text, and audio. 

Indeed, there are many expressive decisions around these generative AI tools. Perhaps most obviously, users of AI tools are making decisions about what exactly they want to create, whether it’s a hit song, a piece of art, or an AI editor. Grok users frequently use Grok to fact-check information on X. 

But the developer of the AI tool also makes many expressive decisions, ranging from what data to train the AI on, how to weight different types of information, any restrictions the developer may wish to impose on the model, or any other editorial decision to reflect the values of the developer. In the words of xAI, they “designed and developed Grok to answer only to evidence and reason, without regard to political correctness, ideological biases, or anything that might distort objective truth.” 

While these editorial and curation decisions involve algorithms and complex AI systems, that in no way makes them less expressive in nature. They are functionally identical to a newspaper editorial board deciding the positions it will take, the guest op-eds it will run or not, and the language choices it makes on controversial subjects. And the Supreme Court has been clear in cases like Hurley v. Irish-Am. Gay, Lesbian, & Bisexual Grp. of Boston that an entity maintains its First Amendment rights even if it is not trying to convey a “particularlized message,” nor does it “forfeit constitutional protection simply by combining multifarious voices” like AI tools do. 

Hurly was prominently and recently cited by Justice Kagan in her decision in the 2024 NetChoice cases that rebuffed laws trying to govern the speech of social media platforms in Florida and Texas, making it clear that the expressive decisions of technology companies are protected by the First Amendment. And like in Colorado, Hurley also involved a state trying to punish a speaker using its anti-discrimination law. 

Discrimination for Me But Not for Thee 

Colorado’s law imposes various requirements on AI developers to prevent and warn others about how their tools may be used for algorithmic discrimination. It, however, follows the trend of previous Colorado laws by forbidding or punishing certain kinds of speech the state disfavors while allowing speech the state favors. Colorado defines algorithmic discrimination as “any condition in which the use of an [AI] system results in an unlawful differential treatment or impact that disfavors an individual or group.” But Colorado explicitly exempts from this definition forms of discrimination that the state believes are beneficial, including discrimination “to increase diversity or redress historical discrimination.” 

In other words, Colorado will punish AI tools that generate speech the state believes to be bad forms of discrimination while permitting such tools to generate speech that supports state-favored forms of discrimination. 

Of course, nothing stops Colorado from enforcing proper, existing laws against those who use AI tools for illegal and malicious purposes. But Colorado’s AI law goes beyond that by forcing AI tools to preemptively take action to audit and assess their systems to prevent any outputs that the state could consider to be illegal discrimination. In other words, the law isn’t needed to punish illegal and discriminatory uses of AI—the law is written to force AI developers to spend significant resources to try to identify and prevent all the various ways their tool could be used illegally. AI tools are being and will be used in countless ways, so this law will inevitably result in developers restricting lawful speech from their AIs in the name of compliance, especially given Colorado’s view of what constitutes harmful discrimination. 

And Colorado clearly has strong views on what kinds of speech are harmful and discriminatory, as seen in its multiple losses at the Supreme Court. 

In the 2018 Masterpiece Cakeshop case, the Supreme Court found that the Colorado Civil Rights Commission had shown “clear and impermissible hostility” towards beliefs they disliked. But this was ultimately a narrow ruling focused on Colorado’s conduct, not necessarily the law itself. 

But then, in 2023, the Supreme Court more robustly ruled against Colorado’s application of its anti-discrimination laws in 303 Creative. In this case, the state accepted that 303 Creative, a custom wedding website designer, was engaging in expressive conduct but argued that anti-discrimination efforts were a compelling enough interest to limit such free expression. The Supreme Court disagreed, holding that the “First Amendment prohibits Colorado from forcing a website designer to create expressive designs speaking messages with which the designer disagrees.” It should be clear that decisions about what an AI tool will or will not generate similarly constitute expressive design choices. 

And most recently in 2026, the Supreme Court held in Chiles that Colorado’s law prohibiting conversion therapy was unconstitutional in its application to a Christian counselor. The Court held that the law not only attempted to regulate professional speech but also did so in a way that was not viewpoint-neutral. In its 8–1 decision, Justice Gorsuch wrote, “The First Amendment stands as a bulwark against any effort to prescribe an orthodoxy of views, reflecting a belief that each American enjoys an inalienable right to speak his mind and a faith in the free marketplace of ideas as the best means for finding truth. Laws like Colorado’s, which suppress speech based on viewpoint, represent an egregious assault on both commitments.” 

Justice Kagan concurred, writing that “Of course, it does not matter what the State’s preferred side is” because the First Amendment would also protect counselors from conservative laws that sought to outlaw pro-transitioning speech. Again, it is not difficult to see how Colorado’s AI law similarly fails this test of viewpoint neutrality in permitting AI tools to support certain forms of discrimination while punishing tools that could be used to support other forms. 

Trying to Put the State in Charge of Online Speech and Knowledge 

The impact of this case is even more serious than those of prior cases because AI has spread into nearly every facet of life. Website designers, custom cake bakers, and therapists are significant but still fairly specific areas where the Supreme Court ultimately favored free expression. AI, though, is quickly becoming the means through which we access most information online and through which a great deal of expression is created or edited. 

As Greg Lukianoff writes, AI is, “in an important sense, the new libraries. And if the great knowledge machines of our age are incentivized to say not what is true, but what will not get them sued, we are all in serious trouble.” Google’s AI summaries, Grok’s fact checks, Midjourney’s image generation, Claude Code’s creation of new computer codes, Chat GPT’s analysis of documents, and every other AI tool of expression and knowledge—including open-source tools—would face legal risks if their tools are able to generate answers that the state of Colorado doesn’t like. 

When the tools of modern expression and knowledge are subject to onerous regulations that punish speech disfavored by the state, then we are handing the government control over the very information we consume and the content we can generate regarding controversial issues. And if allowed to stand, conservative states can easily pass a bill that takes a different view of what AI tools should be forbidden from generating. 

For example, Texas’s AI bill, TRAIGA, was inspired by Colorado’s approach and will impose extensive and often vague regulatory burdens on AI developers, though at least the final version of the TRAIGA also limited punishment to cases where there is some intent to discriminate. But like Colorado, these vague compliance mandates will lead developers to restrict the expression of their AI tools, this time to please Texan regulators and policymakers. Such a patchwork of regulation will cripple AI innovation and its expressive potential. 

As I noted earlier, this has massive implications for how the courts think about individual states imposing their rules on interstate commerce. Together with the clear threat to free expression and inquiry, this is yet another lawsuit challenging a Colorado law that deserves to succeed.