Appeal from the United States District Court for the District of Maryland, at Baltimore. Ellen Lipton Hollander, Senior District Judge. (1:25-cv-00596-ELH)

Argued: September 11, 2025 Decided: April 10, 2026

Before DIAZ, Chief Judge, and WILKINSON, NIEMEYER, KING, GREGORY, AGEE, WYNN, THACKER, HARRIS, RICHARDSON, QUATTLEBAUM, RUSHING, HEYTENS, BENJAMIN, and BERNER, Circuit Judges.

Preliminary injunction vacated by published opinion. Judge Heytens announced the judgment of the court and delivered the opinion of the court with respect to Parts I, II, and III, which Chief Judge Diaz and Judges King, Gregory, Wynn, Thacker, Harris, Benjamin, and Berner joined, and an opinion with respect to Part IV, which Chief Judge Diaz and Page 2Judge Harris joined. Judge Wilkinson wrote an opinion concurring in the judgment, which Judges Niemeyer, Agee, and Rushing joined. Judge Richardson wrote an opinion concurring in the judgment, which Judges Wilkinson, Niemeyer, Agee, Quattlebaum, and Rushing joined. Judge Quattlebaum wrote an opinion concurring in the judgment, which Judges Richardson and Rushing joined. Judge King wrote an opinion concurring in part, dissenting in part, and dissenting from the judgment, which Judges Gregory, Wynn, Thacker, Benjamin, and Berner joined. Judge Wynn wrote an opinion, which Judges King, Thacker, Benjamin, and Berner joined.

ARGUED: Jack E. Starcher, UNITED STATES DEPARTMENT OF JUSTICE, Washington, D.C., for Appellants. Alethea Anne Swift, DEMOCRACY FORWARD FOUNDATION, Washington, D.C., for Appellees. ON BRIEF: Yaakov M. Roth, Principal Deputy Assistant Attorney General, Eric D. McArthur, Deputy Assistant Attorney General, Gerard Sinzdak, Simon Jerome, Jacob Christensen, Civil Division, UNITED STATES DEPARTMENT OF JUSTICE, Washington, D.C., for Appellants. Brian A. Sutherland, Anna-Rose Mathieson, COMPLEX APPELLATE LITIGATION GROUP LLP, San Francisco, California; Mark B. Samburg, Emma R. Leibowitz, Simon C. Brewer, Robin F. Thurston, DEMOCRACY FORWARD FOUNDATION, Washington, D.C., for Appellees.

Page 3TOBY HEYTENS, Circuit Judge:

Three organizations sued to stop the Social Security Administration from giving U.S. DOGE Service personnel access to sensitive personal information about millions of Americans. The district court granted a preliminary injunction, which the Supreme Court stayed pending this appeal and any further Supreme Court review. We now vacate the current preliminary injunction and return the case to the district court for further proceedings.

It would be hard to participate in American life without interacting with the Social Security Administration. The agency issues Social Security numbers and oversees retirement, disability, and survivor benefits for nearly 72 million people. To do so, it collects and retains sensitive information about nearly everyone in the United StatesSocial Security numbers, citizenship status, birth dates, bank account numbers, tax information, medical history, and more. For decades, the agency guarded that data scrupulously, and Americans trusted it to do so.

On January 20, 2025, the President signed an executive order creating the U.S. DOGE Service and charging it with making government technology more efficient. See Exec. Order No. 14,158, 90 Fed. Reg. 8441 (Jan. 20, 2025). DOGE personnel quickly made their way to the Social Security Administration and sought “unprecedented” access to agency systems, including non-anonymized personally identifiable information. JA 103-04, 107-08. A standoff ensued, and career officials resigned. A new acting administrator was installed and granted DOGE the sweeping access it sought.

Page 4Three organizations representing a combined seven million Americans sued to prevent DOGE from accessing their members’ personally identifiable information. When the case was filed and in the original preliminary injunction proceedings, plaintiffs’ theory of the case was not that DOGE had misused the information or disclosed it (accidentally or otherwise) to malicious actors. Instead, plaintiffs argued that handing over non-anonymized and highly sensitive information to DOGE was itself unlawful.

After hours of hearings and hundreds of pages of analysis, the district court issued the preliminary injunction we review here. The Supreme Court stayed that preliminary injunction and directed that the stay would remain in effect until the completion of all appellate review—including by the Supreme Court—of the district court’s order. 145 S. Ct. 1626, 1626 (2025). We have jurisdiction to review the district court's order under 28 U.S.C. § 1292(a)(1), and we review the grant of a preliminary injunction for abuse of discretion, see Ashcroft v. ACLU , 542 U.S. 656, 664 (2004). As always, that standard is “necessarily” satisfied if the district court’s ruling is “based . . . on an erroneous view of the law.” Cooter & Gell v. Hartmarx Corp. , 496 U.S. 384, 405 (1990).

“A preliminary injunction is an extraordinary remedy never awarded as of right.” Winter v. Natural Res. Def. Council, Inc. , 555 U.S. 7, 24 (2008). To obtain a preliminary injunction, a plaintiff “must establish that he is likely to succeed on the merits, that he is likely to suffer irreparable harm in the absence of preliminary relief, that the balance of equities tips in his favor, and that an injunction is in the public interest.” Id. at 20. The Supreme Court has condemned efforts to replace these 'well-established principles of Page 5equity' with 'broad classifications' about when injunctions may issue. eBay Inc. v. MercExchange, L.L.C. , 547 U.S. 388, 391–93 (2006).

Language in this Court’s recent decision in American Federation of Teachers v. Bessent ( AFT ), 152 F.4th 162 (4th Cir. 2025), can be read as departing from these principles.¹ In some situations-those 'when a plaintiff must prevail on several independent issues' to win its case-the AFT opinion describes “a multiplicative problem” where the plaintiff’s “overall” likelihood of success “is the product of [its] probability of success on each of the independent, dispositive issues.” Id. at 169-70 (quotation marks removed). As the 'probabilities are multiplied,' AFT continues, 'their product shrinks rapidly,” a point the opinion illustrates with “[a]n example” where a plaintiff has “a 75% chance of prevailing on five independent issues' and must 'prevail on all of them to receive relief.” Id. at 170 & n.4. In such situations, states AFT , “[t]he plaintiff must . . . show an extremely high likelihood of success on each individual issue in order to have a normal likelihood of success overall.' Id. at 170 (emphasis added).

Today, we disavow any suggestion that district courts should assign numerical probabilities to a plaintiff's chances of success on each issue and then multiply those Page 6probabilities together to determine whether the plaintiff’s “overall odds” of success are high enough to warrant a preliminary injunction. AFT , 152 F.4th at 170. We have no quarrel with the general proposition that it can be harder to win a complex case with multiple issues than a straightforward case that turns on a single issue. But the traditional four-factor test for granting a preliminary injunction is a creature of equity, not 'a mechanical algorithm,' and attempts to reduce it to one risk 'confusion worse confounded.' Delaware State Sportsmen’s Ass’n v. Delaware Dep’t of Safety & Homeland Sec. , 108 F.4th 194, 203 (3d Cir. 2024) (first quote), cert. denied sub nom. Gray v. Jennings , 145 S. Ct. 1049 (2025); Shady Grove Orthopedic Assocs., P.A. v. Allstate Ins. Co. , 559 U.S. 393, 404 (2010) (quotation marks removed) (second quote).

AFT suggests district courts should perform unfamiliar tasks for dubious benefits. The law often deals in probabilities that are 'incapable of precise definition or quantification into percentages,' Maryland v. Pringle , 540 U.S. 366, 371 (2003) (discussing probable cause), and we doubt the value of asking district courts to spend time pondering whether a plaintiff has a 75% (versus a 70% or 80%) chance of prevailing on a given issue. See Kevin M. Clermont, A Theory for Evaluating Evidence Against the Standard of Proof , 127 Penn. St. L. Rev. 345, 367 (2023) (noting that '[c]ognitive limitations leave humans able only weakly to judge likelihood on any sort of scale”).² In addition, AFT never defines what it means for two issues to be “independent” in a legal or Page 7mathematical sense (see 152 F.4th at 167, 169–70), nor does it address how district courts should deal with the familiar statistical problem known as conditional probability.³ What should district courts do when the parties disagree about whether (and if so, how) the plaintiff’s success on one issue would impact its odds of success on another? Are some arguments so weak that a court may definitively reject them without asking how they impact the plaintiff’s “overall odds” of prevailing in the suit? AFT , 152 F.4th at 170. And what about novel legal issues or those of first impression, where a district court may struggle to identify the precise probability of one outcome or another?

Better, we think, to stick with the traditional approach. Consistent with Winter , we reiterate that plaintiffs seeking a preliminary injunction must show they are likely to succeed on the merits of their lawsuit. Plaintiffs need not clear a different or additional hurdle in cases involving multiple issues or defenses. All statements to the contrary in AFT are abrogated. See McMellon v. United States , 387 F.3d 329, 332 (4th Cir. 2004) (en banc) (expressing “no doubt” that the en banc court may “overrule” an earlier panel’s decision in a different case).

A district court may not grant a preliminary injunction-or any other form of relief—unless the plaintiff has sufficiently demonstrated Article III standing. See Murthy Page 8v. Missouri , 603 U.S. 43, 49–50 (2024) (holding that a court of appeals erred by affirming a preliminary injunction because the plaintiffs lacked standing). To establish standing, a plaintiff must show: “(i) that he suffered an injury in fact that is concrete, particularized, and actual or imminent; (ii) that the injury was likely caused by the defendant; and (iii) that the injury would likely be redressed by judicial relief.' TransUnion LLC v. Ramirez , 594 U.S. 413, 423 (2021). “At the preliminary injunction stage,” a “plaintiff must make a clear showing that she is likely to establish each element of standing.” Murthy , 603 U.S. at 58 (quotation marks removed). Before entering a preliminary injunction, the district court concluded that plaintiffs had sufficiently established standing to pursue at least some of their claims. Reviewing that issue de novo, see South Carolina v. United States , 912 F.3d 720, 726 (4th Cir. 2019), we agree.⁴

The dispute before us focuses on 'the first and foremost of standing's three elements”: injury in fact. Spokeo, Inc. v. Robins , 578 U.S. 330, 338 (2016) (brackets and quotation marks removed). “[U]nder Article III, an injury in law is not an injury in fact.” TransUnion , 594 U.S. at 427. Rather, to establish Article III injury, a plaintiff must plead and prove that it has suffered (or will imminently suffer, absent a court’s intervention) a concrete injury. See id. An injury is 'concrete' if it 'has a close relationship to a harm traditionally recognized as providing a basis for a lawsuit in American courts.” Id. at 424 (quotation marks removed).⁵

Like the district court, we conclude that disclosing plaintiffs’ members’ personally identifiable information to DOGE inflicts a harm that is a 'close . . . analogue' to the common law tort of intrusion upon seclusion. TransUnion , 594 U.S. at 424–25 (identifyingPage 10 “intrusion upon seclusion” as a harm “traditionally recognized as providing a basis for lawsuits in American courts” and citing Gadelhak v. AT&T Servs., Inc. , 950 F.3d 458, 462 (7th Cir. 2020), an opinion by then-Judge Barrett that held a plaintiff who received unwanted text messages had standing); see JA 1352–83. We thus abrogate our Court’s contrary conclusion in AFT . See 152 F.4th at 171-74 (concluding that plaintiffs in an analogous case “seemingly lack standing”).

Intrusion upon seclusion is an “intentional[] intru[sion], physical[] or otherwise, upon the solitude or seclusion of another or his private affairs or concerns” that “would be highly offensive to a reasonable person.' Restatement (Second) of Torts § 652B (A.L.I. 1977) (Second Restatement); see Gadelhak , 950 F.3d at 462 (citing the same Restatement); Gov’t Reply Br.⁴ (agreeing the Court should use “the formulation of the tort given in the Restatement (Second) of Torts'). The tort is not limited to entering someone’s house or peering through their windows. Rather, it includes “other form[s] of investigation or examination” of “private concerns,” including opening someone’s mail, going through their wallet, examining their bank account, or 'compelling [them] by a forged court order to permit an inspection of [their] personal documents.' Second Restatement § 652B cmt. b. Intrusion upon seclusion does not require the tortfeasor to go on to misuse the information it learns or disseminate it to the public; instead, the unjustified intrusion upon the plaintiff’s privacy is the harm. See § 652B cmts. a & b.; Persinger v. Southwest Credit Sys., L.P. , 20 F.4th 1184, 1191–92 (7th Cir. 2021).

Much like rifling through someone’s wallet, bank account, or personal documents, granting unauthorized and unwarranted access to a person’s sensitive personal information Page 11is an intentional intrusion into “private affairs or concerns.” Second Restatement § 652B. And that is what plaintiffs say happened here. According to plaintiffs, the Social Security Administration handed over their members’ most sensitive personal data to people (DOGE team members) who were not legally authorized to access it. Whether plaintiffs “would prevail in a lawsuit for common law invasion of privacy is irrelevant.' Persinger , 20 F.4th at 1192. Instead, it is enough that the injury plaintiffs have identified “pose[s] the same kind of harm that common law courts recognize.” Gadelhak , 950 F.3d at 463.⁶

The government and AFT identify various counterarguments. We are not persuaded by any of them.

The government’s lead argument is that the presence or absence of an Article III injury cannot turn on which government employees access plaintiffs’ members’ personally identifiable information. As the government points out, plaintiffs raise no challenge to non-DOGE Social Security Administration employees accessing plaintiffs' members' sensitive personal data. And if plaintiffs have no problem with that, the government asserts, Page 12their members cannot suffer concrete harm for Article III purposes when DOGE employees do the same.

That argument suffers from a familiar flaw: confusing a possible “weakness on the merits with [an] absence of Article III standing.' Davis v. United States , 564 U.S. 229, 249 n.10 (2011). “For standing purposes, we accept as valid the merits of [plaintiff’s] legal claims . . . .' FEC v. Cruz , 596 U.S. 289, 298 (2022). And here, plaintiffs' merits theory is that DOGE team members—unlike other Social Security Administration employees—are not authorized to access this sensitive personal information, which is what renders such access an unlawful and highly offensive 'intrusion.' See DOJ v. Reporters Comm. for Freedom of the Press , 489 U.S. 749, 763–64 (1989) (information may be “private” if it is “intended for or restricted to the use of a particular person or group or class of persons,” because “the extent of the protection accorded a privacy right at common law rested in part on the degree of dissemination of the allegedly private fact' (quotation marks removed)).

Consider the following hypothetical. A psychiatrist asks a patient to record their deepest, darkest thoughts and fears in a journal. As part of their treatment, the patient gives the psychiatrist the journal to review. Without authorization from the patient or any other source of law, the psychiatrist intentionally gives the journal to a janitor who works for the same office as the psychiatrist. The janitor reads it. Has the patient suffered a sufficiently concrete injury to support Article III standing? Absolutely. Despite the government's contrary insistence at oral argument, it matters a great deal—both in life and in law— who in the office reads the patient’s journal. And under plaintiffs’ merits theory (which, again, we must assume is valid at this stage of the analysis), DOGE is the nosy janitor.

Page 13The government also insists this theory of injury “closely resembles the harms the Supreme Court deemed insufficiently concrete in TransUnion .” Gov’t Br. 25. Once again, we disagree.

The TransUnion plaintiffs relied on an analogy to a different common law tort: defamation. See 594 U.S. at 432. As the Court noted, '[p]ublication is 'essential to liability’ in a suit for defamation.” Id. at 434 (quoting Restatement of Torts § 577 cmt. a. (A.L.I. 1938)). The bulk of the TransUnion plaintiffs therefore lacked standing because the allegedly 'inaccurate information' in the defendant's 'internal credit files' was not disclosed to any third party and the plaintiffs “did not factually establish a sufficient risk” of a future disclosure 'to support Article III standing.' Id. at 434-35, 437-38.

The case before us differs in nearly every relevant respect. Plaintiffs rely on a different common law analogy-the tort of intrusion upon seclusion. See TransUnion , 594 U.S. at 425 (identifying 'intrusion upon seclusion' as an 'example' of a 'harm[] traditionally recognized as providing a basis for lawsuit in American courts'). It is black-letter law that intrusion upon seclusion, unlike defamation, has no publication requirement, nor does it require that the underlying information be false or defamatory. Rather, it is '[ t ] he intrusion itself [that] makes the defendant subject to liability, even though there is no publication” or disclosure to any third party. Second Restatement § 652B cmt. b (emphasis added).

AFT identified two other reasons that plaintiffs in a similar case could not analogize to intrusion upon seclusion. First, the opinion stated that, for purposes of that tort, “it is not the information obtained, but the knowledge that the third party is engaged in targeted Page 14snooping, that causes the harm.' AFT , 152 F.4th at 172. Second, the opinion reasoned that plaintiffs had not been “the subjects of targeted ‘investigation[s] or examination[s] into [their] private concerns,'' id. (quoting Second Restatement § 652B cmt. b), because “[e]ach [p]laintiff’s information [was] one row in various databases that [were] millions upon millions of rows long,” id. at 172.

With respect, we disagree. For one thing, none of the authorities cited in AFT say that it is the plaintiff’s knowledge rather than the defendant’s actions that constitute the relevant harm for purposes of intrusion upon seclusion. To the contrary, the relevant Restatement provision provides six examples where the defendant 'has invaded' the plaintiff's privacy. Second Restatement § 652B, cmt. b, illus. 1-5; cmt. c, illus. 7. In two-thirds of those examples (Illustrations 2, 3, 4, and 7), it is never stated—and far from obvious—that the plaintiff is aware of the defendant’s wrongful actions.⁷

In addition, neither the Restatement’s text, nor its comments, nor its illustrations say the snooping must be 'targeted' in the sense that AFT used that term. To be sure, the Restatement requires a defendant to have “ intentionally intrud[ed] . . . upon the solitude or seclusion of another,' Second Restatement § 652B (emphasis added), and all the accompanying illustrations involve a single defendant violating the privacy of a single plaintiff. But AFT never explains why invading the privacy of millions of people isPage 15 “different in kind” from invading the privacy of just one—much less in a way that helps the defendant . 152 F.4th at 172. True, this case does not involve 'reporters, detectives, and paparazzi.' Id. But here, as elsewhere, a history-and-tradition approach asks us to identify the relevant “principles” rather than to search for “a dead ringer or a historical twin.” United States v. Rahimi , 602 U.S. 680, 692 (2024) (quotation marks removed). We thus conclude that plaintiffs have carried their burden of showing they are likely to establish each element of standing. See Murthy , 603 U.S. at 58.

Having satisfied ourselves that the district court had subject matter jurisdiction, we ask whether it exceeded its discretion by entering the preliminary injunction under review here. We conclude that plaintiffs did not establish—based on the record then before the district court-that they satisfied the second Winter factor. We thus vacate the current preliminary injunction without addressing the remaining factors. See Winter , 555 U.S. at 23–24.⁸

Although Winter 's second factor is sometimes shorthanded as 'irreparable injury,'Page 16 see, e.g. , Grupo Mexicano de Desarrollo, S.A. v. Alliance Bond Fund, Inc. , 527 U.S. 308, 315 (1999), that brief description does not tell the whole story. The role of a preliminary injunction is to protect the plaintiff from suffering new or additional irreparable harm between the time the preliminary injunction is entered and the case's final resolution. See University of Texas v. Camenisch , 451 U.S. 390, 395 (1981) ('The purpose of a preliminary injunction is merely to preserve the relative positions of the parties until a trial on the merits can be held.”). And, like any other injunction, a preliminary injunction cannot reach back in time to prevent or undo irreparable harm that has already occurred. See, e.g. , City of Los Angeles v. Lyons , 461 U.S. 95, 102–03, 105 (1983). For those reasons, the key question here is whether a preliminary injunction will prevent plaintiffs from suffering new or additional irreparable harm while they litigate this case to final judgment.

What is more, not all harm-no matter how serious in the real world-is “irreparable” in a legal sense. Even a plaintiff who has a concrete injury for Article III standing purposes may only obtain the “extraordinary remedy” of a preliminary injunction Page 17if that injury cannot be remedied during the ordinary course of litigation. Winter , 555 U.S. at 24; see, e.g. , Di Biase v. SPX Corp. , 872 F.3d 224, 232-33, 235 (4th Cir. 2017) (concluding plaintiffs had standing but not irreparable harm); Moms for Liberty v. Wilson Cnty. Bd. of Educ ., 155 F.4th 499, 512-14 (6th Cir. 2025) (same); Alliance for Retired Ams. v. Bessent , 770 F. Supp. 3d 79, 107-08 (D.D.C. 2025) (same). If there is even a “possibility” that “adequate compensatory or other corrective relief will be available at a later date,” a court should hesitate before concluding a plaintiff’s harm is “irreparable” for purposes of granting a preliminary injunction. Sampson v. Murray , 415 U.S. 61, 90 (1974) (quotation marks removed).

Applying those standards here, we conclude plaintiffs have not satisfied Winter 's second factor.

We start by identifying the only theory of irreparable harm that is properly before us. The district court did not grant this preliminary injunction on the theory that plaintiffs’ members would be harmed by some downstream misuse or public disclosure of their personal data. See supra note 8. Instead, the irreparable harm on which the district court relied mirrors plaintiffs' Article III injury: that DOGE violates plaintiffs' members' privacy by accessing their sensitive personal data without lawful authorization, and that privacy violation is “irreparable” in a legal sense.

The difficulty with that argument is that there are two forms of corrective relief that may be available down the line: money damages and a reparative permanent injunction.

To begin, the Privacy Act-one of the statutes plaintiffs accuse defendants of Page 18violating—authorizes damages for “intentional or willful” violations.⁵ U.S.C. § 552a(g)(4). And if “harm suffered can be remedied by money damages at the time of judgment,” a plaintiff “must overcome the presumption that a preliminary injunction will not issue.' Di Biase , 872 F.3d at 230.

That presumption stands unrebutted here. Plaintiffs insist that any damages would be 'insufficient' and 'difficult to ascertain.' Pls.' Br. 61 (quoting Multi-Channel TV Cable Co. v. Charlottesville Quality Cable Operating Co. , 22 F.3d 546, 551–52 (4th Cir. 1994) (quotation marks removed)). But plaintiffs make little effort to explain why that is so. Nor do plaintiffs explain why two district court judges—in similar cases involving DOGE’s access to government databases—erred in concluding that those plaintiffs did not satisfy the second Winter factor because damages could be available later. See University of Cal. Student Ass’n v. Carter , 766 F. Supp. 3d 114, 123 (D.D.C. 2025); American Fed'n of Lab. & Cong. of Indus. Orgs. v. Department of Lab. , No. 25-cv-339, 2025 WL 1783899, at *14 (D.D.C. June 27, 2025). Plaintiffs are right that 'those decisions concerned different agencies, different systems of record, and different allegations and evidence.” Pls.’ Br. 64. But plaintiffs never explain why those differences (real as they may be) have anything to do with the legal question of whether the availability of damages prevents plaintiffs from establishing the sort of irreparable harm Winter requires.

To the extent that damages alone might not be enough, plaintiffs also fail to explain why a reparative permanent injunction would be legally inadequate to remedy any lingering harm. The district court entered its preliminary injunction based on a record that, at the time, stated that the allegedly unlawful disclosure had been made to a small group of Page 19people within the government. See JA 1334 (district court describing “[t]he SSA DOGE Team' as 'five special government employees and six detailees from other federal agencies'). On that record, the district court could have—if plaintiffs ultimately prevailed on the merits—ordered the relevant employees to destroy any illegally obtained data or work derived from such data. See, e.g. , Church of Scientology of Cal. v. United States , 506 U.S. 9, 13 (1992) (noting that a court could “order[] the Government to destroy or return any and all copies” of unlawfully obtained material); see also Restatement (Third) of Torts: Remedies § 44 cmt. c (A.L.I., Tentative Draft No. 2, 2023) (explaining that a reparative injunction orders a defendant “to reverse or undo all or part of the violation of plaintiff’s rights, or to reverse or undo some or all of the harm” but “still looks to the future, by preventing the future bad effects of [the defendant’s] past violations”). That is the theory on which a third district court judge concluded the plaintiffs in yet another similar case failed to show irreparable harm. See Alliance for Retired Ams. , 770 F. Supp. 3d at 108. Once again, plaintiffs fail to explain why that logic is faulty as a legal matter.

Finally, the elephant in the room. As noted earlier, the role of a preliminary injunction is to protect a plaintiff from suffering further irreparable harm while the case is litigated. But because of the particular procedural posture of this case, the district court’s preliminary injunction cannot currently protect anyone from anything and no decision we issue today has the power to change that fact. The Supreme Court stayed the preliminary injunction, and the preliminary injunction will resume in force—if ever—only after events that would post-date our decision and over which we have no control. See 145 S. Ct. at 1626 (directing that the stay will remain in effect “pending the disposition of the appeal Page 20in the United States Court of Appeals for the Fourth Circuit and disposition of a petition for a writ of certiorari, if such a writ is timely sought. ' (emphasis added)). Although plaintiffs filed their brief after the Supreme Court's stay order, that briefing does not grapple with whether (and if so, how) that order impacts their ability to satisfy Winter 's second factor.

For these reasons, we hold that plaintiffs did not show—based on the record before the district court when it entered this preliminary injunction-that they were 'likely to suffer irreparable harm in the absence of preliminary relief.' Winter , 555 U.S. at 20 (emphasis added). We thus vacate the preliminary injunction without addressing whether plaintiffs satisfy the remaining three Winter factors.

* * *

The order granting a preliminary injunction is vacated and the case is returned to the district court for further proceedings consistent with this opinion.

SO ORDEREDPage 21 WILKINSON, Circuit Judge, with whom Judges NIEMEYER, AGEE and RUSHING join, concurring in the judgment:

I concur in the judgment and believe that the Supreme Court’s issuance of the stay in this case is by far the most salient factor dictating our denial of the preliminary injunction. To conclude otherwise would relegate the Supreme Court's stay order to a shallow exercise. That is a course only the most audacious inferior court would seek to follow.

Here the parties are the same as in the case before the Supreme Court. They are still disputing the same matter. Indeed, this present case lies in direct sequence with the Court’s stay proceedings. While my friend Judge Wynn laments “a system in which unexplained orders silently control future cases,” this is not one of those “future cases,” but the very same case as that before the Court. Wynn Op at 79 n.4. The Supreme Court’s stay would only have been issued if the Court believed that the Nken factors were met. At the preliminary-injunction phase, we now face the nearly identical Winter factors. Indeed, both the Nken and Winter factors examine likelihood of success on the merits. Nken v. Holder , 556 U.S. 418, 426 (2009); Winter v. Nat. Res. Def. Council, Inc. , 555 U.S. 7, 20 (2008). In the present case, the Supreme Court has determined that this likelihood does not exist and/or that the equities lie heavily in the government’s favor; we cannot come to a different conclusion at this stage without directly contravening the Court’s order. If we heard this case on the merits, we might find ourselves in a different position. Probabilistic determinations need not be dispositive of the merits themselves, and thus interim orders Page 22have only persuasive authority in a decision on the final judgment. See Trump v. Boyle , 145 S. Ct. 2653, 2654 (2025).¹ I believe Judge Richardson's opinion ably captures these implications of the current procedural posture, and, for that reason, I concur in it.

I write separately, however, to emphasize that we apply interim orders just as we would any of the Court’s precedents in the normal course of business. Under any theory of stare decisis , the precedential weight of a Supreme Court opinion is inextricably linked to the factual and legal similarity between the past and present. See Epic Sys. Corp. v. Lewis , 584 U.S. 497, 510 (2018) (“The law of precedent teaches that like cases should generally be treated alike . . . .'). The operative question-interim order or not-is whether a precedent sits too far afield of the case at hand so as to be distinguishable. When a case is a perfect mimicry of that before the Supreme Court, then the answer is obvious. Just as identical DNA sequences produce identical twins, identical circumstances should produce identical judicial dispositions.

Presently, we face the legal and factual twin of the Supreme Court’s order. That counsels following that very order. I acknowledge that a correction to the record after the en banc argument has generated some disagreement as to the facts before our court, but, as my colleague Judge Heytens rightly points out, when reviewing a district court’s grant of a preliminary injunction, we look only to facts before the district court at the time of that Page 23decision. Heytens Op. at 15–16 n.8. The new revelations here were not part of the district court’s preliminary-injunction analysis, and thus these new facts should not and cannot become part of our review. The information might become relevant upon a future appeal if the district court amended its preliminary injunction or issued a new preliminary or permanent injunction upon remand. But right now, the relevant record is only that which the district court considered below-the uncorrected record identical to that before the Supreme Court.²

Moreover, as aforementioned, we operate under a parallel legal standard to the one the Court employed in its stay order.³ The binding authority of that order is thus easy to Page 24see in the present circumstances. However, whether the Supreme Court’s interim order in one discrete case should always bind an appellate court's decision in a separate case involving arguably similar facts and legal issues seems a broader proposition that we need not embrace. We cannot forecast all future circumstances, and the applicability of precedent best remains a case-by-case determination.

Even so, this decisional power must not be used to disrupt the vertical hierarchy of Article III courts. Procedural posture notwithstanding, we cannot ascribe to a Supreme Court decision no significance whatsoever. Interim orders are not scrimmages; they are real proceedings with legal effect, and they may have appreciable bearing in such postures as we now find ourselves. And because this case offers no true point of distinction from the Supreme Court’s stay proceedings, we must afford its order proper binding weight in our preliminary-injunction analysis.

I regret my dear colleague Judge Wynn’s rhetorical assault upon the Supreme Court and my friend Judge King’s support of it. It is one thing to regret “emergency motions made under intense time pressure.' Wynn Op. at 79. It is quite another to repeatedly lambast the Court for the “unexplained and summary nature” of its orders and to declare that 'interim orders announced without reasons can just as easily be ignored without explanation,” which more than hints that this and future courts would be willing to do so. Id. at 80; see also King Op. at 76–77. And while Judge Wynn states that “a stay may be a strong signal” as to the ultimate outcome on the merits, his opinion then proceeds to ignore that signal in this very case. Wynn Op. at 87. My brothers King and Wynn have delivered Page 25a stern warning to the Supreme Court. This creeps too near the water’s edge of defiance for my comfort.

To be sure, in an ideal world, every ruling would be accompanied by full briefing, oral argument, lengthy and reasoned opinions, and the like. To truncate or dispense with that process on too frequent a basis would indeed reflect an arbitrariness that risks public faith in the judicial process. But there is another side to this. The Supreme Court is the one judicial body that can establish a degree of uniformity in the application of what is, after all, our national law. And if the Court senses something is very wrong, it cannot then be right to just let it go. Or if the rulings of lower courts are not only disparate but chaotic, is the Court just to shrug it away? Better by far to issue a stay with the expectation that its assessment will have at least some effect in those instances where cases are factually and legally similar. The stay holds matters in place, gives lower courts the chance to analyze and reflect, and achieves a measure of uniformity pending the ultimate decision on the merits. It is a useful device when used sparingly, and if the danger of overuse is arbitrariness, the risks of underuse are anarchic.

My colleagues jump too quickly onto the bandwagon of those who condemn each and every use of the Supreme Court’s emergency docket, no matter how justified. The unfortunate effect of Judge King and Wynn’s approach will be to aggrandize the role of the lower federal courts at the expense of the Supreme Court’s own place in a hierarchical judicial system. And here, I repeat that the striking similarities in the legal analysis before the Supreme Court and our own require adherence to the stay order's necessary implications. This conclusion does not, as my colleagues insinuate, raise the specter of an Page 26absolute rule prohibiting us from ever interpreting the Supreme Court’s interim orders. As with any issue of precedential weight, the consideration for interim decisions is simply one of balance. Balance between respecting the Court's pronouncements and crediting any unique circumstances of the case before us. Here, where those circumstances are lacking, the arrow points unmistakably to our responsibility to follow the Supreme Court. It matters not that we might have decided the issue differently. A certain modesty is required of our station.

Page 27RICHARDSON, Circuit Judge, with whom WILKINSON, NIEMEYER, AGEE, QUATTLEBAUM, and RUSHING, Circuit Judges, join, concurring in the judgment:

Article III of the Constitution sets up two categories of courts: “one supreme Court” and other “inferior Courts.” U.S. Const. art. III, § 1. This Court is an inferior one. This judicial hierarchy commands vertical stare decisis-when the Supreme Court speaks, inferior courts must listen. Of course, many cases feature novel legal or factual issues that require inferior courts to exercise independent judgment. But in other cases, the Supreme Court makes the answer easy. This is one such case.

In this appeal, we review a district court’s grant of a preliminary injunction against a government agency. The merits involve several interesting-and challenging-legal issues. But the outcome of this appeal should be neither interesting nor challenging. That’s because the Supreme Court already answered the interim question before us in this very case. SSA v. Am. Fed’n of State, Cnty., & Mun. Emps. , 145 S. Ct. 1626, 1626 (2025). So this case is “squarely controlled” by the Supreme Court’s earlier interim order granting a stay. Trump v. Boyle , 145 S. Ct. 2653, 2654 (2025). Following the Supreme Court's direction, I would vacate the preliminary injunction.

Fortunately, our en banc Court reaches that result today. But along the way, the Court makes several unforced errors.

This opinion proceeds in four parts. I first describe this case’s background. Second, I explain why the Supreme Court’s interim order in this case controls the outcome here. Third, I reiterate that judges are not exempt from everyday truths about statistical probability, even when acting in equity to assess a party’s likelihood of success. Fourth, IPage 28 conclude that Plaintiffs are not likely to succeed on the merits because they have not shown standing or irreparable harm.

When President Trump took office, he laid out an ambitious agenda to improve governmental efficiency. As part of that agenda, he established the Department of Government Efficiency (“DOGE”), which he tasked with “modernizing Federal technology and software to maximize governmental efficiency and productivity.' See Exec. Order No. 14,158, 90 Fed. Reg. 8441 (Jan. 20, 2025). President Trump directed agency leaders across the federal government to establish “DOGE Teams” within their agencies and-'to the maximum extent consistent with law'-to provide DOGE with access to agency data systems. Id. at 8441-42.

In compliance with this presidential directive, the Social Security Administration established an internal DOGE Team.¹ The SSA DOGE Team consists of eleven individuals with various technical specialties. These DOGE Team members were onboarded as SSA employees and received the same level of privacy and ethics training as other SSA employees. Eventually, SSA granted DOGE Team members the same level of access that dozens of other SSA employees possess—access to various internal records containing personally identifiable information, including Social Security numbers, bank information, medical records, and addresses.

Page 29Frustrated with this development, three membership organizations sued to block the DOGE Team from accessing SSA systems, claiming violations of the Privacy Act and the Administrative Procedure Act (“APA”). The district court granted a preliminary injunction barring DOGE Team members from accessing personally identifiable information and requiring destruction of any such information already obtained.

In a stark departure from our established procedure, our Court sua sponte ordered initial hearing en banc to consider the government's motion to stay the preliminary injunction. Am. Fed'n of State, Cnty. & Mun. Emps. v. SSA , No. 25-1411, 2025 WL 1249608, at *6 (4th Cir. Apr. 30, 2025) (Richardson, J., dissenting). When a majority of our en banc Court voted against a stay, the Supreme Court stepped in to grant the stay. SSA v. Am. Fed’n of State, Cnty. & Mun. Emps. , 145 S. Ct. 1626, 1626 (2025). We now review the government’s direct appeal of that same preliminary injunction.

In recent years, the Supreme Court has often addressed consequential issues through its emergency docket. Through interim orders, the Court makes initial judgments disposing of applications for temporary relief. Because these orders are issued on tight schedules and often feature less discussion than opinions from the Court’s merits docket, scholars and lower courts long wondered whether these interim orders were binding in other cases.

Over the years, some scholars suggested thoughtful frameworks for assessing an interim order’s precedential value. See, e.g. , Trevor N. McFadden & Vetan Kapoor, The Precedential Effects of the Supreme Court’s Emergency Stays , 44 Harv. J.L. & Pub. Pol’y 827 (2021). And members of our Court have debated whether Supreme Court stays should Page 30affect our judgment. Compare CASA de Md. v. Trump , 971 F.3d 220, 230 (4th Cir. 2020) (arguing that, while we may “have the technical authority” to disregard a Supreme Court stay, “every maxim of prudence suggests that we should decline to take [that] aggressive step'), with id. at 281 n.16 (King, J., dissenting) (arguing that 'assigning such significance to perfunctory stay orders is problematic”). Suffice it to say, for years, reasonable minds could disagree about whether Supreme Court interim orders bind lower courts. No longer.

The Supreme Court has now spoken.² In Trump v. Boyle , 145 S. Ct. 2653 (2025), the Supreme Court made clear that its interim orders have precedential force. Boyle is itself a short, unsigned interim order. But it packs a punch. While some questions may remain, Boyle tells us three things about Supreme Court interim orders: (1) they can 'squarely control” certain other cases; (2) they “are not conclusive as to the merits”; and (3) they “inform how a court should exercise its equitable discretion in like cases.” Id. at 2654. Let’s break down each point.

First, interim orders can control the outcome of other cases. In Boyle , the Supreme Court stayed a district court's permanent injunction that barred the President from Page 31removing certain members of an independent agency. Id. ; see Boyle v. Trump , 791 F. Supp. 3d 585 (D. Md. 2025). In its short stay order, the Court stated that the case was “squarely controlled by Trump v. Wilcox .' Boyle , 145 S. Ct. at 2654 (citing Wilcox , 145 S. Ct. 1415 (2025)). Wilcox was an earlier interim order that-like Boyle -stayed a district court injunction that barred the President from removing members of two different independent agencies. Even though the cases involved different agencies, the Court determined that Wilcox controlled Boyle . So the Court explicitly embraced the precedential effect of its interim orders. But when are interim orders binding?

Second, Boyle teaches that interim orders 'are not conclusive as to the merits.' Boyle , 145 S. Ct. at 2654. This makes sense. Supreme Court stay decisions apply the familiar Nken stay factors, the first of which requires a court to consider whether the stay applicant “has made a strong showing that he is likely to succeed on the merits.” Nken v. Holder , 556 U.S. 418, 434 (2009). As detailed below, this factor requires a court to make only a likelihood determination—an assessment of the probability of success, not a final adjudication on the merits. So the Supreme Court’s interim orders can be thought of as “probabilistic holdings.” Nat’l Institutes of Health v. Am. Pub. Health Ass’n , 145 S. Ct. 2658, 2664 (2025) (Gorsuch, J., concurring in part). Much like a preliminary injunction itself, these predictive judgments will not bind lower courts making final decisions on the merits. That is why Boyle affirmed that interim orders 'are not conclusive as to the merits.” Boyle , 145 S. Ct. at 2654 (emphasis added). But if they don't bind as to the ultimate decision on the merits, what do they bind? That brings us to Boyle ’s final lesson.

Page 32Third, interim orders “inform how a court should exercise its equitable discretion in like cases.' Id. at 2654. In cases that are sufficiently similar, interim orders have precedential force over a lower court’s exercise of its “equitable discretion.” For example, when a lower court considers whether to grant a stay or to issue a preliminary injunction in a case with similar issues as a Supreme Court interim order, the interim order will “squarely control” the lower court’s later decision.

Taking these three lessons together, the key takeaway from Boyle is this: Supreme Court interim orders bind lower courts at the preliminary stage in like cases.³

In most cases, applying this rule of precedential force will be straightforward. We consider whether an earlier Supreme Court interim order is “like” the case at hand. In doing so, we ask whether meaningful differences exist between the interim order and the case before us. This may—I repeat, may—be made more challenging by the terse nature of some interim orders. But even terse orders can provide significant information. And where an order gives information about what the Supreme Court decided, lower courts must listen. For example, when the Court stays a preliminary injunction, it has necessarily decided that the Nken factors support a stay in that circumstance. This conclusion tells us that the defendant has a likelihood of success on the merits of the appeal and would likely Page 33face irreparable harm absent a stay.⁴ See Nken , 556 U.S. at 434. A lower court considering similar questions cannot ignore the necessary import of the Court’s ruling.

With Boyle ’s lessons in mind, we can now consider the appeal at hand. We review the district court's grant of a preliminary injunction barring the DOGE Team from accessing certain government records. But we don’t do so on an empty slate. The Supreme Court already intervened to stay this very same injunction . See SSA v. Am. Fed’n of State, Cnty., & Mun. Emps. , 145 S. Ct. 1626, 1626 (2025). So, following Boyle , we must ask whether the case at hand is “like” the case in which the Court intervened. Here, this is easy. To ask the question is to answer it: This is the same case with the same issues, the same factors, and the same procedural context. So the Court’s stay decision must “inform” how we 'exercise [our] equitable discretion' in this posture. Boyle , 145 S. Ct. at 2654.

In its interim order, the Court explicitly stated that the Nken factors favored granting a stay. Id. So the Court ruled for the government after considering: “(1) whether the stay applicant has made a strong showing that he is likely to succeed on the merits; (2) whether the applicant will be irreparably injured absent a stay; (3) whether issuance of the stay will Page 34substantially injure the other parties interested in the proceeding; and (4) where the public interest lies.' Nken , 556 U.S. at 434 (quotation marks omitted).

Consider for a moment the first Nken factor-the applicant's likelihood of success on the merits. The Supreme Court has made clear that this factor asks whether the applicant “is likely to prevail on the merits of the issue before [the court], not whether he is likely to prevail on the merits of the underlying suit.” Trump v. CASA, Inc. , 606 U.S. 831, 860 (2025). In this case, the 'issue' before the Supreme Court was the validity of the preliminary injunction. So the Court necessarily concluded that the government was likely to succeed in this appeal to vacate the injunction.⁵ Boyle tells us that this type of Supreme Court predictive judgment binds us when faced with a similar judgment. So the Court’s stay “squarely controls” this appeal, and the preliminary injunction cannot stand.⁶

The Supreme Court’s interim order tells us the required outcome of this appeal— we must vacate the preliminary injunction. But the order doesn’t tell us why. This leaves Page 35us with two lawful options: (1) simply point to the Supreme Court’s decision and dispose of the case on precedent; or (2) independently analyze a ground for reaching the result the Court reached. But a third option is off the table—we may not defy the Supreme Court by upholding the injunction. That should be even clearer when the Supreme Court already exercised its equitable discretion in the very case that we now review.

Our Court takes the second permissible approach of reaching a conclusion consistent with the Supreme Court's ultimate decision. While the Court makes some mistakes along the way, our decision to vacate the injunction below is consistent with the Supreme Court’s directive.

Some judges may be tempted to take the forbidden third option. But to do so is to assert that inferior courts are not required to follow the Supreme Court’s directives. Even when an inferior judge might independently reach a different conclusion, we remain bound by the Supreme Court. This should come as no surprise—that is a situation inferior courts face every day. Article III establishes a hierarchical court system. While our en banc Court is not bound by existing Circuit precedent, we are never free to disregard Supreme Court precedent—no matter how vigorously we might disagree. That’s the very nature of vertical stare decisis.

But before Judge Heytens's opinion concludes, correctly in my view, that the Plaintiffs have not adequately shown that they will suffer irreparable harm absent a preliminary injunction, it reaches out to discuss an issue that isn’t relevant to the opinion. It argues against what I have called the “multiplicative problem.” See American Fed’n of Page 36Teachers v. Bessent ( AFT ), 152 F.4th 162, 170 (4th Cir. 2025).⁷ That critique is the very definition of dicta-it discusses judicial standards for considering Winter 's first factor while expressly stating that the decision does not rely on that factor. Still, because his opinion purports to “abrogate” AFT 's discussion of the multiplicative problem, I want to explain why basic statistical probabilities are not abrogated when one puts on a black robe.⁸

Let’s start with the basics. To grant a preliminary injunction, a district court must decide that the plaintiff has satisfied Winter 's four-factor test. See Winter v. Natural Res. Def. Council, Inc. , 555 U.S. 7, 20 (2008). “This requires the plaintiff to show (1) that they are likely to succeed on the merits, (2) that they are likely to suffer irreparable harm in the absence of preliminary relief, (3) that the balance of equities tips in their favor, and (4) that the injunction is in the public interest.' AFT , 152 F.4th at 168–69 (citing Winter , 555 U.S. at 20).

It's hard for a plaintiff to get a preliminary injunction. But that shouldn't be surprising— Winter tells us that a “preliminary injunction is an extraordinary remedy never awarded as of right.” Winter , 555 U.S. at 24. One challenge plaintiffs face is that 'each of [the] four factors must be satisfied to obtain preliminary injunctive relief.” Henderson for NLRB v. Bluefield Hosp. Co., LLC , 902 F.3d 432, 439 (4th Cir. 2018). On the flip side,Page 37 “ denying a preliminary injunction only takes the rejection of a single factor.” Frazier v. Prince George’s Cnty. , 86 F.4th 537, 544 (4th Cir. 2023). It is because of this “asymmetry” that AFT noted that plaintiffs 'seeking a preliminary injunction thus face an inherently uneven playing field.' AFT , 152 F.4th at 169.

A similar asymmetry often appears within the first Winter factor-the likelihood of ultimate success on the merits.⁹ This 'likelihood' analysis requires a court to make a predictive judgment about a party's future chances of success. But it does not bind a district court on the ultimate merits determination. See Samuel L. Bray, The Purpose of the Preliminary Injunction , 78 Vand. L. Rev. 809, 817 (2025). And that makes sense. The predictive likelihood analysis can only be tentative: Things may change as new evidence is presented, the court can change its mind on an issue, and new precedent can alter the legal landscape. Put simply, the likelihood analysis requires judges to bake in some room for uncertainty.

In essence, Winter 's first factor asks: What is the probability that the movant will ultimately succeed on the merits? A court therefore must be guided by basic principles of probability. It would be wise, then, to understand the basic concept of probability. And part of probability-the multiplicative problem-is worth discussing here. A plaintiff often faces multiple independent barriers to success. In such cases, the plaintiff’s ultimate Page 38success on the merits is equal to the product of the plaintiff’s likelihood of success on each independent issue. This multiplicative problem means that a plaintiff’s chances of future success on the merits get progressively harder the more issues he must win.¹⁰

The multiplicative problem is inherent in Winter 's first factor-you cannot simply wish it away. So in cases with multiple barriers to success, district court judges shouldn’t ignore these basic mathematical principles. They should instead consider the multiplicative problem when analyzing the likelihood of a plaintiff’s ultimate success on the merits. Doing so is not only helpful; it is necessary for making accurate predictive judgments about a plaintiff’s likelihood of ultimately succeeding on the merits.

To be clear, I have not suggested that district court judges must assign a specific number to the plaintiff’s probability of success on each issue and then multiply them out. Though it might sometimes be helpful for a district court judge to do so as part of its weighing process, it won’t be necessary in every case. In many cases—as in AFT itself, where this Court assigned no specific probability for each issue-it will be enough to identify the structural problems facing the plaintiff and conclude that the plaintiff cannot meet Winter 's first factor. See AFT , 152 F.4th at 177. In other words, the multiplicative Page 39approach provides a framework to guide judges’ equitable discretion in deciding whether to grant a preliminary injunction.

Writing for the Court, Judge Heytens’s opinion rejects the multiplicative approach but assures us that it has “no quarrel with the general proposition that it can be harder to win a complex case with multiple issues than a straightforward case that turns on a single issue.” See Majority Op. at 6. But it fails to acknowledge why . It is precisely because there are several ways for a plaintiff to fail but only one way to win. And as the chance of failure on each issue rises, the plaintiff’s case gets increasingly “harder to win.” That, in brief, is the multiplicative problem. It makes no sense to accept the conclusion (that cases with multiple barriers to success are harder to win) but reject the premise on which it is based (the multiplicative problem).

Nor does the multiplicative approach require judges to “perform unfamiliar tasks for dubious benefits.” See Majority Op. at 6. For starters, the underlying tasks are not 'unfamiliar' to judges and lawyers. Lawyers often need to give their clients concrete estimates of a client’s likelihood of success in a given case. See Opinion of QUATTLEBAUM, J. at 47–52 (concurring in the judgment). And judges in civil bench trials have to determine whether a plaintiff has adequately proved the merits to different standards of proof-such as the preponderance standard, which requires the weight of evidence supporting a claim to be greater than 50%. True enough, challenges may arise when a judge addressing a preliminary injunction applies the multiplicative approach to its fullest extent—by assigning probabilities to each issue. But that concern stems not from the multiplicative problem but from Winter 's demand that courts predict the probability Page 40that the movant will ultimately succeed on the merits. It may be that judges are bad at making predictions of future success. But Winter demands it. The multiplicative approach merely provides a helpful guide in doing so.

The Court suggests that judges would have difficulty applying the multiplicative approach because of “the familiar statistical problem known as conditional probability.” Not so. Conditional probability is not a statistical “problem.” It’s a statistical truth. The actual 'problem'-as the case the opinion relies upon illustrates-is not conditional probability itself but when judges ignore it. Al-Adahi v. Obama , 613 F.3d 1102, 1105 (D.C. Cir. 2010) ('Those who do not take into account conditional probability are prone to making mistakes in judging evidence.”). And that is precisely what the majority demands that courts do.

Judges regularly deal with conditional probability—which considers the likelihood of dependent events—even if not by name. Conditional probability is used whenever a judge considers how different pieces of evidence fit together to suggest a given outcome, as when a judge assesses probable cause. United States v. Prandy-Binett , 5 F.3d 558, 559 (D.C. Cir. 1993) (Randolph, J., concurring in the denial of rehearing) (“[J]udges assessing probable cause necessarily deal with conditional probabilities.”). To illustrate, consider a hypothetical search for a bank robber. The sole fact that someone has a ski mask outside the bank does not establish probable cause that they robbed the bank-ski masks are common outside of Aspen banks in the winter. Nor is it alone sufficient that a given person is found outside a just-robbed bank on a hot summer day. But when the person with the ski mask is also the person found outside the just-robbed bank on a hot summer day, the Page 41probability that the robber has been located substantially increases. To isolate each fact without considering the others is to ignore conditional probability. We rightly reject that approach, recognizing that even judges are capable of considering basic issues of conditional probability.

Perhaps I am overly optimistic about judges. But it sure seems to me that we trust ourselves to consider a little math in other contexts, too. Take the immigration context, where judges have to predict the likelihood of an asylum applicant’s future torture. When more than one source of potential torture exists, we have demanded compliance with basic statistics by aggregating the probability of torture from each source. See Lopez-Sorto v. Garland , 103 F.4th 242, 255 (4th Cir. 2024).¹¹ That is, a government agency may not require a single source of potential torture to clear the 50% likelihood threshold on its own. We demand, at least in that context, that the agency consider the conditional probability: How likely is it that the applicant will be tortured given the combined risk of torture A, torture B, and torture C. In other words, what is the overall likelihood based on the aggregate of three estimated likelihoods. To isolate each risk of torture is to ignore Page 42conditional probability. We demand some statistical literacy from agencies, and we should expect it from judges, too. 12

It is also wrong to suggest that the benefits of applying the multiplicative approach are “dubious.” As the Court sees it, the process for considering the likelihood of success in the preliminary injunction context must be an unstructured, amorphous guessing game. But that is no virtue. Considering basic statistical truths allows for more accurate, transparent, and careful analysis—benefits that both judges and parties should welcome.

Nonetheless, the Court concludes that it is “better . . . to stick with the traditional approach” to analyzing the likelihood inquiry. See Majority Op. at 7. To be clear: I do not suggest a departure from the traditional approach. Nor do I suggest “a different or additional hurdle” or a “heightened standard” for obtaining a preliminary injunction. See Majority Op. at 7 (first quote); AFT , 152 F.4th at 178 (King, J., dissenting) (second quote). I simply identify that the multiplicative problem is inherent in the “traditional approach” when multiple barriers to success exist.

Page 43The multiplicative approach is also fully consistent with “well-established principles of equity.' See eBay Inc. v. MercExchange, L.L.C. , 547 U.S. 388, 391-93 (2006). Far from being “a mechanical algorithm” that displaces equitable discretion, it is a framework that guides that discretion in evaluating a plaintiff's likelihood of ultimate success. Indeed, equity demands -not forbids-the application of basic probability principles when it asks courts to engage in an inherently predictive inquiry. These do not cease to apply merely because a court sits in equity.¹³ Courts sitting in equity have always made probabilistic judgments about future events; doing so is inherent in deciding whether to grant any forward-looking remedy and in fashioning appropriate relief. Probabilistic reasoning is thus a hallmark of equity.

Consider, for example, the context of bankruptcy, which has long been understood as the province of equity. See Young v. United States , 535 U.S. 43, 49-50 (2002). In bankruptcy proceedings, courts routinely must evaluate proposed reorganization plans by 'form[ing] an educated estimate' of 'the probabilities of [a reorganization's] ultimate success,” i.e. , the likelihood that creditors will be paid. Protective Comm. for Indep. Stockholders of TMT Trailer Ferry, Inc. v. Anderson , 390 U.S. 414, 424 (1968). In doing so, courts must assess the likelihood of various wildly uncertain and interlocking issues, while also considering “the complexity, expense, and likely duration of [] litigation” and “all other factors relevant to a full and fair assessment.” Id . These predictive judgments are obviously difficult and uncertain, but equity nonetheless demands them. Thus, Page 44probabilistic reasoning has always been implicit in equity’s power to grant forward-looking remedies.¹⁴ And, because the multiplicative approach is a basic rule of probability, it fits comfortably within a court’s equitable role.

Recall that the Supreme Court’s interim order in this case controls the outcome here. But for the sake of clarity, I reiterate that a variety of reasons explain why the Supreme Court may have decided that the government was likely to succeed in this appeal. A brief note on two of those: irreparable harm and standing.¹⁵

First, irreparable harm. I agree with the conclusion in Part IV of Judge Heytens’s opinion that Plaintiffs have not met Winter 's second factor. The irreparable harm analysis does not simply ask whether a plaintiff’s harm will be severe . Instead, it ensures that Page 45injunctive relief is being used to preserve a court’s ultimate remedial options. See Samuel L. Bray, The Purpose of the Preliminary Injunction , 78 Vand. L. Rev. 809, 824 (2025) (“The court asks whether it needs to act now, with a preliminary injunction, to preserve its ability to act in the future.”). So a plaintiff must show that his harm will no longer be remediable in the absence of a preliminary injunction. Plaintiffs fail to make that showing here. Recall Plaintiffs’ theory of harm. They allege harm arising from the knowledge that DOGE Team members may see their personal information. If there is a reason to think that monetary damages or a permanent injunction could not remedy this harm, it does not now occur to me. Plaintiffs’ failure to show irreparable harm is sufficient to require us to vacate the preliminary injunction.

But Plaintiffs should also lose on Winter 's first factor. Consistent with my analysis in AFT , these Plaintiffs have not shown that they likely have standing because they lack a concrete injury in fact. See TransUnion LLC v. Ramirez , 594 U.S. 413, 423-27 (2021).¹⁶ Plaintiffs assert the harm of certain government employees having unlawful access to personal information. And this harm, Plaintiffs argue, bears a close relationship to the common law privacy tort called intrusion upon seclusion. But that’s wrong. The harm involved in an intrusion upon seclusion claim arises from the “knowledge that a third party is engaged in targeted snooping.” AFT , 152 F.4th at 172. Plaintiffs at most allege that a Page 46handful of additional government employees now have generalized access to a vast database that includes their personal information. That claimed harm is different in kind from the sort of investigation traditionally involved in an intrusion upon seclusion claim.¹⁷

Because of this key distinction, Plaintiffs have failed to make a sufficient showing that they likely have standing at this stage of the proceedings. This is another reason to vacate the preliminary injunction.

* * *

Our en banc Court rightly vacates the preliminary injunction granted below. While several independent reasons support this conclusion, one stands above them all: the Supreme Court’s stay of this same injunction requires us to. Still, I regret that our Court purports to reject basic principles of probability that would—and already do—guide district courts in the very task that equitable discretion requires: assessing a plaintiff’s likelihood of future success.

Page 47QUATTLEBAUM, Circuit Judge, with whom RICHARDSON and RUSHING, Circuit Judges, join, concurring in the judgment:

Sometimes, appellate judges can forget what the real world of practicing law is like. This is one of those times. While I agree with the majority’s disposition, I also agree with Judge Richardson on the standard required for preliminary injunctions. But I write separately only to address one of the reasons the majority gives for rejecting what Judge Richardson calls the “multiplicative problem.” The majority suggests that likelihood of success on the merits cannot be effectively assessed in an objective, numerical way. In fact, it quotes a law review article stating that “[c]ognitive limitations leave humans able only weakly to judge likelihood on any sort of scale.” Maj. Op. at 6 (citation omitted). In my view, it'd be a mistake to abrogate American Federation of Teachers v. Bessent , 152 F.4th 162 (4th Cir. 2025), because we don’t think judges can effectively assess likelihood of success in numerical terms. Why do I say that? Because I know firsthand that lawyers around the country do this very thing every day. And if lawyers regularly assess probability of success numerically, judges—who have more time and resources than most lawyers— should be able to do it too.

To explain, consider a plaintiffs’ lawyer who is deciding whether to take a product liability case. The potential client explains how the accident happened. And he then explains that because of it, he incurred $40,000 in medical bills and lost $10,000 of wages from being out of work. Plus, he experienced pain and suffering from his injuries. Based on experience, the lawyer estimates that if she wins at trial, an estimated verdict is roughly three times economic damages. That means she’d predict a verdict of roughly $150,000 or Page 48something in that range. The lawyer also knows that to win, she’ll have to hire product design and accident reconstruction experts, which she estimates will cost $50,000. She figures the other costs she’ll have to pay (an investigator, deposition costs, etc.) are about $15,000. She knows she’ll have to put in lots of her own time, and it’ll likely be a couple of years before the case is tried.

Should she take the case? If she wins, she’ll get a fee of 1/3 of the recovery, or $50,000, leaving $100,000 for the potential client. From that, the client must repay the $65,000 in costs, so his net recovery is $35,000. The lawyer would need to decide if a potential $50,000 fee is worth the amount of work she’d have to take the case to trial or obtain a favorable settlement. And the client would have to decide if the potential to recover $35,000 is worth the time and emotional toll of litigation.

But there is another critical factor to consider before taking the case—what is the likelihood the client will succeed-either at trial or through settlement? The lawyer assesses the likelihood of success by evaluating factors such as the law applicable to the potential claims, the quality of the evidence and her experience with and knowledge of similar cases in that jurisdiction. Her assessment of the likelihood of success then goes into any cost-benefit analysis.

For example, imagine the lawyer assesses the case as a virtual slam dunk. She might say the likelihood of winning is 90%. In doing this, she isn't selecting 90% in a mathematically certain way. No one would say that predicting likelihood of success is that precise. The lawyer might just as well have picked 95% or 85%. But she wouldn’t pickPage 49 60% if she thinks the case is a slam dunk. In other words, the lawyer will likely place some numerical value or range on her chance of success.

Using 90% to make the point, the lawyer might then reduce the predicted recovery by 10% to $135,000. The client would get 2/3, or $90,000, and the lawyer would get 1/3 or $45,000. But the costs must be repaid. Subtracting $65,000 from the client’s portion, the client’s net recovery would be $25,000. So, with a 90% estimated likelihood of success, it’s still likely that the ultimate recovery for the client will be more than the costs. In other words, some upside is still there.

But not so if the lawyer assesses the chance of success as a toss-up. In that situation, the lawyer might estimate the likelihood of success as 50%. Again, she could just as easily have picked 55% or 45%, but, if she thinks the case is a toss-up, she won’t pick 90%. Using 50%, the predicted recovery would be $75,000. That means the likely fee for the lawyer is $25,000, and the likely recovery for the client is only $50,000. But don’t forget the costs. The $65,000 in litigation expenses must either be repaid by the client or eaten by the lawyer. Either way, one will be underwater.

Based on this type of cost-benefit analysis, the lawyer might take the case if she thinks it’s a slam dunk. But she might very well decide not to pursue the case if she thinks it’s a toss-up.¹

Page 50While this example is hypothetical, the fact that lawyers conduct this type of decision-making is not. It’s a common way lawyers and clients make informed decisions about a case—what the majority says judges can’t do effectively. Plaintiffs’ lawyers in the real world can’t afford to disavow numerical assessments of likelihood of success.

And it's not just plaintiffs' lawyers who make probabilistic assessments about likelihood of success. Consider a product liability defendant. Assume the plaintiff and the plaintiffs’ lawyer from our prior example decided to pursue the case. After some discovery, the company that made the product wants to decide whether to go to trial or settle the case. The company knows it will cost $75,000 in legal fees, experts and other costs from that point forward to defend the case through trial. And they agree that if they lose, the likely verdict would be roughly $150,000 or something in that range. The general counsel asks the company’s defense lawyer for a recommendation.

Just as it was for the plaintiffs’ lawyer, the estimate of the likelihood of success is critical to the advice the lawyer will give. If the lawyer thinks the company is almost certainly going to lose, she might assess the likelihood of loss at 90% (based on the slam dunk explanation described above). In that situation, she might recommend the company settle for something less than its total expected losses from trial-which include the $75,000 in projected defense costs plus the $135,000 expected value of the verdict (90%Page 51 of $150,000). But if the lawyer thinks the company has a 90% chance of winning, she’d make a very different settlement recommendation.²

According to the majority, these sort of numerical likelihood-of-success assessments cannot be effectively made. What if after reading the majority opinion, the defense lawyer says to her company client that asked her to assess the likelihood they would win at trial, “you’re asking me the wrong question—I just read the Fourth Circuit’s opinion that says it’s just too hard to give you percentages on likelihood of success. So, while I think we’ll win, I can’t give you any numbers.”

What is the company going to do then? Without some sense of numerical percentages, the client can’t make informed decisions. Frustrated, it’ll likely fire the lawyer who follows the majority’s thinking and replace her with someone who can give it more concrete help.

One last example. Mediation is a common feature of modern litigation. In some instances, district courts require mediation. Others utilize it as a case management tool. Even our court reviews certain appeals to determine whether mediation may be helpful. See 4th Cir. R. 33. During mediation, mediators and parties constantly assess the likelihood of success on issues in numerical terms. It happens every day in much the same way I have Page 52described already. If mediators and parties are assessing likelihood of success in numerical terms to engage in a process that federal judges promote, I don’t think we as judges should say it’s inappropriate for us to do so in our assessment of the very same issues.

To conclude, I agree that we should vacate the preliminary injunction and remand to the district court. So, I concur in the judgment. On the standard for preliminary injunctions, however, I agree with Judge Richardson. But regardless of how we decide that issue, it shouldn’t be because we can’t effectively assess likelihood of success numerically. That makes us seem out-of-touch with the real world, where lawyers do every day just what the majority says can't be effectively done.

Page 53KING, Circuit Judge, with whom Judges GREGORY, WYNN, THACKER, BENJAMIN, and BERNER join, concurring in part, dissenting in part, and dissenting from the judgment:

When the district court issued its preliminary injunction, the facts then known to the district court were bad enough. The Social Security Administration (“SSA”) had abruptly opened all its records to affiliates of the President’s then-new Department of Government Efficiency (“DOGE”) despite the DOGE affiliates’ lack of vetting, lack of training, and lack of any demonstrated need for the vast and extremely sensitive personal information that fills the SSA records.

The facts now known are much worse!

As SSA recently revealed in a “Notice of Corrections to the Record,” a significant portion of the information provided by SSA and the other defendants in the preliminary injunction proceedings was patently false. The Notice of Corrections confesses repeated violations of the district court’s prior temporary restraining order (the “TRO”) and multiple instances of the DOGE affiliates’ misuse and mishandling of SSA records. Moreover, the Notice of Corrections belies SSA’s entire justification for opening its records to the DOGE affiliates — that the DOGE affiliates are regular SSA employees working under SSA’s supervision, in accordance with its rules, and on its behalf — by exposing that the DOGE affiliates are actually rogue actors whose activities are hidden from SSA itself.

After receiving SSA’s Notice of Corrections and on the motion of the plaintiffs, the district court promptly corrected the record on appeal. That leaves our en banc Court with at least two legitimate options for disposing of this appeal. We could (1) assess the merits of the preliminary injunction on the basis of the corrected record or (2) remand, without Page 54assessing the preliminary injunction’s merits, so that the district court may decide anew whether to award injunctive relief on the basis of the corrected record and subsequent developments.

I would pursue option (1) — assessing the merits of the preliminary injunction on the basis of the corrected record — and I would thereby affirm the preliminary injunction without hesitation. Candidly, I would affirm even if it were the erroneous original record that controls the analysis. The very able district judge acted with exceptional thoughtfulness in issuing the preliminary injunction, committing no legal error or otherwise abusing her discretion.

Regrettably, however, my friend Judge Heytens pursues neither option (1), option (2), nor any other legitimate option for disposing of this appeal. Instead, in Part IV of his opinion, Judge Heytens improperly disregards the corrected record and wrongly relies on the erroneous original record to assess the preliminary injunction’s merits. Compounding that misstep, Judge Heytens then unjustifiably rules that the district court erred in crediting the plaintiffs’ showing of irreparable harm, such that the preliminary injunction must be vacated.

In these circumstances, I concur solely in Parts I through III of Judge Heytens’s opinion. I am compelled to dissent from Part IV of the opinion, as well as the resultant Page 55judgment of our en banc Court vacating the preliminary injunction because of purported error by the district court.¹

In explaining my views, I begin with a discussion of the relevant factual and procedural history. This covers the erroneous original record that was before the district court when it issued the preliminary injunction, as well as the corrected record now before us on appeal. Additionally, I address two extra-record whistleblower reports, which concern potentially significant events that allegedly occurred in the preliminary injunction’s wake.

From evidence in the erroneous original record that has not been corrected and thus remains valid, the district court knew when it issued its preliminary injunction that, in early February 2025, SSA had accorded DOGE affiliates unfettered access to the Social Security records of essentially everyone in our Country. Those records contain a mass of personal information of the utmost sensitivity, including the following:

• ● Names and Social Security numbers;
• ● Names of parents and their Social Security numbers;

Page 56● Dates and places of birth; ● Phone numbers and home addresses; ● Data regarding citizenship, ethnicity, race, and sex; ● Birth and marriage certificates; ● School and family court records; ● Employment and pension records; ● Details of driver's licenses, credit cards, and bank accounts; ● Tax and earnings information; and ● Extensive medical and mental health records, documenting treatments, hospitalizations, prescription medications, test results, and more.

Until SSA opened its records to the DOGE affiliates, there had been good reason for the American people to be confident that SSA was safeguarding the personal information entrusted to it. Since SSA’s inception in 1935, a bedrock principle of the agency had been to ensure the confidentiality and security of its records. Moreover, SSA had been known to carefully abide by the various regulations and statutes enacted to reinforce its record-protecting obligations, including the Privacy Act of 1974.

Of especial relevance here, the Privacy Act limits SSA's internal disclosure of records containing personal information to solely those SSA employees “who have a need for the record in the performance of their duties.” See 5 U.S.C. § 552a(b)(1). As such, SSA implemented strict policies of “need to know” (according access to only employees with a need for it) and “least privilege” (granting just the minimum access necessary and Page 57anonymizing data whenever possible), along with “separation of duties” (denying users enough privileges to misuse a system on their own) and “zero trust” (performing regular security audits, assuming breaches, and scrutinizing requests for information). SSA also required stringent background investigations and training before an employee was granted access, as well as regular recertifications and retraining thereafter.

Notably, those policies had long applied to SSA employees performing the same work in which the DOGE affiliates are now purportedly engaged, that being technology upgrades and fraud, waste, and abuse detection. But once the DOGE affiliates arrived at the agency, SSA flouted its own rules and opened all its records to the DOGE affiliates, without even requiring them to undergo standard background checks and training.

Furthermore, although the district court afforded SSA ample opportunities in the preliminary injunction proceedings to do so, SSA never provided a clear, consistent, or convincing explanation why the DOGE affiliates need unfettered access to the personal information contained in the SSA records, as required by the Privacy Act. Nor did SSA demonstrate that the DOGE affiliates — unlike SSA employees before them — cannot accomplish their work with “least privilege” access and thus largely anonymized data. At most, SSA’s evidence merely suggested that a lack of unfettered access to non-anonymized data may cause the work to take longer.

On April 17, 2025, based on the erroneous original record, the district court issued its preliminary injunction, supported by a thorough and cogent 148-page opinion. The preliminary injunction replaced the court's TRO of March 20, 2025, which had been Page 58accompanied by a 137-page opinion. The relief was awarded on claims brought by the plaintiffs — two national labor and membership associations and a grassroots advocacy organization, including lead plaintiff American Federation of State, County and Municipal Employees, AFL-CIO ('AFSCME') - on behalf of their millions of members.²

In issuing the preliminary injunction, the district court focused on the plaintiffs’ claims under the Administrative Procedure Act for violations of the Privacy Act and for arbitrary and capricious agency action, premised on SSA’s opening of its records to DOGE affiliates who have no need for the personal information contained therein and who lack standard background checks and training. With respect to those claims, the court carefully confirmed the plaintiffs’ Article III standing to sue and concluded that the plaintiffs have satisfied the traditional Winter preliminary injunction standard. See Winter v. Nat. Res. Def. Council , 555 U.S. 7, 20 (2008) ('A plaintiff seeking a preliminary injunction must establish [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.”).

The preliminary injunction, like the TRO before it, both ends DOGE’s unfettered access to SSA records and requires DOGE affiliates to disgorge and delete the personal information already in their possession or under their control. Additionally, the Page 59preliminary injunction prohibits DOGE affiliates from installing any software at SSA, directs them to remove any software previously installed by them or on their behalf, and bars them from accessing, altering, or disclosing any SSA computer or software code.

The preliminary injunction does not, however, completely block DOGE’s records access or stop its purported work on technology upgrades and fraud, waste, and abuse detection. Rather, the preliminary injunction allows SSA to provide DOGE affiliates, once properly trained and vetted, with access to all anonymized data, as well as to any discrete non-anonymized data actually needed to perform legitimate work.

Seeking to be freed from the modest strictures imposed by the district court, SSA and the other defendants noted this appeal from and sought an immediate stay of the preliminary injunction on the basis of the erroneous original record. By majority votes of all active judges, we agreed on April 30, 2025, to initial en banc consideration of the stay motion and to deny the requested stay. We then agreed on May 6, 2025, to an initial hearing en banc on the merits of this appeal.³

On June 6, 2025, however, the Supreme Court stayed the preliminary injunction, relying on the erroneous original record. See SSA v. AFSCME , 145 S. Ct. 1626, 1626 (2025). In so doing, the Court did not explain its ruling and instead provided only this brief discussion of the Nken stay factors:

Page 60When considering whether to grant a stay, this Court looks to four factors: “(1) whether the stay applicant has made a strong showing that he is likely to succeed on the merits; (2) whether the applicant will be irreparably injured absent a stay; (3) whether issuance of the stay will substantially injure the other parties interested in the proceeding; and (4) where the public interest lies.” Nken v. Holder , 556 U.S. 418, 434 (2009) (quoting Hilton v. Braunskill , 481 U.S. 770, 776 (1987)). After review, we determine that the application of these factors in this case warrants granting the requested stay. We conclude that, under the present circumstances, SSA may proceed to afford members of the SSA DOGE Team access to the agency records in question in order for those members to do their work.

Id. Pursuant to the Court's decision, the preliminary injunction is stayed pending our Court's disposition of this appeal and any subsequent Supreme Court proceedings. Id.

It was not until January 16, 2026 — many months after the district court issued the preliminary injunction and our en banc Court denied a stay (in April 2025), the Supreme Court awarded a stay (in June 2025), and we conducted the oral argument in this appeal (on September 11, 2025) — that SSA filed its “Notice of Corrections to the Record” in the district court. On January 21, 2026, the district court corrected the record on appeal by supplementing it with the Notice of Corrections. See Fed. R. App. P. 10(e)(2)(B) ('If anything material to either party is omitted from or misstated in the record by error or accident, the omission or misstatement may be corrected and a supplemental record may be certified and forwarded . . . by the district court[.]”). The corrected record was thereafter transmitted to our Court on February 9, 2026.

From the corrected record, we now know that SSA and the other defendants provided patently false information to the district court in the preliminary injunction Page 61proceedings.⁴ We thus know that the prior rulings in this matter - the district court's issuance of the preliminary injunction, our Court's denial of a stay, and the Supreme Court’s grant of a stay — were rendered on a materially erroneous record. And we know that, going forward, we should not accord the defendants any benefit of the doubt or readily trust in anything they say.

Regarding the false information proffered in the preliminary injunction proceedings, we specifically know that SSA falsely assured the district court of compliance with the TRO of March 20, 2025, and that SSA failed to disclose the full extent of the DOGE affiliates’ systems access. We further know that SSA falsely advised the district court that no DOGE affiliate other than those assigned to SSA had ever had access to SSA records, that SSA had safeguards in place that would prevent DOGE affiliates from violating SSA security protocols and integrating SSA systems with outside servers, and that the DOGE affiliates' work at SSA had been limited to technology upgrades and fraud, waste, and abuse detection within the agency.

Correspondingly, we know that SSA found itself obliged to belatedly confess the following, which includes repeated violations of the TRO and multiple incidents of the DOGE affiliates’ misuse and mishandling of SSA records:

• ● The TRO was violated by SSA's failure to terminate the DOGE affiliates' unfettered records access until approximately noon on March 24, 2025, as well as by a DOGE affiliate’s searches of SSA records for personal information earlier that morning;

• ● Page 62The TRO was again violated when SSA gave a DOGE affiliate access to certain records containing personal information from March 26 to April 2, 2025, and yet again when SSA gave a DOGE affiliate access to different records containing personal information from April 9 to June 11, 2025, through the period that the preliminary injunction was unstayed;
• ● SSA failed to disclose that it had granted DOGE affiliates systems access enabling them to, inter alia, exchange data with each other in a 'shared workspace' and access personal information via a 'data visualization tool”; and
• ● It was unknown to SSA at the time of the preliminary injunction proceedings but later discovered that:
• -On March 3, 2025, an SSA DOGE affiliate copied DOGE affiliates with the DOGE umbrella organization and the Department of Labor on an SSA email to the Department of Homeland Security, attaching an encrypted and passwordprotected file believed to contain personal information derived from the SSA records of some 1,000 people;
• -From March 7 to 17, 2025, DOGE affiliates shared SSA data through the third-party server “Cloudflare”; and
• -On March 24, 2025, acting in his official capacity with SSA, a DOGE affiliate entered a 'Voter Data Agreement' with an unnamed 'political advocacy group' for the purpose of proving voter fraud and overturning certain state election results.

In an effort to downplay the import of its admissions, SSA has emphasized that it is unsure if one of the DOGE affiliates granted access to personal information in the wake of the TRO actually engaged with any personal information. Nor has SSA been able to determine exactly what was contained in the encrypted and password-protected file provided to the outside DOGE affiliates or whether the password to view the file was shared with them. Nor has SSA been able to ascertain exactly what SSA data was shared Page 63to Cloudflare or whether that data yet exists on that server. Nor has SSA seen evidence that the Voter Data Agreement resulted in the sharing of SSA data with the political advocacy group.

But this most assuredly does not, as SSA would have it, somehow render the corrected record inconsequential. Rather, it lays bare what are probably the most damning facts now known. That is, the DOGE affiliates are not regular SSA employees working under SSA’s supervision, in accordance with its rules, and on its behalf. Rather, they are rogue actors whose activities are hidden from SSA itself.

Beyond the corrected record, there are the potentially significant events that allegedly occurred in the preliminary injunction's wake, as reflected in the two whistleblower reports. The first of those reports was made to the U.S. Office of Special Counsel and members of Congress in late August 2025 by Charles Borges, then SSA’s Chief Data Officer, alleging “serious data security lapses” on the part of DOGE and its affiliates.

The Borges report specifies that, shortly after the Supreme Court stayed the preliminary injunction in early June 2025, DOGE affiliates evidently authorized themselves to create a copy of SSA’s Numerical Identification System (“NUMIDENT”) database, which contains the detailed and extremely sensitive personal information relating to each of the more than 450 million Social Security numbers ever issued. According to the Borges report, the DOGE affiliates then transferred the NUMIDENT data to a highly vulnerable “cloud environment” controlled by DOGE and beyond SSA oversight, exposing Page 64the data to misuse by not only DOGE, but also identity thieves, blackmailers, and other bad actors.

Questioned about the Borges report during the September 2025 oral argument in this appeal, counsel for SSA and the other defendants emphasized that there was no evidence the NUMIDENT data had actually been compromised. But as the report conveys, Borges could not ascertain whether the NUMIDENT data had been compromised because he was locked out of the DOGE-controlled cloud server. And as my friend Judge Thacker pointed out at the oral argument, the best defendants' counsel could say is that the NUMIDENT data apparently had not been compromised “yet.” See Oral Argument at 25:42, AFSCME v. SSA , No. 25-1411 (4th. Cir. Sept. 11, 2025), https://www.ca4.uscourts.gov/OAarchive/mp3/25-1411-20250911.mp3.

Since then, the plaintiffs have moved in the district court for the court to lift its stay of its own proceedings pending this appeal and to authorize limited discovery into the information disclosed by SSA’s “Notice of Corrections to the Record.” In support of their motion, the plaintiffs recently submitted to the court a recent news article regarding the second whistleblower report, made to SSA’s inspector general in January 2026. That report is said to allege that a former DOGE software engineer told coworkers that he possessed copies of the NUMIDENT and another SSA database — including a copy of at least one of those databases on a thumb drive - that he planned to share with his new private employer, a government contractor.

To be clear, I recognize that the whistleblower reports are not part of the evidence before us and concern events that allegedly occurred in the preliminary injunction’s wake.

Page 65The evidence before us is instead that contained in the corrected record, pertaining to events that predate the preliminary injunction. And while the whistleblower reports are notable for the troubling questions they raise about the ongoing activities of DOGE affiliates at SSA, the corrected record already shows the DOGE affiliates’ misuse and mishandling of SSA records, as well as their freedom from SSA oversight and control.

That brings me to the explanation of my areas of agreement — and disagreement — with my friend Judge Heytens. First and foremost, I explain why I dissent from Part IV of Judge Heytens’s opinion and the resultant judgment vacating the preliminary injunction. Thereafter, I explain why I concur in Parts I through III of the opinion.

In Part IV of his opinion, Judge Heytens insists that — although we now know that the original record is replete with false information — our Court is constrained to assess the merits of the district court's preliminary injunction on the basis of the erroneous original record. Relying on that record, Judge Heytens rules that the district court erred in crediting the plaintiffs’ showing of irreparable harm, such that the preliminary injunction must be vacated.

Judge Heytens’s first mistake is that he disregards the corrected record and instead assesses the merits of the preliminary injunction on the basis of the erroneous original record. In so doing, Judge Heytens relegates SSA’s “Notice of Corrections to the Record”Page 66 to a footnote, despite recognizing that “[t]he government’s recent acknowledgments are alarming and raise serious questions about its earlier conduct before the district court.' See ante 15 n.8. As Judge Heytens sees it, “even though the [Notice of Corrections] has been made part of the official record on appeal, our task in this appeal is to review the record that was before the district court at the time the preliminary injunction was entered.” Id. (internal quotation marks omitted). To Judge Heytens, the corrected record is merely something that the district court may consider on remand — along with the whistleblower reports and other subsequent developments — in conjunction with “any future requests for appropriate relief or corrective action.' Id.

a.

For support of his approach, Judge Heytens invokes the decisions of the Sixth Circuit in Wilson v. Williams , 961 F.3d 829 (6th Cir. 2020), and the Tenth Circuit in Verlo v. Martinez , 820 F.3d 1113 (10th Cir. 2016). But those decisions neither involved nor expressly addressed the role of a record that was corrected on appeal. And as it turns out, they actually invalidate Judge Heytens's reliance on the erroneous original record and instead require consideration of the corrected record.

To elaborate, the passages of the Wilson and Verlo decisions highlighted by Judge Heytens recognize that the review of a preliminary injunction by a court of appeals must be based on the facts as of the time the preliminary injunction was issued by the district court, and not on subsequent developments. See Wilson , 961 F.3d at 833 (observing that “our task is to review the record that was before the district court at the time the preliminary injunction was entered” (citation modified)); Verlo , 820 F.3d at 1125 (recognizing same Page 67and thus that it would be improper 'to consider events occurring after the preliminary injunction hearing to determine whether the district court abused its discretion in issuing the preliminary injunction”). Our Court has heeded the same uncontroversial principle, limiting our review to the facts as of the time the preliminary injunction was issued. See, e.g. , Wetzel v. Edwards , 635 F.2d 283, 286 (4th Cir. 1980) (observing that we “decide[] only whether the grant of interlocutory relief . . . was appropriate under the facts of [the] case,” i.e., “whether the record shows an abuse of discretion by the district court”).

Pursuant to Wilson , Verlo , and similar authorities such as Wetzel , I readily acknowledge that we cannot consider any evidence, including the whistleblower reports, relating to events that have allegedly occurred since the district court issued the preliminary injunction before us today. The corrected record, however, does not concern events that occurred in the preliminary injunction’s wake.

Rather, under Federal Rule of Appellate Procedure 10(e)(2), the corrected record is the compilation of the facts as of the time the preliminary injunction was issued. That is, the corrected record presents the facts as they truly were, correcting the false version of the facts portrayed by the erroneous original record due to “material” information having been “omitted from or misstated in [that] record by error or accident.” See Fed. R. App. P. 10(e)(2). Simply put, the corrected record presents the facts as they truly were, while the erroneous original record depicts the facts as they never were. It is therefore the corrected record — not the erroneous original record — that is the district court’s record of the preliminary injunction proceedings.

Page 68So, when Wilson and Verlo recognize that a court of appeals must “review the record that was before the district court at the time the preliminary injunction was entered,” they necessarily require our consideration of the corrected record herein. Contrary to Judge Heytens’s reading of them, Wilson and Verlo in no way mandate reliance on the erroneous original record and the false version of the facts it portrays.

Judge Heytens’ reliance on the erroneous original record not only lacks support in the authorities he cites, but also flouts the appellate rules. Federal Rule of Appellate Procedure 10(e)(2)(B) authorizes the district court to correct material errors in the record even after it has been forwarded to the court of appeals. And our corresponding local rule favors having the district court resolve “in the first instance” any “[d]isputes concerning the accuracy . . . of the record on appeal.” See 4th Cir. R. 10(d). As reflected in that rule, we are so concerned with the record's accuracy that we deem it 'unnecessary to seek permission” to correct the record and allow correction “at any time during the appellate process.' Id.

By disregarding the corrected record and instead relying on the erroneous original record, Judge Heytens renders the foregoing rules utterly meaningless. Moreover, he encourages egregious litigation conduct, in that he rewards SSA and the other defendants for providing false information to the district court.

The end result is that Part IV of Judge Heytens’s opinion constitutes an illegitimate advisory decision that turns on facts that we now know were never true. Of course, “Article III does not assign to federal courts any power to address hypothetical circumstances, give Page 69advisory opinions, or resolve abstract disputes.” See B.R. v. F.C.S.B. , 17 F.4th 485, 493 (4th Cir. 2021). But Judge Heytens does just that by assessing the merits of the preliminary injunction on the basis of the erroneous original record.

It bears emphasizing that Judge Heytens’s reliance on the erroneous original record is not only wholly improper, but also completely unnecessary. As I stated at the outset of this opinion, there are at least two valid options for disposing of this appeal, those being (1) assessing the merits of the preliminary injunction on the basis of the corrected record or (2) remanding, without assessing the preliminary injunction’s merits, so that the district court may decide anew whether to award injunctive relief on the basis of the corrected record and subsequent developments.

Perhaps Judge Heytens would be uncomfortable pursuing option (1) and relying on the corrected record because of the magnitude of the corrections. If so, an appropriate alternative would be to pursue option (2) and refrain from assessing the preliminary injunction’s merits at all. Indeed, there is precedent for such a course, at least in situations where there have been significant developments following the district court’s award of the injunctive relief under review. See, e.g. , McLeod v. Gen. Elec. Co. , 385 U.S. 533, 535 (1967) (ordering the vacatur of the district court’s temporary injunction and a remand to that court so that it could “determine in the first instance the effect of [a] supervening event upon the appropriateness of injunctive relief”); City of Pontiac Retired Emps. Ass'n v. Schimmel , 751 F.3d 427, 429 (6th Cir. 2014) (per curiam order of en banc court) (recognizing that “[l]egal, factual, and equitable considerations have developed Page 70significantly since the district court denied the plaintiffs' request for a preliminary injunction,” and thus vacating the district court’s denial of injunctive relief and remanding for further proceedings).

Nevertheless, Judge Heytens rejects the valid options and sets on his improper course. He proceeds by wrongly relying on the erroneous original record to assess the merits of the preliminary injunction.

Judge Heytens’s next mistake is his Part IV ruling that the district court erred in crediting the plaintiffs’ showing of irreparable harm, such that the preliminary injunction must be vacated. In reaching that ruling, Judge Heytens declines to consider the theory, supported by the corrected record, “that plaintiffs’ members would be harmed by some downstream misuse or public disclosure of their personal data.” See ante 17. Instead, Judge Heytens confines his irreparable harm analysis to the theory that the plaintiffs were compelled by the erroneous original record to rely on: “that DOGE violates plaintiffs’ members’ privacy by accessing their sensitive personal data without lawful authorization.” Id.

In fairness to the plaintiffs, I underscore that they predicted from the outset of this action — as it turns out, accurately — that DOGE affiliates would misuse and improperly disclose the personal data contained in SSA records. Accordingly, the plaintiffs raised a theory of irreparable harm in the preliminary injunction proceedings premised on the potential for (but not any actual) misuse and improper disclosure. The district court understandably disallowed that theory, explaining in its preliminary injunction opinion that Page 71the “risk [was then] too speculative to constitute irreparable harm.” See AFSCME v. SSA , 778 F. Supp. 3d 685, 776 n.56 (D. Md. 2025). In other words, the plaintiffs were thwarted by a lack of evidence that the corrected record now provides, i.e., evidence that by the time of the preliminary injunction proceedings, misuse and improper disclosure had already occurred.

Even considering the erroneous original record, however, Judge Heytens's irreparable harm ruling is wrong. That is, Judge Heytens unjustifiably concludes that the plaintiffs have not made the requisite showing that they are likely to suffer irreparable harm in the absence of a preliminary injunction, in that they have not demonstrated that future money damages or a reparative permanent injunction would be insufficient to remedy their injuries.

On the issue of future money damages, Judge Heytens improperly faults the plaintiffs for failing to substantiate their contention that such 'damages would be insufficient and difficult to ascertain.” See ante 18 (internal quotation marks omitted). And on the issue of a reparative future injunction, Judge Heytens first misreads the erroneous original record — misconstruing it to show that SSA’s opening of its records to DOGE affiliates constituted only an “allegedly unlawful disclosure . . . made to a small group of people within the government” — and then relies on that misinterpretation to wrongly conclude that any harm could thus easily be remedied by eventually “order[ing] the relevant employees to destroy any illegally obtained data or work derived from such data.' Id. at 18-19.

Page 72As the district court underscored in its preliminary injunction opinion, this case has never been about something as simple as DOGE affiliates having a single look at a given SSA record. See AFSCME , 778 F. Supp. 3d at 776-78 (district court’s irreparable harm analysis). From the start, it has been about DOGE affiliates — whether or not they have vetting, training, or need — being accorded ongoing and unfettered access to “some of the most sensitive personal information imaginable.' Id. at 777 (internal quotation marks omitted). Consequently, “[m]oney damages cannot rectify this invasion of privacy.” Id. at 778 (citing numerous persuasive authorities). And waiting for a reparative permanent injunction “lets the proverbial genie out of the bottle,” such that “the plaintiffs will already have suffered irreparable harm.” Id. (quoting Am. Fed’n of Teachers v. Bessent , No. 251282, 2025 WL 1023638, at *10 (4th Cir. Apr. 7, 2025) (Berner, J., dissenting from the denial of initial hearing en banc)).

Lastly, I note one more point on which I disagree with Judge Heytens. In his irreparable harm analysis, Judge Heytens suggests that the Supreme Court's stay may foreclose a showing of irreparable harm. See ante 19 (deeming it “the elephant in the room' that, because of the Supreme Court's stay, 'the district court's preliminary injunction cannot currently protect anyone from anything and no decision we issue today has the power to change that fact”).

As I see it, the Supreme Court’s stay has no role in the irreparable harm analysis. Our Court is tasked today with deciding whether the district court abused its discretion in issuing the April 2025 preliminary injunction based on the facts as of that time. See Wetzel , 635 F.2d at 286 (limiting our review to the facts as of the time the preliminary injunction Page 73was issued). The Supreme Court’s stay of June 2025 is not such a fact. Moreover, to treat it as one would be just another way of saying what my friends Judge Wilkinson and Judge Richardson wrongly assert in their opinions herein: that the Supreme Court's stay absolutely requires our Court to vacate the preliminary injunction. See infra Part III (joining Judge Wynn in rejecting Judge Wilkinson’s and Judge Richardson’s assertions regarding the Supreme Court’s stay).

* * *

At bottom, because Judge Heytens improperly assesses the merits of the preliminary injunction on the basis of the erroneous original record, and because in that assessment he unjustifiably rules that the district court erred in crediting the plaintiffs' showing of irreparable harm, I am left to dissent from Part IV of his opinion. I also must dissent from the resultant judgment of our en banc Court vacating the preliminary injunction due to purported error by the district court.

Turning to Parts I through III of Judge Heytens’s opinion, I concur in those aspects of the opinion for the following reasons. Part I is just a succinct and unobjectionable discussion of the background of this matter.

Part II rightly abrogates the mathematical standard that was created from whole cloth by Judge Richardson in a similar DOGE case — American Federation of Teachers v. Bessent , 152 F.4th 162, 169-71 (4th Cir. 2025) (' AFT ”) — for analysis of the likelihood of success on the merits in stay and preliminary injunction proceedings. I have steadfastly advocated rejecting that baseless heightened standard, see, e.g. , id. at 178-79 (King, J., Page 74dissenting), and I am relieved that our Court is taking the opportunity to do so today. Whether the merits of the preliminary injunction are being reviewed on the basis of the erroneous original record (as Judge Heytens does) or the corrected record (as I would do), 'we cannot decide whether the district court abused its discretion without ensuring it applied the correct legal standards.” See ante 5 n.1 (citation modified). In Part II of his opinion, Judge Heytens confirms that the district court appropriately applied the traditional Winter standard and in no way erred by failing to utilize Judge Richardson's mathematical standard. Id. (“Because we disagree with (and abrogate) the relevant portions of AFT , we conclude the district court committed no legal error in this regard.”).

As for Part III of Judge Heytens's opinion, it correctly rules that the plaintiffs possess Article III standing to sue, abrogating the contrary and erroneous standing ruling in AFT , see 152 F.4th at 171-74. Importantly, when evaluating standing, a court “must look to the facts at the time the complaint was filed.” See Wild Va. v. Council on Env’t Quality , 56 F.4th 281, 293 (4th Cir. 2022). Thus, for purposes of the standing inquiry, Judge Heytens properly considers the plaintiffs’ theory of the case when it was initiated, i.e., “that handing over non-anonymized and highly sensitive information to DOGE was itself unlawful.” See ante 4 (recognizing that the plaintiffs’ original theory “was not that DOGE had misused the information or disclosed it (accidentally or otherwise) to malicious actors”).

To be sure, the evidence in the corrected record — that DOGE affiliates actually have misused and improperly disclosed personal information contained in SSA records — would support an even stronger theory with respect to Article III standing to sue.

Page 75Nevertheless, as Judge Heytens recognizes in Part III of his opinion, the plaintiffs have established their standing even without such a theory.

Finally, I briefly address the opinions of my friends Judge Wilkinson, Judge Richardson, and Judge Quattlebaum. As previously mentioned, Judge Wilkinson and Judge Richardson assert that the Supreme Court’s stay absolutely requires our Court to vacate the district court's preliminary injunction. But pursuant to Judge Richardson's opinion, they would vacate the preliminary injunction in any event, on the premise that the plaintiffs have failed to establish Article III standing to sue and have failed to make an adequate showing of irreparable harm based on the erroneous original record. Additionally, Judge Richardson and Judge Quattlebaum seek to defend the heightened mathematical standard that was conjured up by Judge Richardson in AFT .

My views on the bulk of those issues — Judge Richardson’s mathematical standard, Article III standing to sue, and irreparable harm — have already been specified herein. Again, I concur in Judge Heytens’s opinion insofar as he rejects the improper mathematical standard and confirms the plaintiffs' Article III standing to sue, but I dissent from his opinion insofar as he assesses the merits of the preliminary injunction on the basis of the erroneous original record and concludes that the district court erred in crediting the plaintiffs’ showing of irreparable harm.

I thus now focus on the issue of whether the Supreme Court's stay absolutely requires our Court to vacate the district court’s preliminary injunction. According to Judge Page 76Wilkinson and Judge Richardson, it does because of the overlap between the Nken stay factors and the Winter preliminary injunction standard. I disagree with their theory for two reasons.

First, the proper outcome of this appeal cannot possibly be preordained, in that the Supreme Court had to rely on the erroneous original record in issuing its stay ruling, while we must assess the merits of the preliminary injunction on the basis of the corrected record. Simply put, the facts now known are materially different from the facts considered by the Supreme Court, foreclosing any notion that the Court’s Nken stay analysis controls our Winter preliminary injunction assessment.

Indeed, neither Judge Wilkinson nor Judge Richardson has a satisfying response to this obvious problem with their theory. Each instead insists that our Court must consider the erroneous original record (as the Supreme Court did when it entered its stay), latching onto the groundless statement in Judge Heytens’s opinion that “our task in this appeal is to review [the erroneous original record, as] the record that was before the district court at the time the preliminary injunction was entered.” See ante 15 n.8 (internal quotation marks omitted).

As for my second reason for disagreeing with the theory advanced by Judge Wilkinson and Judge Richardson, I would not believe us to be bound by the Supreme Court’s stay even if there had been no corrections to the record after that stay was entered. That is because — notwithstanding that there is overlap — there are also key distinctions between the Nken stay factors and the Winter preliminary injunction standard. In this Page 77regard, I largely rely on the cogent opinion of my friend Judge Wynn, which I am pleased to join.

I add only that I reject the proposition advanced by Judge Wilkinson and Judge Richardson that, in the wake of the Supreme Court’s unexplained stay decision, our options are limited to either (1) summarily vacating the preliminary injunction or (2) choosing some basis for a vacatur. That proposition may seem reasonable to Judge Wilkinson, Judge Richardson, and my colleagues who have joined their opinions, but none of them would uphold the preliminary injunction anyway.

I, on the other hand, have discerned no meritorious ground for vacating the preliminary injunction. As such, I would see it as an abdication of my commission as an Article III judge and a betrayal of my conscience to rubber stamp a vacatur or to concoct an excuse for one.

Pursuant to the foregoing, I concur in Parts I through III of Judge Heytens’s opinion. I dissent, however, from Part IV of the opinion and from the resultant judgment of our en banc Court vacating the preliminary injunction because of purported error by the district court. Again, I commend the work of the district court and would affirm the preliminary injunction.

Page 78WYNN, Circuit Judge, writing separately, with whom Judges KING, THACKER, BENJAMIN, and BERNER join: 1

I write to voice my disagreement with the separate opinions of my fine concurring colleagues Judge Wilkinson and Judge Richardson. Together, they advocate a step that judges need not take, and in my view, should not take. That’s because by treating the Supreme Court’s short interim order in Trump v. Boyle as setting precedent, they blur a line the Supreme Court itself has carefully preserved—the line between provisional relief and precedential lawmaking.²

In fact, the Supreme Court could not have been clearer on that point: “[O]ur interim orders are not conclusive as to the merits . . . .” Trump v. Boyle , 145 S. Ct. 2653, 2654 (2025). Those words are not a rhetorical aside. They reflect a long-settled understanding of the judicial role. Interim orders exist to manage risk, preserve institutional interests, and prevent irreparable harm while legal questions mature through the ordinary appellate process. They are not designed to announce governing law, and the Court has never said they do.

Page 79My concurring colleagues rely on the Supreme Court's statement in Boyle that interim orders “inform how a court should exercise its equitable discretion in like cases.” 145 S. Ct. at 2654. That’s true—but it is not enough. To inform discretion is not to control it. Equity has always allowed room for judgment, nuance, and factual distinction. But converting guidance into commands drains equity of its very character and substitutes rigidity where flexibility was intended.³

There is a deeper concern. Interim orders are frequently issued without full briefing and without oral argument. That counsels caution, not expansion. To treat interim orders as binding precedent abandons our long-held jurisprudence of deciding constitutional law through reasoned opinions, not emergency motions made under intense time pressure. More profoundly, it would weaken the public’s confidence in the integrity of our judicial system's commitment to deliberation and transparency.⁴

Page 80And even if Boyle reflects an expansion in the use and importance of interim orders, as my concurring colleagues suggest, that expansion would in no way apply to being able to predict how the Supreme Court will treat these issues in future years. Nothing in their unexplained and summary nature confers interim orders the status of precedent that a reasoned opinion commands in this Court. They bind no future Justice, constrain no future Court, and provide no principle capable of consistent application. In short, interim orders announced without reasons can just as easily be ignored without explanation—thereby undermining public confidence and eroding trust in the integrity of judicial decision making.

If anything risks aggrandizing the lower courts, it is the invitation to treat silence as license—to fill gaps in Supreme Court reasoning with our own assumptions about what must have been decided. That is not deference, it is judicial activism at its worst.

With these preliminary thoughts in mind, I now turn to the particular interim order in this matter.

As an inferior court, we are bound by the Supreme Court’s commands. Payne v. Taslimi , 998 F.3d 648, 655 n.4 (4th Cir. 2021). But a judicial opinion can be “binding” in several ways.

Of course, when the Supreme Court issues an opinion that contains analysis, we are bound both by 'the result' and by the reasoning that was 'necessary to that result.' Seminole Tribe of Fla. v. Florida , 517 U.S. 44, 67 (1996); see Payne , 998 F.3d at 655.

Page 81And even when a Supreme Court opinion contains no reasoning at all, it is binding as to the outcome of an issue in a particular case. See Ramos v. Louisiana , 590 U.S. 83, 104-05 (2020); Mandel v. Bradley , 432 U.S. 173, 176 (1977) (holding that summary affirmances “prevent lower courts from coming to opposite conclusions on the precise issues presented and necessarily decided by those actions”).

Yet those summary rulings “do not have the same precedential value as does an opinion of [the Supreme] Court after briefing and oral argument on the merits.' Lunding v. N.Y. Tax Appeals Tribunal , 522 U.S. 287, 307 (1998) (cleaned up). In part, that is because “[i]t is usually a judicial decision’s reasoning—its ratio decidendi -that allows it to have life and effect in the disposition of future cases.' Ramos , 590 U.S. at 104. Indeed, the Supreme Court has held that certain issues cannot be made precedential through implication, even if necessary to the judgment. See, e.g. , Ariz. Christian Sch. Tuition Org. v. Winn , 563 U.S. 125, 144-45 (2011) (holding that decisions omitting a jurisdictional discussion do “not stand for the proposition that no defect existed,” and the “Court would risk error if it relied on assumptions that have gone unstated and unexamined”).

Then, there is dicta. Dicta of the Supreme Court, 'although non-binding,' has “considerable persuasive value.” In re Bateman , 515 F.3d 272, 282 (4th Cir. 2008). At times, we have even treated 'carefully considered language of the Supreme Court' as “authoritative,” even if it is “technically dictum.” Wynne v. Town of Great Falls , 376 F.3d 292, 298 n.3 (4th Cir. 2004) (citation omitted).

The question here is how to treat the short stay order by the Supreme Court.

Page 82We are, of course, bound by the outcome of the order: The preliminary injunction is stayed. Our opinion on the timing of any injunctive relief is irrelevant—that has practical implications for our review because our decision today cannot change the fact that the injunction “cannot currently protect anyone from anything[.]” Maj. Op. at 19.

But as to binding reasoning, there is none in the order, which has only a few sentences to review. First, the order recites the legal standard under Nken v. Holder , 556 U.S. 418 (2009). Soc. Sec. Admin. v. Am. Fed’n of State, Cnty., & Mun. Emps. , 145 S. Ct. 1626, 1626–27 (2025). Then, it states the outcome:

After review, we determine that the application of these factors in this case warrants granting the requested stay. We conclude that, under the present circumstances, SSA may proceed to afford members of the SSA DOGE Team access to the agency records in question in order for those members to do their work.

Id. at 1627. Finally, it instructs us on the length of the stay, which will remain in place “pending the disposition of the appeal” in this Court and the “disposition of a petition for a writ of certiorari.' Id.

The entirety of the order’s reasoning is found in the phrases “[a]fter review” and “under the present circumstances.” It includes no analysis, or even an indication of which legal factors the Supreme Court considered relevant. There is simply nothing in the order that compels the outcome of our review here.

Of course, an interim order has legal effect—it binds us as to its result. But to say that an interim order has effect is not to say that it has reasoning. And without reasoning, its reach extends no further than its result.

Page 83For some of my colleagues, this silence speaks volumes. Specifically, in their concurring opinions, Judge Wilkinson and Judge Richardson conclude that the Supreme Court has told us that this particular stay order compels us to reverse on appeal. I disagree in two important respects.

First, my colleagues point to Trump v. Boyle , a stay order in an unrelated case, 5 to mean that we must reverse here. 145 S. Ct. 2653 (2025). But that is a more liberal reading than the text of that order can support.

In Boyle , the Court explained that its interim orders were “not conclusive as to the merits,” but rather should “inform how a court should exercise its equitable discretion in like cases.' 145 S. Ct. at 2654. There, the Court pointed to a previous interim order's conclusion about the government's risk of harm. Id. Here, we have no such bindingPage 84 'reflect[ions]' regarding the Court's judgment. Id. We are left, instead, to be as “informed” as the outcome of the order can make us.⁶ Id.

I also depart from my concurring colleagues on how they view the overlap of the standards for stays and preliminary injunctions more generally.

My colleagues assert that the question before this Court and the one that was before the Supreme Court are the same and that a “parallel” standard applies. See J. Richardson, Concurring Op. at 27 (asserting that the order 'already answered' the same question “before us in this very case”); J. Wilkinson, Concurring Op. at 23–24 (asserting that the “binding authority” of the stay order is “easy to see” because the standard we apply today is “parallel . . . to the one the Court employed in its stay order”).

But the stay order neither answered the same question nor applied the same standard as we do today.

The question answered by a stay order is whether a preliminary injunction should be enforceable pending appeal. That determination is within the discretion of the court Page 85granting or denying the stay, which will “operate[] upon the judicial proceeding itself” rather than on the parties. Nken , 556 U.S. at 428; see id. at 433.

The question answered by a preliminary injunction is whether a court should “tell[] someone what to do or not to do' pending litigation on the merits. See id. at 428 (explaining that a preliminary injunction is “directed at someone” and “governs that party’s conduct”). It is usually within the discretion of the district court. Frazier v. Prince George’s County , 86 F.4th 537, 543 (4th Cir. 2023).

In answering those questions, the standard that the Supreme Court applies during a stay application review does not match the standard we apply when reviewing the appeal of a preliminary injunction.

First, the Supreme Court’s review encompasses considerations that are unavailable to us. When it evaluates likelihood of success on the merits under Nken , it also considers its own future intentions, including whether there is “a reasonable probability” that it will grant certiorari and “a fair prospect” that the Court will reverse. Hollingsworth v. Perry , 558 U.S. 183, 190 (2010); see Does 1-3 v. Mills , 142 S. Ct. 17, 18 (2021) (Barrett, J., concurring) (explaining that this factor includes “a discretionary judgment about whether the Court should grant review in the case'). Without any explanation and before the Supreme Court’s review of the injunction, we cannot know if it based its decision on its intention to create new law. We, of course, must apply the current law. See Agostini v. Felton , 521 U.S. 203, 237 (1997).

Second, though the Nken and Winter factors have 'substantial overlap,' they are not 'one and the same[.]' Nken , 556 U.S. at 434. Consider, for example, success on the merits.

Page 86When considering a preliminary injunction, succeeding on the merits means succeeding on the ultimate merits of the case. See Winter v. Nat. Res. Def. Council, Inc. , 555 U.S. 7, 20 (2008). But a court considering a stay is two levels deep in its probabilistic determination, asked to divine the future from a crystal ball within a crystal ball: It predicts whether the stay applicant is likely to succeed on appeal by showing that the district court erred in applying the Winter factors, which themselves include a prediction about success on the ultimate merits. See Ruckelshaus v. Monsanto Co. , 463 U.S. 1315, 1316 (1983). Thus, the Supreme Court predicted—but did not decide—the result of the appeal of the preliminary injunction.

Most importantly, collapsing these standards would jettison “the whole idea” of a stay, which is “to hold the matter under review in abeyance” until the appellate court has “sufficient time to decide the merits” of the appeal. Nken , 556 U.S. at 432. Thus, a stay should “preserve the availability of a subsequent review,” not “render it redundant” by effectively deciding its outcome. Id. ; cf. Cook County v. Wolf , 962 F.3d 208, 234 (7th Cir. 2020) ('There would be no point in the merits stage if an issuance of a stay must be understood as a sub silentio disposition of the underlying dispute.').

Indeed, the stay order here explicitly contemplates our review. Am. Fed’n , 145 S. Ct. at 1627 (staying the preliminary injunction “pending the disposition of the appeal in the United States Court of Appeals for the Fourth Circuit”). Had the Supreme Court instead wanted to grant certiorari before judgment, it could have done so. Sup. Ct. Rule 11; see, e.g. , Noem v. Doe , -- S. Ct. --, 2026 WL 731088, at *1 (Mar. 16, 2026) (treating an Page 87application for stay as a petition for a writ of certiorari before judgment and granting the petition). It did not.

To be sure, a stay may be a strong signal that the panel will ultimately reverse the injunction. But it does not compel that outcome because the stay analysis involves 'predicting rather than deciding' the outcome of the appeal. E. Bay Sanctuary Covenant v. Biden , 993 F.3d 640, 660 (9th Cir. 2021); see, e.g. , Dist.⁴ Lodge of the Int'l Ass'n of Machinists v. Raimondo , 40 F.4th 36, 38 (1st Cir. 2022) (after the court granted a stay, “the handwriting was on the wall for the appeal itself,' yet 'nevertheless, the possibility remained” for success). In simpler terms, a signal, however strong, is not a command. Courts do not adjudicate by semaphore. Binding law speaks with words, not signals.

A review of stays and subsequent reviews in the intermediate appellate courts makes clear the absurdity of a stay that compels the outcome of an appeal. Indeed, many courts have subsequently affirmed a preliminary injunction that was previously stayed. See, e.g. , City & County of San Francisco v. U.S. Citizenship & Immigr. Servs. , 981 F.3d 742, 749, 763 (9th Cir. 2020); United States v. City of New Haven , 447 F.2d 972, 973 (2d Cir. 1971).

Ultimately, our discussion reflects a prudential concern that we not “cultivate the appearance” of defiance. CASA , 971 F.3d at 230. With rather strong rhetoric, my friend Judge Wilkinson says, “My colleagues jump too quickly onto the bandwagon of those who condemn each and every use of the Supreme Court’s emergency docket, no matter how justified.' J. Wilkinson, Concurring Op. at 25. But my view does not condemn the emergency docket—it simply declines to transform it. The emergency docket serves an Page 88important function—resolving urgent disputes. It does not, without more, create precedent. To insist upon that distinction is not a criticism, it is a classification.⁷

Here, we have an unclear signal from the Supreme Court about what it may do in the future. Whatever my colleagues may think about the merits of this particular preliminary injunction and the weight we may give to the Supreme Court’s stay, the order did not resolve this appeal entirely, and we are “duty-bound to give [it] a thorough and conscientious review.' CASA , 971 F.3d at 230.