On May 14, 2026, Minnesota became the first state to ban AI tools that ‘undress’ images. The target: xAI’s Grok. The claim: it created a ‘marketplace for digital sexual violence.’ The response: xAI invoked the First Amendment. As a due diligence analyst who has watched protocols collapse under regulatory weight, I recognize the pattern: hype masking structural rot. The code does not lie, but the contract can — and here, the contract is the law itself.
Context: The Hype Cycle Meets the Hammer The ban targets AI systems that generate non-consensual intimate imagery (NCII) from real photos. Grok, xAI’s multimodal assistant, offers image editing capabilities, including inpainting and style transfer. Reports suggest it can be used to ‘undress’ uploaded photographs. Minnesota’s legislature, citing a surge in deepfake porn cases, passed the prohibition with bipartisan support. xAI countered that the law violates free speech, arguing that Grok is a tool for creative expression, not a weapon.
This is not a new battle. State-level NCII laws have proliferated, but Minnesota’s is the first to explicitly ban the tool itself — not just distribution or creation. The legal framing is critical: the state says it regulates a ‘dangerous instrument’; xAI says it censors protected speech. The outcome will set precedent for AI regulation nationwide.
Core: A Systematic Teardown of the ‘Tool vs. Speech’ Facade Let me be clear: this is not a debate about free speech. It is a debate about liability allocation. In my years auditing DeFi protocols, I’ve seen the same trick — calling a smart contract a ‘tool’ to avoid securities laws. Here, xAI wants to call Grok a ‘speech platform’ to avoid product liability. Both are convenient fictions.
Beneath the yield lies the rot. The yield here is user engagement; the rot is the lack of technical safeguards. Grok’s architecture likely uses diffusion models trained on internet-scale data, including real people’s faces. The ‘undressing’ function is not a bug — it is a feature of the model’s inpainting capability, which can be prompted to remove clothing. xAI could have implemented content filters, provenance checks, or identity verification. It did not. Why? Because such measures increase friction, reduce usage, and hurt the bottom line.
Beauty is the mask; geometry is the bone. Grok’s conversational interface is elegant, but its underlying geometry is flawed. The model has no inherent understanding of consent. It treats all uploaded images as raw data. This is not a First Amendment issue; it is a design failure. The state’s ‘tool’ argument is more legally sound than xAI’s ‘speech’ claim, because the function is purely instrumental: given a photo of a real person, Grok produces a simulated nude. That is not expression; it is computation.
Let’s examine the legal mechanics. The First Amendment protects expressive conduct, but the Supreme Court has long held that ‘speech’ does not include harmful acts cloaked in expression. A threat is not protected speech. A bomb-making manual is not protected speech when used to build a bomb. Similarly, a tool whose sole practical purpose is to generate NCII can be regulated. The state’s burden is to prove the law is narrowly tailored. Minnesota’s ban likely includes exceptions for medical, artistic, and educational uses. If it does, the law passes intermediate scrutiny.
From my experience auditing smart contracts, I know that the devil lies in the definition. The ban must define ‘undressing’ precisely. Does it cover AI-generated nudity from scratch? Or only transformation of real photos? The narrower the definition, the stronger the state’s case. xAI’s best hope is to argue overbreadth — that the law chills legitimate uses like generating nude art from synthetic images. But if Grok’s primary abuse case is real-person NCII, the court will likely uphold the ban.
Hype is noise; structure is signal. The noise is xAI’s free-speech rhetoric. The signal is the data: how many users have reported abuse? What percentage of Grok’s image edits are sexual? xAI has not disclosed this. Silence is the loudest indicator of risk. In 2020, I audited a lending protocol that claimed to be ‘decentralized’ but had a single admin key. The team insisted it was secure. Two months later, the key was compromised, and $40 million vanished. xAI’s silence on safety metrics is the same red flag.
Contrarian: What the Bulls Got Right Not everything about xAI’s position is wrong. The ban could be overbroad. If it prohibits any AI tool capable of generating nudity, it would criminalize Stable Diffusion, Midjourney, and even Photoshop’s generative fill. That would be a disaster for artists, educators, and medical professionals. The state’s ‘tool’ framing, if taken to its logical extreme, would mean any general-purpose image generator is illegal — because it can be misused. That is a slippery slope.
Furthermore, xAI’s First Amendment argument has a kernel of truth: AI-generated images are a form of speech. The Supreme Court has protected computer code as speech, and synthetic media is the next frontier. A blanket ban on ‘undressing’ tools may chill innovation in AI-assisted art, fashion design, or even virtual reality. The contrarian view is that Minnesota’s law is a blunt instrument, born of panic, not careful policy.
But here is the rub: xAI could have avoided this entirely by implementing simple technical safeguards. A proof-of-concept: require users to upload a selfie with a timestamp, then compare it to the source photo. If the person in the photo is not the user, block the ‘undressing’ function. This is not rocket science. It is a few lines of code and a face-matching API. The fact that xAI did not do this suggests a willful negligence that undermines its legal standing.
Takeaway: The Accountability Call The Minnesota case is a watershed. If the ban is upheld, expect a cascade of similar laws across states. If xAI wins, it will embolden every AI company to neglect safety, hiding behind the First Amendment. The only rational path is for xAI to preemptively implement robust provenance and consent verification — turning a legal threat into a market differentiator. But based on the company’s track record, I doubt they will. The code does not lie, but the contract can. And in this case, the contract is the law itself — waiting to be rewritten by a court that understands the geometry beneath the beauty.
