X.AI LLC v. Ellison

UNITED STATES DISTRICT COURT

DISTRICT OF MINNESOTA

X.AI LLC,

Plaintiff, v.

Keith Ellison, in his official capacity, as Attorney General of Minnesota,

Defendant.

Civil No. 26-3425 (DWF/DTS)

MEMORANDUM

OPINION AND ORDER

INTRODUCTION

At the center of this lawsuit is a new Minnesota law, Minn. Stat. § 325E.91 ("H.F. 1606"). H.F. 1606, 94th Leg., Reg. Sess. (Minn. 2026) (to be codified at Minn. Stat. § 325E.91). H.F. 1606 prohibits companies that own specific artificial intelligence ("AI") technology products, like Plaintiff X.AI LLC's ("xAI") Grok Imagine, from allowing users of their technology to generate certain sexual images and images of other intimate parts. xAI brought this action against Defendant Keith Ellison, in his official capacity as Attorney General of Minnesota (the "State"), arguing that H.F. 1606 violates the First Amendment. This matter is before the Court on xAI's motion for a preliminary injunction blocking the law, which the State opposes.¹For the reasons set forth below, the Court denies the motion.

1

2

BACKGROUND

I. The Law—H.F. 1606

H.F. 1606 provides, in relevant part, the following:

Nudification prohibited. (a) A person who owns or controls a website, application, software, program, or other service must not:

(1) allow a user to access, download, or use the website, application, software, program, or other service to nudify an image or video; or

(2) nudify an image or video on behalf of a user.

. . . .

Exemption. The prohibitions [above] do not apply when the website, application, software, program, or other service requires the technical skill of a user to nudify an image or video.

Minn. Stat. § 325E.91, subdivs. 2, 3 (2026). The law defines "nudify" as:

[T]he process by which . . . an image or video is altered or generated to depict an intimate part not depicted in an original unaltered image or video of an identifiable individual; and . . . the altered or generated image or video is so realistic that a reasonable person would believe that the intimate part belongs to the identifiable individual.

Id., subdiv. 1(d). "Intimate part" has the meaning given in Minn. Stat. § 609.341, subdiv. 5, which provides: "'Intimate parts' includes the primary genital area, groin, inner thigh, buttocks, or breast of a human being."

H.F. 1606 imposes consequences for noncompliance. Relevant to this motion, the law allows the attorney general to enforce the law against violators and seek civil penalties up to $500,000 for each violation, as well as injunctive relief and other remedies. Minn. Stat. § 325E.91, subdiv. 5 (2026); see also Minn. Stat. § 8.31, subdiv. 3

(2025).

3

The legislative history of H.F. 1606 sheds light on the range of harms caused by AI-nudification technology that was considered by the Minnesota legislature. For example, at the Commerce Finance and Policy Committee hearing,²three women testified to the extensive harms they suffered after a perpetrator used nudification technology to create realistic pornographic material (images and/or videos) using their likenesses. The harms included missed work, fear for family members and in particular their children, fear of leaving the house, and the emotional toll of wondering who has seen or who will see convincing, sexually graphic images and videos of them. This particular perpetrator created non-consensual pornographic images and videos of over 80 women.

In addition, material provided during the same hearing revealed widespread use of nudification technology now that it is available via various apps and websites. RAINN, an anti-sexual violence organization that operates the National Sexual Assault Hotline, shared that in one month alone in 2023, there were 24 million users of nudification apps. And in 2023, over 95 percent of manipulated images were nonconsensual, sexually explicit, and featured women.

The committee also heard evidence that this technology harms both adults and children. With respect to children, AI-nudification technology is being exploited to create child sexual abuse material, and there has been a marked increase in AI-generated child sexual abuse material on dark web forums, as well as on the clear web, including

4

commercial sites. The Internet Watch Foundation ("IWF") identified a "pedophile guide"

that encourages the use of nudification tools to generate material and to blackmail children, and found perpetrators online expressing their affection for nudification technology. In addition, the IWF highlighted that in the first half of 2024, the center received 6,800 reports flagging child pornography created with generative AI, and through June of 2025, tips increased to over 400,000 reports. The committee also learned that nudification technology presents a real and growing threat in schools. AI software has been used to fabricate sexually explicit images of students and those images can then be quickly circulated through group chats on various social media platforms. AI-nudification technology has resulted in a flood of sexualized images and allowed the harmful images to spread online and, in many cases, remain online forever. At a Judiciary Finance and Civil Law Committee hearing,³the Senior Legislative Policy Counsel for RAINN submitted the results of a study conducted after a social media site launched an AI tool allowing users to edit images and post the images with one click. This study examined images created in the first 11 days after the AI tool was launched and found that approximately 65 percent of the photos created were photorealistic sexualized images—amounting to a total of 3 million such images at a pace of 190 images per minute. Roughly 23,000 of the sexualized images were of children.

5

Seeking to address the flood of harmful images and damaging effects of AI- nudification technology, on April 23, 2026, the Minnesota House passed the anti- nudification bill at issue in this case. Minn. H.J., 94th Leg., Reg. Sess. 6139-40 (2026). The Minnesota Senate passed the bill unanimously on April 29, 2026. Minn. S.J., 94th Leg., Reg. Sess. 9257-58 (2026). The Governor signed the bill into law on May 7, 2026. Minn. H.J., 94th Leg., Reg. Sess. 7315 (2026). The law went into effect on August 1, 2026. Minn. Stat. § 325E.91 (2026).

II. xAI and Grok Imagine

xAI is an AI developer with products that include the family of models known as Grok. (Doc. No. 1 ("Compl.") ¶ 12.) Grok includes an image and video generation model called Grok Imagine. (Id.) Grok Imagine allows users to create images and videos by typing or dictating a short description of the image or video they wish to create and an image or video arrives in seconds. (Id. ¶ 23; Doc. No. 14 ¶ 7.) Grok Imagine is available in all fifty states, including Minnesota. (Compl. ¶¶ 12, 24.) xAI submits that the images and videos that users create with Grok Imagine span the full range of visual culture, including art, animated family photographs, memes, product mockups, political satire, and religious imagery. (Id. ¶ 21.)

Grok Imagine can be used to create sexual images through its nudification tools. xAI maintains that, through its terms of service, it prohibits users from engaging in "any illegal, harmful, or abusive activities, or any other behavior that violates [its] Acceptable Use Policy." (Id. ¶ 25.) xAI further asserts that its policy prohibits, among other things, nudification of a real person's image, altering a real person's image to depict them in an

6

intimate or sexual context, and sexualizing or exploiting children. (Id.) xAI stresses that it enforces its rules against users who violate the terms through various measures, including account suspensions, account terminations, and reporting suspected child sexual abuse material to law enforcement. (Id. ¶ 26.)⁴

Despite xAI's efforts to rein in its users' nefarious use of xAI's technology, the State points to evidence that the steps xAI is currently taking are insufficient to address the unique harms that stem from the technology. For example, the State cites to at least three lawsuits filed against xAI involving allegations that the photographs of children were used to create child pornography on Grok Imagine. Amended Complaint, Doe 1 v. X.AI Corp., No. 5:26-cv-2246 (N.D. Cal. July 7, 2026); Amended Complaint, Does v. X.AI, No. 4:26-cv-750 (E.D. Ark. Aug. 17, 2026); Amended Complaint, Does v. X.AI, No. 4:26-cv-772 (E.D. Ark. Aug. 17, 2026). In addition, the State notes that SpaceX, xAI's parent company, has set aside a fund with more than $500 million for potential litigation losses stemming from Grok Imagine's "Spicy" and "Unhinged" modes. Maxwell Zeff & Paresh Dave, SpaceX Listed Grok's 'Spicy' Mode as a Risk in Its IPO Filing, Wired (May 20, 2026, at 08:43 PM), https://www.wired.com/story/spacex-ipo- grok-spicy-mode-risks [https://perma.cc/4DXQ-33WW].

7

In this matter, xAI argues that H.F. 1606 violates the First Amendment, both facially and as applied, and moves for a preliminary injunction blocking the law. (Doc. No. 11.) The State opposes the motion. (Doc. Nos. 20, 38.)

DISCUSSION

"A preliminary injunction is an extraordinary remedy . . . ." Winter v. Nat. Res. Def. Council, Inc., 555 U.S. 7, 24 (2008). A district court has "broad discretion" in determining whether to issue an injunction. United Indus. Corp. v. Clorox Co.,

140 F.3d 1175, 1179 (8th Cir. 1998). Exercising its "equitable discretion," eBay Inc. v. MercExchange, L.L.C., 547 U.S. 388, 391 (2006), a district court may grant a preliminary injunction when a movant demonstrates that: (1) it is likely to succeed on the merits;

(2) it is likely to suffer irreparable harm in the absence of preliminary relief; (3) the balance of equities tips in its favor; and (4) an injunction is in the public interest. Winter,

555 U.S. at 20; see also Dataphase Sys., Inc. v. C L Sys., Inc., 640 F.2d 109, 113 (8th Cir. 1981) (en banc). The party requesting the injunctive relief bears the "complete burden"

of proving that a preliminary injunction should be granted. Gelco Corp. v. Coniston Partners, 811 F.2d 414, 418 (8th Cir. 1987).

I. Delay and Irreparable Harm

As a threshold matter, the State argues that xAI's motion for preliminary injunction should be denied for lack of diligence and irreparable harm. "[A] party requesting a preliminary injunction must generally show reasonable diligence."

Benisek v. Lamone, 585 U.S. 155, 159 (2018) (per curiam). The issue of delay is related to the issue of irreparable harm as "it has long been recognized that delay in seeking

8

relief vitiates much of the force of allegations of irreparable harm." Novus Franchising, Inc. v. Dawson, 725 F.3d 885, 894 (8th Cir. 2013) (citation modified). In addition, irreparable harm occurs when a party lacks an adequate remedy at law, "typically because its injuries cannot be fully compensated through an award of damages." Gen. Motors Corp. v. Harry Brown's, LLC, 563 F.3d 312, 319 (8th Cir. 2009). "[A] failure to demonstrate irreparable harm, standing alone, may be a sufficient basis to deny preliminary injunctive relief." Caballo Coal Co. v. Ind. Mich. Power Co., 305 F.3d 796, 800 (8th Cir. 2002).

H.F. 1606 was signed into law on May 7, 2026. Nearly three months later and only three days before the law was set to take effect, xAI filed this motion. xAI has not offered an explanation for its delay, other than to generally suggest that it takes time for laws like H.F. 1606 to reach the attention of corporate management and for management to green light the resources to challenge the law. (Doc. No. 45 at 32-33.) Acknowledging that there is no bright-line rule marking when a motion for a preliminary injunction must be filed in a case such as this, there is also no serious dispute that xAI is a sophisticated and well-resourced litigant with the means to move quickly when its interests are at stake. If xAI genuinely feared irreparable harm, the Court does not doubt it would have acted more quickly to file this action and move for injunctive relief. The fact that it did not do so strongly suggests a lack of irreparable harm.⁵

9

In addition to xAI's delay, xAI has failed to otherwise demonstrate a threat of irreparable harm. First, xAI points to the potential imposition of a $500,000 penalty per violation as a potential measure of irreparable harm. This penalty, however, is monetary in nature. Even if penalties were assessed, which is unlikely because xAI has disabled its nudification tool in Minnesota (Doc. No. 40 ¶ 19), those financial penalties could be recouped in the event that the law is eventually invalidated. Therefore, this monetary penalty, even if substantial, is not irreparable. See, e.g., Corning Sav. & Loan Ass'n v. Fed. Home Loan Bank Bd., 562 F. Supp. 279, 283 (E.D. Ark. 1983) (explaining that monetary loss constitutes irreparable harm only where the movant's very existence is threatened).

Second, xAI argues it will suffer commercial injury absent a preliminary injunction. In support, it points to the declarations of Naser Baseer, xAI's product counsel team lead. (Doc. Nos. 14, 40.) Baseer attests that on August 1, 2026, xAI implemented additional controls on Grok Imagine's image-editing feature for users in Minnesota in response to H.F. 1606. (Doc. No. 40 ¶ 19.) Baseer adds that in order to implement those controls, xAI had to design, test, and deploy geographically targeted controls within Grok Imagine's image-editing pipeline. (Id.) Baseer submits that between July 24, 2026 and August 1, 2026, xAI engineers spent "a substantial amount of their time" on the controls that diverted them from other projects. (Id.) In addition, xAI submits that since it has implemented the new controls in Minnesota, the rate of that motions for TROs are analyzed under the same framework as a motion for preliminary injunction).

10

moderation in Minnesota for image-editing prompts has risen over 10 percent and that this increase will result in the loss of Minnesota users and related business for xAI. (Id. ¶¶ 22, 24.)

These examples of "commercial injury" do not demonstrate a threat of irreparable harm. The Court acknowledges that xAI has taken action to adjust its AI tools to avoid violating H.F. 1606 and to manage an increased rate of moderation. And the parties agree that any associated cost will not be compensable by the State if the law is invalidated. Even so, xAI has not demonstrated that these costs are irreparable. With respect to the adjustment of its AI tools, by xAI's own account, the adjustment has already been made, and xAI does not provide evidence that there will be any future cost. In addition, xAI claims that it is expending substantial resources, but the evidentiary support for this is vague and conclusory. Merely stating so is not enough. As to the possibility that increased moderation will result in lost Minnesota users and related business, any such loss is purely speculative. xAI provides a single declaration that contains vague and speculative substantiation of this claimed harm. In addition, H.F. 1606 applies to all companies providing its users with AI-tools that allow for nudification. Therefore, there will be no loss of users vis-à-vis any other competing company. Finally, xAI argues that a violation of the First Amendment alone demonstrates irreparable harm. xAI cites Maixner v. Eidsness for the proposition that the loss of First Amendment freedoms, even for a short time, constitutes irreparable injury. 182 F.4th 736, 738 (8th Cir. 2026). In Maixner, the Eighth Circuit recognized this general rule as it has been applied in cases where it is clear that First Amendment interests are either

11

threatened or in fact being impaired at the time relief was sought. Id. at 738-39 (citing Elrod v. Burns, 427 U.S. 347 (1976)). The Eighth Circuit, however, also recognized that even in the First Amendment context, a party must make a clear showing of likely irreparable harm to justify the imposition of a preliminary injunction. Id. at 739 (first citing Kohls v. Ellison, 166 F.4th 728, 732-33 (8th Cir. 2026) (concluding that the district court did not abuse its discretion in denying extraordinary preliminary relief in light of the delay in bringing the request); and then citing Hotchkiss v. Cedar Rapids Cmty. Sch. Dist., 115 F.4th 889, 894 (8th Cir. 2024) (affirming the district court's denial of a preliminary injunction in a case involving the First Amendment where the movant failed to make an adequate showing of irreparable harm)). This is because, as explained by the Supreme Court, "in a First Amendment case[,] 'a preliminary injunction does not follow as a matter of course from a plaintiff's showing of a likelihood of success on the merits [and] a court must also consider'" the remaining factors, including likelihood of irreparable harm and the balance of the equities. Id. (quoting Benisek, 585 U.S. at 158). Importantly, even in the First Amendment context, a court may independently deny a motion for a preliminary injunction when a moving party acts with unreasonable delay. See Hotchkiss, 115 F.4th at 894.

For the above reasons, the Court concludes that xAI has both unreasonably delayed in bringing this motion and has failed to otherwise demonstrate a sufficient threat of irreparable harm. The motion for preliminary injunction is properly denied for these reasons alone. Even so, the Court considers the balance of harm and public interest.

12

II. Balance of Harms and Public Interest

When a preliminary injunction is sought against the government, the balance of harms and public interest merge. Morehouse Enters., LLC v. Bureau of Alcohol, Tobacco, Firearms & Explosives, 78 F.4th 1011, 1018 (8th Cir. 2023). In considering these two factors together, the Court considers whether the movant's likely harm without a preliminary injunction exceeds the nonmovant's likely harm with a preliminary injunction. Cigna Corp. v. Bricker, 103 F.4th 1336, 1347 (8th Cir. 2024). The purpose is not to determine the rights of the parties, but to balance the equities as the litigation moves forward. Id.

On the record before it, the Court finds that the balance of harms and public interest tips steeply in favor of the State. The State has enacted, democratically and nearly unanimously, a civil statute aimed at protecting the public from the undisputed harm stemming from AI-nudification technology and the resulting proliferation of photorealistic sexualized images. It is evident, and xAI does not dispute, that the State has an interest in curbing the generation of those images. The State's choice to prohibit the provision of AI tools to users who create such images is a direct response to the unique harms stemming from the new technology and the State's conclusion that curtailing user violations after harmful images are circulated is not enough. On the other side of the State's interest lies xAI's alleged harm. Again, xAI's evidence of irreparable harm is unconvincing. In addition, xAI's claim that its First Amendment interests are threatened such that preliminary injunctive relief is warranted is belied by their own delay. And when comparing the interests of the State with any potential harm to xAI, the

13

Court finds without hesitation that the balance of the equities weighs heavily in favor of the State. See, e.g., Maryland v. King, 567 U.S. 1301, 1303 (2012) ("[A]ny time a State is enjoined by a court from effectuating statutes enacted by representatives of its people, it suffers a form of irreparable injury." (alteration in original) (quoting New Motor Vehicle Bd. of Cal. v. Orrin W. Fox Co., 434 U.S. 1345, 1351 (1977))).

CONCLUSION

The parties sharply contest the strength of xAI's First Amendment claim. xAI of course argues that it is likely to succeed on the merits. The State responds by challenging xAI's standing to bring a First Amendment claim on behalf of its users, disputing xAI's own First Amendment interest at stake, and arguing that H.F. 1606 is valid under any level of scrutiny. The constitutional issues raised by the parties are complex, particularly when considered in the context of this new technology and the risks that it poses to the public. These issues deserve, and will receive, full consideration. To be sure, a determination of xAI's constitutional claims will be made in the near future because the State has filed a motion to dismiss that places these issues front and center. In the future, the Court may also consider whether xAI is entitled to permanent injunctive relief on the merits. Until then, because xAI has not demonstrated irreparable harm and the balance of harms tips steeply in favor of the State, xAI's motion for a preliminary injunction is denied.