xAI has spent more than six months insisting Grok’s safeguards were fine. It took a Minnesota statute with a $500,000-per-image penalty to get the company to admit, in writing, that it’s preparing to change the product.
That admission is buried in a lawsuit Elon Musk's company filed Monday against the state of Minnesota, arguing that its ban on nudification technology is unconstitutional. The filing is a First Amendment challenge on its face. Read it closely and it’s also a confession about what the company was willing to ship until the math got scary.
The number that changed xAI’s mind
Minnesota’s law takes effect August 1. It threatens firms like xAI with fines of up to $500,000 for every single harmful output discovered in the state. No distribution requirement. The output counts even if only the user ever saw it.
xAI ran the arithmetic in its own complaint, and the escalation is the point.
“A company whose users request just ten images in violation of the statute would face exposure up to $5 million in civil penalties alone. A company with a thousand violative images could be fined up to $500 million. And a business whose users created a hundred thousand images covered by [the law] (not at all unlikely for a publicly available program with millions of users generating billions of images) could owe an eye-popping $50 billion dollars.”
The law also gives victims a right to sue xAI over any individual output, stacking private claims on top of civil penalties.
So the company blinked. “Confronted with $500,000-per-image strict liability and no safe harbor, xAI has no practical choice but to restrict Grok Imagine‘s image-editing features in various ways when the statute takes effect on August 1, 2026,” xAI argued. “Protected speech freely available before the law takes effect will thus be chilled.”
What it would rather do instead
Given the choice, xAI says it would change nothing at all. Its preferred approach is the one it’s been running: terms of use stipulating that users could be banned for using Grok to make CSAM or other kinds of non-consensual intimate images.
“But for [the law] and its penalties, xAI would continue to offer the editing feature exactly as it does today,” the company said.
That’s a policy document as a safety mechanism. It’s also the status quo that produced the arrests.
Suing your own users as a legal strategy
Earlier this month, xAI started suing users who are allegedly circumventing its safeguards to generate AI CSAM. The company now cites those suits as evidence in the Minnesota case, arguing that going after bad actors proves it’s already doing everything in its power and that Grok therefore deserves safe harbor.
It’s a tidy loop: sue the users, then point at the lawsuit to argue you shouldn’t be liable for what the users did.
Last week, xAI got sued itself after tipping off police to a well-known Arkansas child photographer who allegedly used Grok to sexualize images he took of hundreds of clients. A complaint from a 10-year-old Jane Doe alleges the photographer made hundreds of images of Doe and thousands of images and videos of other kids.
Doe’s lawyer told us many of those kids are preparing to file their own lawsuits, and those claims could end up being added to a potential class action filed in March by young girls in Tennessee.
The definition fight, and why it’s the whole case
Most of xAI’s argument comes down to how Minnesota defines “intimate parts.”
The company prefers the definition in anti-revenge porn laws like the Take It Down Act, which xAI said relied on a narrower definition to criminalize distribution of both real and AI-generated “intimate visual depictions.” Under that federal law, harmful images are those exposing “the genitals, pubic area, anus, or partially or fully exposed nipple of an individual.”
Minnesota pulled its definition from a criminal statute covering “nonconsensual touching” of the “inner thigh, buttocks, or breast of a man or woman, as well as the groin and primary genital area.”
xAI’s objection is that a touching statute translates badly to image generation. The definition “accordingly bans ordinary depictions of men without shirts, people in shorts or swimsuits, and other body parts routinely displayed in public—far beyond what an ordinary person would consider ‘nudification,'” the company argued.
The company also argued the law sweeps in nude images generated with “artistic, scientific, political, satirical, educational, medical, or religious value,” and hammered on scope: “liability attaches even if the depicted persons consented —or created the image themselves —and even if the image is never shared.”
Notice which examples xAI picked
Here’s where the filing gets careful in a way that’s worth reading twice.
xAI’s illustrations of protected expression are political: mocking politicians by reimagining them in Speedos or as Sumo wrestlers, both of which would produce exposed thighs and male breasts. The complaint tiptoes around bikinis, which is notable given that much of the Grok backlash followed an X post Musk made advertising Grok’s ability to put anyone in a bikini.
The closest xAI came to acknowledging that women and girls were targeted by its scandal was citing an output that Donald Trump generated, showing him with his political allies shirtless in a pool with what xAI suggested was an AI-generated woman in a bikini.
The complaint similarly avoids much discussion of the arrests of Grok users. Doe’s lawyer suggested to us those are happening at an increasing pace, as xAI appears to be partnering more closely with law enforcement in some cases.
244 arrests out of 70,000 reports
xAI has been criticized for failing to help police ID one Grok user who was eventually arrested. In its lawsuit, the company said it’s motivated to catch bad actors, and that out of 70,000 reports of suspected Grok-generated CSAM, at least 244 cases led to arrests.
The company offers that figure as proof of what it calls state-of-the-art or “near-perfect” detection of harmful outputs. Seventy thousand reports is also the denominator.
Minnesota isn’t budging
The state’s position is that it can block images at the source, before distribution, before the lifelong harms to kids that the lawsuits describe.
Asked for comment, the office of Minnesota Attorney General Keith Ellison confirmed the state will likely defend the law.
“We’re waiting to be served and will review the lawsuit and response in Court,” Ellison said. “But I know that using AI to generate nude images of people against their will is appalling. There are plenty of worthy debates to have about AI policy. This is not one of them. AI nudification robs the target of their dignity and can cause immense harm on an emotional, personal, and professional level.”
xAI did not respond to our request for comment.
What a win would actually do
xAI argued Minnesota has less restrictive means available, and claimed the Take It Down Act already protects users from the harms of distribution. If the court agrees, Minnesota could be enjoined from enforcing the law against any nudification app. Not just Grok. Any of them.
Without an injunction, “providers may well have no choice but to alter their services in Minnesota rather than run the risk that users will evade their controls,” xAI argued. “And that would hamper all uses of AI image-generation tools, including those that [the law] does not purport to regulate.”
Beyond the safe harbor claim, xAI argued that Grok’s process for creating any images or videos is entitled to First Amendment protections, echoing other chatbot makers who have argued that chatbot speech deserves the same protections as human speech.
The test the case turns on
This will come down to narrow tailoring. xAI argued Minnesota’s law can’t survive any level of scrutiny, and even conceded the state’s underlying goal.
“Although Minnesota likely has a compelling interest in preventing individuals from creating nudes of other people without consent and distributing that content, the statute punishes AI platforms for assisting in the creation of a substantial amount of protected speech in relation to that limited goal,” xAI argued.
Strip the constitutional framing and one detail sits there unmoved. xAI told a federal court it will restrict Grok Imagine’s image-editing features on August 1 because a state made harmful outputs expensive. Six months of backlash and probes didn’t do it. A dollar figure did.
