Tracing the ghost in the machine.
In late April 2026, Minnesota became the first U.S. state to ban the distribution and use of AI tools designed to “undress” real people in images. The law, which targets platforms like xAI’s Grok, classifies the technology not as a form of protected speech but as a dangerous instrument. xAI has already signaled a First Amendment challenge, arguing that the ban criminalizes the very act of AI-generated expression. The clash is not just legal—it is a narrative war over what an AI model fundamentally is: a tool for harm, or a canvas for creation.
Context: The Silence Between the Blocks
I’ve been tracking AI-image controversies since 2022, when I audited the early output of Stable Diffusion for a small fund. Back then, the debate was about copyright. Now, it’s about the body. Minnesota’s ban is part of a wave of state-level laws targeting non-consensual intimate imagery (NCII) generated by AI. What makes this law distinct is its focus on the tool itself—not just the distribution of the output, but the very existence of the function that allows a user to upload a photo of a real person and receive a nude version. The state’s attorney general has publicly stated they are not regulating speech; they are regulating a mechanism. This is a crucial distinction. Under U.S. constitutional law, speech restrictions trigger strict scrutiny, while regulation of conduct (like manufacturing a lockpick) often only requires rational basis review. By framing the AI model as a “tool,” Minnesota hopes to avoid the heavy burden of defending a content-based ban.
But the reality is more complex. Grok is not a lockpick. It is a generative model that can produce an infinite variety of outputs. The same latent space that creates a nude of a celebrity can also create a medical illustration for a dermatology textbook. The law’s wording is critical: if it defines “undress” broadly enough to include any image that partially removes clothing, it could sweep in countless legitimate uses. xAI’s best argument is not that they have a right to generate non-consensual nudes—they don’t, and no one would defend that. Their best argument is that the law is overbroad, that it chills protected expression, and that it forces a private company to become the censor-in-chief for an entire state. And that is where the ghost in the machine becomes visible.
Core: The Narrative Mechanism of Compliance
From my perspective as a token fund manager who has spent years analyzing the fragility of decentralized trust, this case is a textbook example of how regulatory pressure reshapes the architecture of a product. xAI faces a choice that every major AI platform will eventually face: split your model per jurisdiction, or build a single global model so heavily filtered that it loses its creative edge. The Minnesota law does not just ask for a content filter—it asks for a “reasonable design” that prevents the undressing function from existing at all. That means the model’s weights must be modified or the API must be geofenced with near-perfect accuracy. Geofencing is leaky. Model modification is expensive and fragile.
Based on my audit experience, I can tell you that the real cost here is not legal fees—it’s the loss of the product’s soul. Grok was marketed as an unfiltered, “truth-seeking” AI. If it now has to embed a moral override for every prompt involving a real person’s image, the model’s behavior becomes a patchwork of legal compromises. The code is law, but trust is fragile—and once users realize that the AI is not free but constrained by a patchwork of state laws, the narrative of radical transparency collapses. The market will punish xAI not because of the ban, but because of the betrayal of its own brand promise.
Moreover, the legal lens reveals a deeper technical reality: the “tool” vs “speech” distinction is a false binary. A generative AI model is a dual-use artifact—it can be both a tool for creating speech and a tool for inflicting harm. The U.S. Supreme Court has long held that the First Amendment does not protect instructions on how to commit a crime, but it does protect the medium of expression. If Minnesota’s law is upheld, it will set a precedent that states can regulate the underlying capabilities of AI models, not just the outputs. This is the true narrative shift: from regulating content to regulating potential. And that is a much more dangerous precedent for the entire AI industry.
Contrarian: The Myth of Decentralized Perfection
Here is the counter-intuitive angle that most analysts miss: the First Amendment challenge might actually be the weakest part of xAI’s defense. Why? Because the public sentiment against non-consensual deepfake nudes is overwhelming. The law has a powerful emotional narrative—protecting real people’s dignity from a machine that can strip them without consent. xAI’s attempt to frame this as a free speech issue risks painting them as defenders of a “digital sexual violence marketplace,” a phrase the article itself uses. The contrarian truth is that xAI’s best path is not to fight the law on principle, but to embrace the regulation and turn it into a competitive advantage.
Consider: if xAI voluntarily builds the most robust provenance and consent-checking system in the industry, they can market Grok as the only AI that respects the “scarce resource” of authenticity. They can implement a system where every image of a real person is watermarked with a cryptographic proof of consent, stored on-chain, and verified before any edit is allowed. This is not a fantasy—it’s a technical architecture that already exists in the NFT world (proof of provenance, soulbound tokens). Why not extend it to AI image generation? The cost is high, but the narrative payoff is enormous: xAI becomes the guardian of digital identity, not the enabler of digital violence. The alternative—fighting a losing legal battle while the media brands them as a purveyor of harm—will destroy their brand value far more than any regulatory compliance cost.
But this requires a fundamental shift in the company’s culture. xAI, like many in the crypto and AI space, has a libertarian streak that resists oversight. The myth of decentralized perfection—that code alone can solve trust—is exposed here. The law is not an enemy; it is a signal of what society values. And right now, society values the protection of the body over the freedom of the machine. The ghost in the machine, if it wants to survive, must learn to listen to the silence between the blocks—the silence of the victims who never asked to be digitized.
Takeaway: Listening to the Silence Between the Blocks
The Minnesota case is a bellwether. If xAI loses, every major AI model will be forced to embed state-level compliance into its core architecture—a fragmentation that will accelerate the call for federal AI regulation. If xAI wins narrowly, it may only delay the inevitable. The real question is not whether AI can generate nudes, but whether the industry can prove that it can be trusted with the most intimate data of all: our likeness. Authenticity is the only scarce resource. And the only way to preserve it is to build systems that treat consent as a non-negotiable line in the sand, not as a legal loophole to be exploited.
As I wrote in my 2021 essay on digital rareness, the value of a token is not in its code but in the community’s belief in its integrity. The same applies here. The code is law, but trust is fragile. And the ghost in the machine is watching.
