Full Opinion

FOR PUBLICATION FILED UNITED STATES COURT OF APPEALS SEP 17 2026 MOLLY C. DWYER, CLERK U.S. COURT OF APPEALS FOR THE NINTH CIRCUIT S.D., individually and on behalf of all others No. 26-4872 similarly situated; J.F., D.C. No. 6:25-cv-01726-CL Plaintiffs - Appellees, v. OPINION MIKE REESE; RYAN LEGORE; MARK NOOTH; ERIN REYES; STATE OF OREGON, by and through the Oregon Department of Corrections; JEREMY WAGNER, Defendants - Appellants, JOHN DOES, 1-5, Defendant. Appeal from the United States District Court for the District of Oregon Mark D. Clarke, Magistrate Judge, Presiding Argued and Submitted August 27, 2026 Pasadena, California Before: Daniel P. Collins, Lawrence VanDyke, and Eric C. Tung, Circuit Judges. Opinion by Judge Tung TUNG, Circuit Judge: The district court in this case issued a class-wide preliminary injunction requiring (among other things) that the State of Oregon presumptively assign more than a hundred “transgender women” prison inmates (male inmates who claim to be female) for placement in the only state-run women’s prison in Oregon. The State of Oregon and Department of Corrections officials (Defendants) appealed. Defendants contend that the district court based its injunction on a clearly erroneous finding that Defendants applied a “default presumption” that “transgender” inmates be assigned to men’s prisons. That “presumption,” in the district court’s view, exposed those inmates to a substantial risk of harm in violation of the Eighth Amendment. We agree with Defendants that the district court committed clear error. The record does not support the district court’s finding that Defendants applied a “presumption.” Nor does the record justify class-wide relief. We vacate the preliminary injunction and direct that the class be decertified. I. Plaintiffs J.F. and S.D. brought a class action against Defendants. Plaintiffs allege violations of the Eighth Amendment (and other causes of action not relevant here) for Defendants’ purported failure to guard against Plaintiffs’ alleged exposure to a substantial risk of harm in men’s prisons. Plaintiffs, claiming to be 2 26-4872 “transgender women” inmates, sought a preliminary mandatory injunction on behalf of a class of “all current and future transgender women in [Oregon Department of Corrections (‘ODOC’)] facilities.” 3-ER-316. Plaintiffs asked, among other things, to be presumptively assigned to the women’s prison. 3-ER- 377. The district court (Clarke, Magistrate Judge) granted the motion and provisionally certified a class. 1-ER-16. The district court acknowledged that issuance of a mandatory injunction must clear a high bar—that “the law and the facts clearly favor Plaintiffs’ position.” S.D. v. Rees, No. 6:25-CV-01726-CL, 2026 WL 1146591, at *8 (D. Or. Apr. 28, 2026) (emphasis added). In the district court’s view, that burden was met with respect to Plaintiffs’ Eighth Amendment claims. Id. at *8–11, 13. Defendants filed a motion for reconsideration. 2-ER- 110–139. The district court denied reconsideration in part and granted it in part, issuing an amended preliminary injunction that largely tracks Plaintiffs’ requests with a few exceptions not relevant here. See S.D. v. Rees by & through the Oregon Dep’t of Corr., No. 6:25-CV-01726-CL, 2026 WL 1960717, at *6 (D. Or. July 7, 2026). The revised mandatory injunction requires that Defendants: 1. Immediately end the categorical housing of transgender women in men’s facilities without individualized safety assessments and instead conduct prompt, case-by-case housing and classification reviews for all transgender women in ODOC custody. The starting point for each assessment shall begin with a presumption of placement consistent with each person’s gender identity, absent a documented and articulable 3 26-4872 security justification. Then, the assessment shall take into consideration the AIC’s [adult-in-custody’s] preferences, their privacy, access to medical support, programming, support-network, safety, or other individualized reasons for placement in any particular facility. 2. Provide safe, non-punitive housing options for transgender women, including (as appropriate) transfer to CCCF [Coffee Creek Correctional Facility] [the women’s facility], voluntary transgender or gender non- conforming housing units within existing facilities, or other protective placements that do not involve involuntary segregation or loss of programming solely because of transgender status. 3. Prohibit housing transgender women with cisgender cellmates known to have a history of sexual or serious physical violence. 4. Ensure that transgender women have access to separate showers or shower schedules that provide reasonable bodily privacy from male prisoners and staff and implement policies to prevent forced exposure of nude bodies to male inmates. 5. Require that unclothed searches of transgender women be conducted by female staff absent extraordinary, documented circumstances and prohibit cross-gender strip searches otherwise. 6. Implement immediate, confidential, and non-retaliatory reporting mechanisms for sexual abuse, harassment, and threats, including effective [Prison Rape Elimination Act] PREA-compliant hotline access and prompt, good-faith investigations of all such reports with measures to protect reporters from retaliation. 7. Provide interim mental health support, include crisis counseling, for transgender women who report sexual abuse or exhibit suicidality related to unsafe housing or retaliation. 8. An AIC may decline any of these protections with informed, voluntary consent after a meaningful review of options. 1-ER-3–4. 4 26-4872 Defendants appealed and sought emergency relief. We granted Defendants’ request for a stay of the preliminary injunction, and of all district court proceedings, pending appeal. Stay Order at 1 (No. 26-4872), Aug. 20, 2026. We now address the merits of the preliminary injunction. II. “A preliminary injunction is an extraordinary remedy never awarded as of right.” Winter v. Nat. Res. Def. Council, Inc., 555 U.S. 7, 24 (2008). To obtain that remedy, a plaintiff must satisfy four factors: (1) “that he is likely to succeed on the merits”; (2) “that he is likely to suffer irreparable harm in the absence of preliminary relief”; (3) “that the balance of equities tips in his favor”; and (4) “that an injunction is in the public interest.” Id. at 20. “We review an order regarding preliminary injunctive relief for abuse of discretion, but review any underlying issues of law de novo.” Norbert v. City & Cnty. of San Francisco, 10 F.4th 918, 927 (9th Cir. 2021) (citation omitted). An abuse of discretion occurs when the district court’s “decision relies on an erroneous legal standard or clearly erroneous finding of fact.” Arc of Cal. v. Douglas, 757 F.3d 975, 983 (9th Cir. 2014) (internal quotation marks omitted). And a finding is clearly erroneous if it is “illogical, implausible, or without support in inferences that may be drawn from the facts in the record.” Id. (citation omitted). 5 26-4872 Plaintiffs fail to show that they are likely to succeed on the merits. The remaining Winter factors weigh against them too. Accordingly, the district court’s preliminary injunction, resting on a contrary conclusion, must be vacated. A. Whether Plaintiffs are “likely to succeed on the merits” is “the most important factor.” California v. Azar, 911 F.3d 558, 575 (9th Cir. 2018) (internal quotation marks omitted). Plaintiffs’ burden “is doubly demanding” in defending a mandatory injunction: They “must establish that the law and facts clearly favor [their] position, not simply that [they are] likely to succeed.” Garcia v. Google, Inc., 786 F.3d 733, 740 (9th Cir. 2015) (en banc); see also Stanley v. Univ. of S. Cal., 13 F.3d 1313, 1320 (9th Cir. 1994) (“A mandatory injunction goes well beyond simply maintaining the status quo pendente lite [and] is particularly disfavored.” (citation omitted)). Plaintiffs cannot show that the law and facts clearly favor their position. First, the injunction was based on a flawed premise that Defendants employed a “default presumption” of placing Plaintiffs (and class members) in men’s prisons. Second, even assuming such a default presumption existed, the record does not support the district court’s conclusion that the presumption produced a substantial risk of harm that Defendants were deliberately indifferent toward. Finally, the 6 26-4872 injunction ordered class-wide relief that was based on an improperly certified class. 1. The district court based its injunction on the following premise: Defendants did not conduct “individualized assessments” but rather applied a “default presumption” that placed Plaintiffs and class members in men’s prisons. S.D., 2026 WL 1146591, at *1 (“It is undisputed in the record before the Court that this default presumption, and their overwhelming placement in men’s prisons, has exposed transgender women inmates to a high risk of violence and sexual assault.”); id. at *10 (“[A] major cause of both the known higher risk of violence and the violence itself, was due to the default housing placement of transgender women in men’s prison facilities.”). That premise lacks record support and is clearly erroneous. The district court improperly disregarded the overwhelming contrary evidence presented by Defendants in favor of baseless speculation. In contending below that Plaintiffs failed to carry their burden to substantiate their claim that Defendants applied a default presumption that assigned class members to men’s prisons, Defendants submitted declarations and policy documents explaining the individualized process that ODOC uses to place “transgender” inmates. Newly admitted inmates are “assessed within 72 hours of arrival . . . and again within 30 days to determine whether they meet specific 7 26-4872 criteria indicating either vulnerability or likelihood of perpetrating sexual abuse.” OB, Add–6. Newly admitted inmates who claim “transgender, nonbinary, genderfluid, or intersex” status are “placed in the infirmary until” a committee “can gather all necessary information to decide placement,” which takes about a week. 2-ER-240. Inmates already in custody but who later “come forward to identify as transgender, nonbinary, genderfluid, or intersex remain in their current facility until [the committee] reviews and makes a placement determination.” 2-ER-240. “When determining appropriate and safe housing, committee members will consider: the AIC’s [adult-in-custody’s] medical history; the AIC’s mental health status; how the AIC self-identifies; the safety of the AIC, other AICs, and staff; observation of the AIC’s behaviors; and any other relevant information. Each housing placement is individualized and determined on a case-by-case basis.” 2- ER-239. In a further effort to refute Plaintiffs’ claim of a “default presumption,” Defendants submitted records summarizing their placement decisions, which are not disputed. Defendants stated that “[t]here are currently 117 transgender women AICs” (male inmates claiming to be women) “housed in ODOC prisons, 26 of the 117 have requested to be housed at the [women’s prison], and 8 are currently 8 26-4872 housed” there. 2-ER-240.1 “Not all transgender female AICs want to be housed at [the women’s prison],” Defendants represented. 2-ER-240. “Some transgender female AICs specifically request placement at men’s facilities, for a variety of reasons, and may be granted such placement when it is appropriate.” 2-ER-240. “Based upon our current tracking information,” Defendants said, “approximately 22 transgender women total have been housed at” the women’s prison. 2-ER-240. In addition, Defendants presented evidence concerning individual placements, which they contended further confirmed that ODOC did not apply any default presumption. In particular, Defendants described the placement decision made with respect to the only named Plaintiff (J.F.) who remains in custody.2 Plaintiff J.F. entered the Department’s custody in 2008 and identified “as transgender” in 2015. 2-ER-240. J.F. requested transfer to a women’s prison in 2021 but was “not eligible to transfer . . . due to safety concerns of the [female inmates]” at the women’s prison. 2-ER-240. J.F. was incarcerated for a sex offense against a disabled female victim who was in the care of . . . J.F.’s mother; the victim had physical limitations and was unable to communicate. The victim had to have surgery to remove an 11-inch plastic object that was lodged in her abdomen by J.F., who 1 Defendants later submitted a supplemental declaration that identified 123 “trans- female” inmates. 2-ER-147. 2 The other named Plaintiff (S.D.) is no longer in custody. See S.D., 2026 WL 1146591, at *4 n.1. 9 26-4872 lodged it into her abdomen via the victim’s vagina. The victim’s injuries were potentially fatal. 2-ER-240–241. Decisions concerning other inmates also clearly reflect individualized review. Defendants pointed to their handling of a transfer request made by L.B., who had “sexually abused a young niece and physically assaulted a girlfriend by grabbing her throat and pushing her up against a wall.” 2-ER-255. As Defendants explained, “L.B.’s request was denied due to [L.B.’s] assaultive history toward women.” 2-ER-255. In another example, Defendants acceded to inmate Doe’s request to be transferred to the women’s prison. 2-ER-255. Months after the transfer, however, Doe was returned to a men’s prison after displaying “aggressive” behavior toward the women inmates, including calling them “derogatory names (‘bitches’, ‘whores’),” and grabbing Doe’s own “[male] genitals” when walking by women inmates; moreover, Doe “request[ed] to return to a male facility.” 2-ER-256. Two and a half years later, Doe received a “vulvoplasty” and is now back in the women’s prison. 2-ER-256. Neither the district court nor Plaintiffs contested these facts, which undermine Plaintiffs’ assertion that Defendants applied a default presumption. The district court nevertheless inferred a “default presumption” from the mere fact that “more than 90% of transgender women are housed in men’s facilities.” S.D., 2026 WL 1146591, at *1; see also id. at *10 (finding a default presumption of 10 26-4872 “placement in men’s facilities” based solely on the fact that “eight out of 117 total transgender AICs are placed at CCCF, which is less than seven percent”). That inference is groundless. The district court assumed that, if the process were truly individualized, it would not have produced an outcome that was, in the court’s view, this lopsided. But no evidence supports the district court’s speculation about how many transgender inmates “should” have ended up in women’s prisons under an individualized review. Indeed, the record refutes it. For example, uncontested evidence showed that only 26 “transgender inmates” (out of 117) requested transfer to the women’s prison, and 8 of them were transferred. The low rate of transfer requests cannot be squared with the district court’s baseless assumption that, under an individualized-assessment system, there would be a much higher rate of actual placement into the women’s prison. The district court admitted to giving “little weight” to the low number of transfer requests, but its only reason for doing so was that it had “no information about the other 91 [inmates], including whether they [we]re aware that requesting a transfer is required, or that it is even an option, to be considered for housing at [the women’s facility].” Id. at *10. The district court’s rationale is legally flawed because it ignores the burden of proof. Plaintiffs bore the burden of proving the default presumption, and any gaps in the evidence therefore counted against them rather than (as the district court would have it) in their favor. Plaintiffs might have 11 26-4872 sought to show that the other 91 inmates who did not request transfer were not aware of the option or did not know how to make such a request. Rather than hold Plaintiffs to their burden, the district court essentially “substitute[d] [its] assumptions for evidence in the record.” United States v. Barnes, 895 F.3d 1194, 1199 (9th Cir. 2018). That it cannot do. Plaintiffs’ citation of cases suggesting the relevance of statistics is misplaced. See Ans. Br. 27 (citing United States v. Ironworkers Local 86, 443 F.2d 544 (9th Cir. 1971); Diaz v. Am. Tel. & Tel. Co., 752 F.2d 1356 (9th Cir. 1985); Obrey v. Johnson, 400 F.3d 691 (9th Cir. 2005)). We do not dispute that statistics can be useful. But even the authority Plaintiffs rely on says, “as is the case with all statistics, their use is conditioned by the existence of proper supportive facts and the absence of variables which would undermine the reasonableness of the inference of [improper conduct] which is drawn.” Ironworkers Local 86, 443 F.2d at 551 (emphasis added) (footnotes omitted). Plaintiffs have not offered “proper supportive facts” or the “absence of variables” suggesting a “presumption.” Id. Nor are Diaz and Obrey on point. Both Diaz and Obrey addressed the relevance of statistical evidence: whether it was “properly subject to discovery,” Diaz, 752 F.2d at 1358, and whether it was “probative” to be admitted, Obrey, 400 F.3d at 697; again, it is not relevance we dispute, but whether the inference drawn from statistics alone had any factual basis here. It did not. 12 26-4872 2. Even assuming the existence of a default presumption, however, the district court’s mandatory injunction would still be unjustified. The record here does not support the conclusion that any presumption reflected Defendants’ deliberate indifference to a substantial risk of harm to Plaintiffs and other class members—a prerequisite to establishing an Eighth Amendment violation. See Farmer v. Brennan, 511 U.S. 825, 828 (1994). The district court concluded that there was a substantial risk of harm because “[n]early 35% of transgender people in state and federal prisons were sexually assaulted between 2007 and 2021, compared to less than 5% of the general prison population who experienced the same.” S.D., 2026 WL 1146591, at *9 (emphasis added). But the Department of Justice study, cited by the district court, was not from 2007 through 2021. Rather, the study covered a much narrower timeframe— 2007, 2008–2009, and 2011–2012. See U.S. Dep’t of Just., Off. of Just. Programs, Sexual Victimization in Prisons and Jails Reported by Inmates, 2011–12, NCJ No. 241399, Supp. Table 1 (2014). The district court’s error is significant: The statistics used long predate the relevant period here—Plaintiff J.F. and other inmates (who submitted declarations) had all requested transfer during or after 2021. That is roughly a decade-long gap. The district court’s use of stale statistics does not justify finding a “substantial risk of harm.” 13 26-4872 The district court’s reliance on the DOJ study is flawed in another respect: The DOJ statistics are cumulative nationwide numbers reflecting reports at both federal and state prisons; the numbers do not necessarily reflect conditions in Oregon prisons, which likely have a different model of prison administration from many other prisons. For instance, the Oregon Department of Corrections has a specific “Transgender and Intersex Committee” that “determine[s] appropriate and safe housing for transgender and intersex adults-in-custody.” 2-ER-239. The district court, to be sure, relied also on declarations of individuals who claim to be exposed to a substantial risk of harm from sexual assault while in men’s prisons. But even if we credited those declarations, individual instances of harm would warrant, at most, individual redress of particular constitutional violations. They would not justify the class relief mandated by the district court. 3. The district court’s injunction is also based on an improperly certified class. The certified class here included “all current and future transgender women in ODOC custody.” S.D., 2026 WL 1146591, at *19. Plaintiffs “must ‘affirmatively demonstrate’ by a preponderance of actual evidence that they satisfy all the Rule 23 prerequisites.” Dickinson v. Trump, 174 F.4th 634, 646 (9th Cir. 2026) (citations omitted). But they failed to satisfy that requirement here. See Fed. R. Civ. P. 23(a)(2) (requiring “questions of law or fact common to the class”). 14 26-4872 Plaintiffs have not shown by a preponderance of the evidence that the class members’ alleged harms, their varying degrees of harm, and the source of their harms are “common” to all class members. Nor have Plaintiffs shown that Defendants acted in a uniform way with deliberate indifference toward each of the class members. At most, the record reflects that different officials acted in different ways toward different prisoners at different times. See 2-ER-240 (declining J.F.’s transfer request because of J.F.’s history of sexually assaulting a disabled female victim); 2-ER-240–41 (granting Doe’s initial transfer request but then returning Doe to a men’s prison after Doe became aggressive toward female inmates, and then ultimately putting Doe back in women’s prison upon Doe’s receiving a vulvoplasty); 2-ER-242 (denying S.S.’s transfer request because S.S. did not make the request until two months before S.S’s release date). That does not support a class. The district court’s own admission that it “has no information about the other 91 AICs” confirms that a class would be inappropriate here. S.D., 2026 WL 1146591, at *10. The district court’s finding of commonality rested on its conclusion that Defendants had a “systemic” policy that “preferences housing [proposed class members], categorically, in men’s prisons, such that they are exposed to a substantial risk of violence and sexual assault.” Id. at *17. But as discussed, the record does not support the existence of such a policy. See supra II.A.1. Wal- 15 26-4872 Mart Stores, Inc. v. Dukes, 564 U.S. 338 (2011), is on point. There, the Supreme Court rejected certification of a class of female employees who purportedly suffered gender discrimination at the hands of store managers. Id. at 344–46, 367. The plaintiffs could not show by a preponderance that the company had a policy of discrimination that would justify the class; what remained were individual allegations of discrimination insufficient to satisfy commonality. Id. at 350–51, 353–55. The same defect pertains here—Plaintiffs have not shown a Department- wide policy of deliberate indifference to justify a class; scattered claims of harm, however severe, will not do. Because the class must be decertified, the class-wide injunction necessarily falls as well. B. While Plaintiffs’ failure to establish a likelihood of success is enough to warrant vacatur of the preliminary injunction, the remaining Winter factors reinforce that conclusion. See Azar, 911 F.3d at 575. Consider first irreparable harm. Plaintiffs’ allegations of harm, even assuming its irreparability, reflect individual harm and do not justify class-wide relief. Meanwhile, the district court’s class-wide injunction would continue to work irreparable harm on Defendants. As the Supreme Court has acknowledged, “it is ‘difficult to imagine an activity in which a State has a stronger interest, or one that is more intricately bound up with state laws, regulations, and procedures, than the administration of 16 26-4872 its prisons.’” Woodford v. Ngo, 548 U.S. 81, 94 (2006) (citation omitted). The State’s interest here is severely disrupted by the district court’s overhaul of Oregon’s prison system. And by imposing a presumption that “transgender women” inmates be assigned to a women’s prison, the district court’s injunction not only displaces the State’s role as prison administrator—it compromises public safety too. The balance of the equities and the public interest thus sharply favor Defendants. Under the district court’s order, more than a hundred inmates currently in men’s prisons could be channeled to and placed in the only state-run women’s prison in Oregon. Those inmates include Plaintiff J.F., who sexually assaulted a woman by lodging a footlong plastic object into that woman’s vagina, and other inmates convicted of violent crimes. 2-ER-240. The district court allows for the presumption to be rebutted by “a documented and articulable security justification” (1-ER-3), but it apparently believed that, in J.F.’s case, no such justification existed. Remarkably, the district court stated that “no explanation has been provided as to why J.F.’s violent history requires [J.F.’s] continued housing in a men’s prison.” S.D., 2026 WL 1146591, at *12. If J.F.’s record does not rebut the district court’s presumption, it is hard to see what would. Consider then the order’s implications: Against the State’s deliberate judgment, the district court could force Oregon to transfer the named Plaintiff to a 17 26-4872 women’s prison because Plaintiff now identifies as a woman—an individual who no one disputes was convicted of perpetrating sexual violence against a woman in a manner most unconscionable. Plainly, that is not in the public interest. And because the district court found that Plaintiff served as an adequate class representative whose claims are typical of the other class members, hundreds of other class members (many with histories of violence against women) could be ordered transferred too. The equities and public interest do not justify that result. * * * The law and facts do not clearly favor the district court’s order here. Nor do the other Winter factors support it. Accordingly, we vacate the district court’s preliminary injunction and remand with instructions to decertify the class. Given the risk that delay could moot the appeal, see Ahlman v. Barnes, 20 F.4th 489, 493–95 (9th Cir. 2021); Norbert v. City & Cnty. of San Francisco, 10 F.4th 918, 926–27 (9th Cir. 2021), the mandate shall issue forthwith. See Fed. R. App. P. 2; Ninth Cir. Gen. Order 4.6.b. PRELIMINARY INJUNCTION VACATED; CLASS DECERTIFIED; CASE REMANDED; MANDATE ISSUED FORTHWITH. 18 26-4872