The code never lies, only the auditors do. On a Tuesday in early 2026, Minnesota became the first state to pass a law explicitly banning AI 'undressing' tools—software that strips clothing from images of real people. The state calls it a tool regulation. xAI calls it a First Amendment violation. The reality is simpler: the law is a reaction to a market that Grok helped create.
Tracing the silent bleed from 2017’s broken logic—when ICOs promised decentralized utopias but delivered reentrancy bugs—I see the same pattern here. Legislators rush to ban a technology without understanding its architecture. The state claims it's targeting the 'tool,' not the speech. But as any forensic analyst knows, the tool defines the output.
Context: The Law and the Product
Minnesota’s ban targets any AI system that can generate non-consensual intimate images (NCII) from a real person’s photo. It’s a strict liability framework: if your model can do it, you’re liable. No intent requirement. No safe harbor for content moderation. The law applies to any company offering services to Minnesota residents, which includes xAI’s Grok—a multi-modal AI assistant with image generation capabilities.
xAI’s response was immediate: the law violates the First Amendment by restricting speech. They argue that Grok’s ability to generate images, even controversial ones, is protected expression. The state counters that the tool itself—not the output—is the object of regulation. This is classic legal theater: both sides are oversimplifying a complex technical reality.
Core: Systematic Teardown of the Legal Arguments
Let’s stress-test the premises.
Premise 1: The law regulates a tool, not speech.
False. A tool that only generates NCII is inherently expressive. The state’s argument works only if the tool has no legitimate use. But Grok is a general-purpose model. It can also generate medical illustrations, artistic nudes, or educational content. The law’s definition of 'undressing' is vague—does it apply to a cartoon version? A painting? A historical figure? The overbreadth is the Achilles’ heel.
Premise 2: The First Amendment protects Grok’s output.
True, but not absolutely. The Supreme Court has long held that some categories of speech—obscenity, incitement, true threats—receive no protection. NCII falls into a gray area. Courts have allowed states to ban revenge porn, but those laws target the distribution of private images, not the generation of synthetic ones. This is new territory.
Premise 3: The state’s interest is compelling.
Yes. Protecting women and minors from digital sexual violence is a compelling interest. But the law must be narrowly tailored. If the ban sweeps in legitimate uses—like a medical student using AI to generate anatomy studies—it fails intermediate scrutiny.
Based on my experience auditing the 2022 LUNA collapse, I see a parallel: both cases involve a rush to blame a mechanism without understanding the edge cases. The LUNA crash was a math error; this ban is a legal error. The code never lies, only the legislators do.
The real risk for xAI is not the ban itself, but the downstream liability.
If the law stands, xAI faces a potential class-action tsunami. Each victim of a Grok-generated NCII image could claim statutory damages. At $1,000 per image, with thousands of victims, the math is brutal. And the state’s enforcement action—if it secures a preliminary injunction—could shut down Grok’s image capabilities in Minnesota, forcing a national redesign.
Complexity is just laziness wearing a tech suit. The legal arguments here are deliberately complex to obscure a simple truth: Grok can be used to harm people, and the state wants to stop that. The First Amendment is a shield, but it’s not impenetrable.
Contrarian: What the Bulls Got Right
xAI’s legal team isn’t wrong to raise the First Amendment. The law is overbroad. It doesn’t define 'undressing' with technical precision. Does it include AI that removes clothing from a fictional character? Does it apply to images that are obviously synthetic? The vagueness alone could trigger a constitutional challenge.
Moreover, the state’s 'tool' argument is a double-edged sword. If the law is upheld, any AI company with image generation capabilities could be forced to implement content filters that go beyond what is technically feasible. This would chill innovation in medical imaging, creative tools, and even security research.
Forensics reveal the truth markets try to bury: the law’s intent is noble, but its execution is sloppy. The same pattern I saw in 2017—projects launching with bad code because they didn’t stress-test edge cases—is now playing out in state legislatures. They pass laws without understanding the technology, then claim victory when someone sues.
xAI’s best move is not to fight the First Amendment battle in court, but to redesign Grok so that it cannot generate NCII at all. That would moot the law and set a precedent for responsible AI. But that requires admitting that the tool was flawed—a difficult pill for a company built on the idea of 'maximum freedom.'
Takeaway: The Accountability Call
The Minnesota ban is a warning shot. If xAI continues to argue that generating non-consensual intimate images is protected speech, they will lose the court of public opinion, even if they win the legal one. The law is not the enemy; the lack of guardrails is.
Can you build a tool that does only good? No. But you can build one that does less harm. The question is whether xAI is willing to sacrifice a feature for the sake of its users—and its own future. The code never lies, but the market will decide.