Justice Breyer
delivered the opinion of the Court.
In 1998, petitioner CIGNA Corporation changed the nature of its basic pension plan for employees. Previously, the plan provided a retiring employee with a defined benefit in the form of an annuity calculated on the basis of his preretirement salary and length of service. The new plan provided most retiring employees with a (lump sum) cash balance calculated on the basis of a defined annual contribution from CIGNA as increased by compound interest. Because many employees had already earned at least some old-plan benefits, the new plan translated already-earned benefits into an opening amount in the employee’s cash balance account.
Respondents, acting on behalf of approximately 25,000 beneficiaries of the CIGNA Pension Plan (which is also a petitioner here), challenged CIGNA’s adoption of the new plan. They claimed in part that CIGNA had failed to give them proper notice of changes to their benefits, particularly because the new plan in certain respects provided them with less generous benefits. See Employee Retirement Income Security Act of 1974 (ERISA), §§ 102(a), 104(b), 88 Stat. 841, [425]*425848, as amended, § 204(h), as added, 100 Stat. 243, and as amended, 29 U. S. C. §§ 1022(a), 1024(b), 1054(h).
The District Court agreed that the disclosures made by CIGNA violated its obligations under ERISA. In determining relief, the court found that CIGNA’s notice failures had caused the employees “likely harm.” The court then reformed the new plan and ordered CIGNA to pay benefits accordingly. It found legal authority for doing so in ERISA § 502(a)(1)(B), 29 U. S. C. § 1132(a)(1)(B) (authorizing a plan “participant or beneficiary” to bring a “civil action” to “recover benefits due to him under the terms of his plan”).
We agreed to decide whether the District Court applied the correct legal standard, namely, a “likely harm” standard, in determining that CIGNA’s notice violations caused its employees sufficient injury to warrant legal relief. To reach that question, we must first consider a more general matter — whether the ERISA section just mentioned (ERISA’s recovery-of-benefits-due provision, § 502(a)(1)(B)) authorizes entry of the relief the District Court provided. We conclude that it does not authorize this relief. Nonetheless, we find that a different equity-related ERISA provision, to which the District Court also referred, authorizes forms of relief similar to those that the court entered. § 502(a)(3), 29 U. S. C. § 1132(a)(3).
Section 502(a)(3) authorizes “appropriate equitable relief” for violations of ERISA. Accordingly, the relevant standard of harm will depend upon the equitable theory by which the District Court provides relief. We leave it to the District Court to conduct that analysis in the first instance, but we identify equitable principles that the court might apply on remand.
I
Because our decision rests in important part upon the circumstances present here, we shall describe those circumstances in some detail. We still simplify in doing so. But [426]*426the interested reader can find a more thorough description in two District Court opinions, which set forth that court’s findings reached after a lengthy trial. See 559 F. Supp. 2d 192 (Conn. 2008); 534 F. Supp. 2d 288 (Conn. 2008).
A
Under CIGNA’s pre-1998 defined-benefit retirement plan, an employee with at least five years’ service would receive an annuity annually paying an amount that depended upon the employee’s salary and length of service. Depending on when the employee had joined CIGNA, the annuity would equal either (1) 2 percent of the employee’s average salary over his final three years with CIGNA, multiplied by the number of years worked (up to 30); or (2) l2/3 percent of the employee’s average salary over his final five years with CIGNA, multiplied by the number of years worked (up to 35). Calculated either way, the annuity would approach 60 percent of a longtime employee’s final salary. A well-paid longtime employee, earning, say, $160,000 per year, could receive a retirement annuity paying the employee about $96,000 per year until his death. The plan offered many employees at least one other benefit: They could retire early, at age 55, and receive an only-somewhat-redueed annuity.
In November 1997, CIGNA sent its employees a newsletter announcing that it intended to put in place a new pension plan. The new plan would substitute an “account balance plan” for CIGNA’s pre-existing defined-benefit system. App. 991a (emphasis deleted). The newsletter added that the old plan would end on December 31, 1997, that CIGNA would introduce (and describe) the new plan sometime during 1998, and that the new plan would apply retroactively to January 1, 1998.
Eleven months later CIGNA filled in the details. Its new plan created an individual retirement account for each employee. (The account consisted of a bookkeeping entry backed by a CIGNA-funded trust.) Each year CIGNA [427]*427would contribute to the employee’s individual account an amount equal to between 3 percent and 8.5 percent of the employee’s salary, depending upon age, length of service, and certain other factors. The account balance would earn compound interest at a rate equal to the return on 5-year treasury bills plus one-quarter percent (but no less than 4.5 percent and no greater than 9 percent). Upon retirement the employee would receive the amount then in his or her individual account — in the form of either a lump sum or whatever annuity the lump sum then would buy. As promised, CIGNA would open the accounts and begin to make contributions as of January 1, 1998.
But what about the retirement benefits that employees had already earned prior to January 1,1998? CIGNA promised to make an initial contribution to the individual’s account equal to the value of that employee’s already-earned benefits. And the new plan set forth a method for calculating that initial contribution. The method consisted of calculating the amount as of the employee’s (future) retirement date of the annuity to which the employee’s salary and length of service already (i. e., as of December 31,1997) entitled him and then discounting that sum to its present (1 e., January 1, 1998) value.
An example will help: Imagine an employee born on January 1, 1966, who joined CIGNA in January 1991 on his 25th birthday, and who (during the five years preceding the plan changeover) earned an average salary of $100,000 per year. As of January 1,1998, the old plan would have entitled that employee to an annuity equal to $100,000 times 7 (years then worked) times 1% percent, or $11,667 per year — when he retired in 2031 at age 65. The 2031 price of an annuity paying $11,667 per year until death depends upon interest rates and mortality assumptions at that time. If we assume the annuity would pay 7 percent until the holder’s death (and we use the mortality assumptions used by the plan, see App. 407a (incorporating the mortality table prescribed by Rev. [428]*428Rul. 95-6,1995-1 Cum. Bull. 80)), then the 2081 price of such an annuity would be about $120,500. And CIGNA should initially deposit in this individual’s account on January 1, 1998, an amount that will grow to become $120,500, 33 years later, in 2031, when the individual retires. If we assume a 5 percent average interest rate, then that amount presently (i. e., as of January 1,1998) equals about $24,000. And (with one further mortality-related adjustment that we shall describe infra, at 429-430) that is the amount, more or less, that the new plan’s transition rules would have required CIGNA initially to deposit. Then CIGNA would make further annual deposits, and all the deposited amounts would earn compound interest. When the employee retired, he would receive the resulting lump sum.
The new plan also provided employees a guarantee: An employee would receive upon retirement either (1) the amount to which he or she had become entitled as of January 1, 1998, or (2) the amount then in his or her individual account, whichever was greater. Thus, the employee in our example would receive (in 2031) no less than an annuity paying $11,667 per year for life.
B
The District Court found that CIGNA’s initial descriptions of its new plan were significantly incomplete and misled its employees. In November 1997, for example, CIGNA sent the employees a newsletter that said the new plan would “significantly enhance” its “retirement program,” would produce “an overall improvement in... retirement benefits,” and would provide “the same benefit security” with “steadier benefit growth.” App. 990a, 991a, 993a. CIGNA also told its employees that they would “see the growth in [their] total retirement benefits from CIGNA every year,” id., at 952a, that its initial deposit “represented] the full value of the benefit [they] earned for service before 1998,” Record E-503 [429]*429(Exh. 98), and that “[o]ne advantage the company will not get from the retirement program changes is cost savings,” App. 993a.
In fact, the new plan saved the company $10 million annually (though CIGNA later said it devoted the savings to other employee benefits). Its initial deposit did not “represen[t] the full value of the benefit” that employees had “earned for service before 1998.” And the plan made a significant number of employees worse off in at least the following specific ways:
First, the initial deposit calculation ignored the fact that the old plan offered many CIGNA employees the right to retire early (beginning at age 55) with only-somewhat-reduced benefits. This right was valuable. For example, as of January 1, 1998, respondent Janice Amara had earned vested age-55 retirement benefits of $1,833 per month, but CIGNA’s initial deposit in her new-plan individual retirement account (ignoring this benefit) would have allowed her at age 55 to buy an annuity benefit of only $900 per month.
Second, as we previously indicated but did not explain, supra, at 428, the new plan adjusted CIGNA’s initial deposit downward to account for the fact that, unlike the old plan’s lifetime annuity, an employee’s survivors would receive the new plan’s benefits (namely, the amount in the employee’s individual account) even if the employee died before retiring. The downward adjustment consisted of multiplying the otherwise-required deposit by the probability that the employee would live until retirement — a 90 percent probability in the example of our 32-year-old, supra, at 427-428. And that meant that CIGNA’s initial deposit in our example — the amount that was supposed to grow to $120,500 by 2031 — would be less than $22,000, not $24,000 (the number we computed). The employee, of course, would receive a benefit in return — namely, a form of life insurance. But at least some employees might have preferred the retirement [430]*430benefit and consequently could reasonably have thought it important to know that the new plan traded away one-tenth of their already-earned benefits for a life insurance policy that they might not have wanted.
Third, the new plan shifted the risk of a fall in interest rates from CIGNA to its employees. Under the old plan, CIGNA had to buy a retiring employee an annuity that paid a specified sum irrespective of whether falling interest rates made it more expensive for CIGNA to pay for that annuity. And falling interest rates also meant that any sum CIGNA set aside to buy that annuity would grow more slowly over time, thereby requiring CIGNA to set aside more money to make any specific sum available at retirement. Under the new plan CIGNA did not have to buy a retiring employee an annuity that paid a specific sum. The employee would simply receive whatever sum his account contained. And falling interest rates meant that the account’s lump sum would earn less money each year after the employee retired. Annuities, for example, would become more expensive (any fixed purchase price paying for less annual income). At the same time falling interest meant that the individual account would grow more slowly over time, leaving the employee with less money at retirement.
Of course, interest rates might rise instead of fall, leaving CIGNA’s employees better off under the new plan. But the latter advantage does not cancel out the former disadvantage, for most individuals are risk averse. And that means that most of CIGNA’s employees would have preferred that CIGNA, rather than they, bear these risks.
The amounts likely involved are significant. If, in our example, interest rates between 1998 and 2031 averaged 4 percent rather than the 5 percent we assumed, and if in 2031 annuities paid 6 percent rather than the 7 percent we assumed, then CIGNA would have had to make an initial deposit of $35,500 (not $24,000) to assure that employee the [431]*431$11,667 annual annuity payment to which he had already become entitled. Indeed, that $24,000 that CIGNA would have contributed (leaving aside the life-insurance problem) would have provided enough money to buy (in 2031) an annuity that assured the employee an annual payment of only about $8,000 (rather than $11,667).
We recognize that the employee in our example (like others) might have continued to work for CIGNA after January 1,1998; and he would thereby eventually have earned a pension that, by the time of his retirement, was worth far more than $11,667. But that is so because CIGNA made an additional contribution for each year worked after January 1, 1998. If interest rates fell (as they did), it would take the employee several additional years of work simply to catch up (under the new plan) to where he had already been (under the old plan) as of January 1,1998 — a phenomenon known in pension jargon as “wear away,” see 534 F. Supp. 2d, at 303-304 (referring to respondents’ requiring 6 to 10 years to catch up).
The District Court found that CIGNA told its employees nothing about any of these features of the new plan — which individually and together made clear that CIGNA’s descriptions of the plan were incomplete and inaccurate. The District Court also found that CIGNA intentionally misled its employees. A focus group and many employees asked CIGNA, for example, to “ ‘[djisclose details’ ” about the plan, to provide “‘individual comparisons,’” or to show “‘[a]n actual projection for retirement.’” Id., at 342. But CIGNA did not do so. Instead (in the words of one internal document), it “ ‘focus[ed] on NOT providing employees before and after samples of the Pension Plan changes.’ ” Id., at 343.
The District Court concluded, as a matter of law, that CIGNA’s representations (and omissions) about the plan, made between November 1997 (when it announced the plan) and December 1998 (when it put the plan into effect) violated:
[432]*432(1) ERISA §204(h), implemented by Treas. Reg. § 1.411(d)-6, 26 CFR § 1.411(d)-6 (2000), which (as it existed at the relevant time) forbade an amendment of a pension plan that would “provide for a significant reduction in the rate of future benefit accrual” unless the plan administrator also sent a “written notice” that provided either the text of the amendment or summarized its likely effects, 29 U. S. C. § 1054(h) (2000 ed.) (amended 2001); Treas. Reg. § 1.411(d) — 6, Q&A-10, 63 Fed. Reg. 68682 (1998); and
(2) ERISA §§ 102(a) and 104(b), which require a plan administrator to provide beneficiaries with summary plan descriptions and with summaries of material modifications, “written in a maimer calculated to be understood by the average plan participant,” that are “sufficiently accurate and comprehensive to reasonably apprise such participants and beneficiaries of their rights and obligations under the plan,” 29 U. S. C. §§ 1022(a), 1024(b) (2006 ed. and Supp. III).
The District Court then turned to the remedy. First, the court agreed with CIGNA that only employees whom CIGNA’s disclosure failures had harmed could obtain relief. But it did not require each individual member of the relevant CIGNA employee class to show individual injury. Rather, it found (1) that the evidence presented had raised a presumption of “likely harm” suffered by the members of the relevant employee class, and (2) that CIGNA, though free to offer contrary evidence in respect to some or all of those employees, had failed to rebut that presumption. It concluded that this unrebutted showing was sufficient to warrant class-applicable relief.
Second, the court noted that § 204(h) had been interpreted by the Second Circuit to permit the invalidation of plan amendments not preceded by a proper notice, prior to the 2001 amendment that made this power explicit. 559 F. Supp. 2d, at 207 (citing Frommert v. Conkright, [433]*433433 F. 3d 254, 263 (2006)); see 29 U. S. C. § 1054(h)(6) (2006 ed.) (entitling participants to benefits “without regard to [the] amendment” in case of an “egregious failure”). But the court also thought that granting this relief here would harm, not help, the injured employees. That is because the notice failures all concerned the new plan that took effect in December 1998. The court thought that the notices in respect to the freezing of old-plan benefits, effective December 31, 1997, were valid. To strike the new plan while leaving in effect the frozen old plan would not help CIGNA’s employees.
The court considered treating the November 1997 notice as a sham or treating that notice and the later 1998 notices as part and parcel of a single set of related events. But it pointed out that respondents “ha[d] argued none of these things.” 559 F. Supp. 2d, at 208. And it said that the court would “not make these arguments now on [respondents’] behalf.” Ibid.
Third, the court reformed the terms of the new plan’s guarantee. It erased the portion that assured participants who retired the greater of “A” (that which they had already earned as of December 31,1997, under the old plan, $11,667 in our example) or “B” (that which they would earn via CIGNA’s annual deposits under the new plan, including CIGNA’s initial deposit). And it substituted a provision that would guarantee each employee “A” (that which they had already earned, as of December 31, 1997, under the old plan) plus “B” (that which they would earn via CIGNA’s annual deposits under the new plan, excluding CIGNA’s initial deposit). In our example, the District Court’s remedy would no longer force our employee to choose upon retirement either an $11,667 annuity or his new-plan benefits (including both CIGNA’s annual deposits and CIGNA’s initial deposit). It would give him an $11,667 annuity plus his new-plan benefits (with CIGNA’s annual deposits but without CIGNA’s initial deposit).
[434]*434Fourth, the court “order[ed] and enjoin[ed] the CIGNA Plan to reform its records to reflect that all class members . . . now receive [the just described] A + B’ benefits,” and that it pay appropriate benefits to those class members who had already retired. Id., at 222.
Fifth, the court held that ERISA § 502(a)(1)(B) provided the legal authority to enter this relief. That provision states that a “civil action may be brought” by a plan “participant or beneficiary ... to recover benefits due to him under the terms of his plan.” 29 U. S. C. § 1132(a)(1)(B). The court wrote that its orders in effect awarded “benefits under the terms of the plan” as reformed. 559 F. Supp. 2d, at 212.
At the same time the court considered whether ERISA § 502(a)(3) also provided legal authority to enter this relief. That provision states that a civil action may be brought
“by a participant, beneficiary, or fiduciary (A) to enjoin any act or practice which violates any provision of this subchapter or the terms of the plan, or (B) to obtain other appropriate equitable relief (i) to redress such violations or (ii) to enforce any provisions of this subchapter or the terms of the plan.” 29 U. S. C. § 1132(a)(3) (emphasis added).
The District Court decided not to answer this question because (1) it had just decided that the same relief was available under § 502(a)(1)(B), regardless, cf. Varity Corp. v. Howe, 516 U. S. 489, 515 (1996); and (2) the Supreme Court has “issued several opinions ... that have severely curtailed the kinds of relief that are available under § 502(a)(3),” 559 F. Supp. 2d, at 205 (citing Sereboff v. Mid Atlantic Medical Services, Inc., 547 U. S. 356 (2006); Great-West Life & Annuity Ins. Co. v. Knudson, 534 U. S. 204 (2002); and Mertens v. Hewitt Associates, 508 U. S. 248 (1993)).
The parties cross-appealed the District Court’s judgment. The Court of Appeals for the Second Circuit issued a brief [435]*435summary order, rejecting all their claims, and affirming “the judgment of the district court for substantially the reasons stated” in the District Court’s “well-reasoned and scholarly opinions.” 348 Fed. Appx. 627 (2009). The parties filed cross-petitions for writs of certiorari in this Court. We granted the request in CIGNA’s petition to consider whether a showing of “likely harm” is sufficient to entitle plan participants to recover benefits based on faulty disclosures.
II
CIGNA in the merits briefing raises a preliminary question. Brief for Petitioners 13-20. It argues first and foremost that the statutory provision upon which the District Court rested its orders, namely, the provision for recovery of plan benefits, § 502(a)(1)(B), does not in fact authorize the District Court to enter the kind of relief it entered here. And for that reason, CIGNA argues, whether the District Court did or did not use a proper standard for determining harm is beside the point. We believe that this preliminary question is closely enough related to the question presented that we shall consider it at the outset.
The District Court ordered relief in two steps. Step 1: It ordered the terms of the plan reformed (so that they provided an “A plus B,” rather than a “greater of A or B” guarantee). Step 2: It ordered the plan administrator (which it found to be CIGNA) to enforce the plan as reformed. One can fairly describe step 2 as consistent with § 502(a)(1)(B), for that provision grants a participant the right to bring a civil action to “recover benefits due . . . under-the terms of his plan.” 29 U. S. C. § 1132(a)(1)(B). And step 2 orders recovery of the benefits provided by the “terms of [the] plan” as reformed.
But what about step 1? Where does § 502(a)(1)(B) grant a court the power to change the terms of the plan as they [436]*436previously existed? The statutory language speaks of “enforc[ing]” the “terms of the plan,” not of changing them. 29 U. S. C. § 1132(a)(1)(B) (emphasis added). The provision allows a court to look outside the plan’s written language in deciding what those terms are, i.e., what the language means. See UNUM Life Ins. Co. of America v. Ward, 526 U. S. 358, 377-379 (1999) (permitting the insurance terms of an ERISA-governed plan to be interpreted in light of state insurance rules). But we have found nothing suggesting that the provision authorizes a court to alter those terms, at least not in present circumstances, where that change, akin to the reform of a contract, seems less like the simple enforcement of a contract as written and more like an equitable remedy. See infra, at 441.
Nor can we accept the Solicitor General’s alternative rationale seeking to justify the use of this provision. The Solicitor General says that the District Court did enforce the plan’s terms as written, adding that the “plan” includes the disclosures that constituted the summary plan descriptions. In other words, in the view of the Solicitor General, the terms of the summaries are terms of the plan.
Even if the District Court had viewed the summaries as plan “terms” (which it did not, see supra, at 433), however, we cannot agree that the terms of statutorily required plan summaries (or summaries of plan modifications) necessarily may be enforced (under § 502(a)(1)(B)) as the terms of the plan itself. For one thing, it is difficult to square the Solicitor General’s reading of the statute with ERISA § 102(a), the provision that obliges plan administrators to furnish summary plan descriptions. The syntax of that provision, requiring that participants and beneficiaries be advised of their rights and obligations “under the plan,” suggests that the information about the plan provided by those disclosures is not itself part of the plan. See 29 U. S. C. § 1022(a). Nothing in § 502(a)(1)(B) (or, as far as we can tell, anywhere else) suggests the contrary.
[437]*437Nor do we find it easy to square the Solicitor General’s reading with the statute’s division of authority between a plan’s sponsor and the plan’s administrator. The plan’s sponsor (e. g., the employer), like a trust’s settlor, creates the basic terms and conditions of the plan, executes a written instrument containing those terms and conditions, and provides in that instrument “a procedure” for making amendments. §402, 29 U. S. C. §1102. The plan’s administrator, a trustee-like fiduciary, manages the plan, follows its terms in doing so, and provides participants with the summary documents that describe the plan (and modifications) in readily understandable form. §§3(21)(A), 101(a), 102, 104, 29 U. S. C. §§ 1002(21)(A), 1021(a), 1022, 1024 (2006 ed. and Supp. III). Here, the District Court found that the same entity, CIGNA, filled both roles. See 534 F. Supp. 2d, at 331. But that is not always the case. Regardless, we have found that ERISA carefully distinguishes these roles. See, e. g., Varity Corp., 516 U. S., at 498. And we have no reason to believe that the statute intends to mix the responsibilities by giving the administrator the power to set plan terms indirectly by including them in the summary plan descriptions. See Curtiss-Wright Corp. v. Schoonejongen, 514 U. S. 73, 81-85 (1995).
Finally, we find it difficult to reconcile the Solicitor General’s interpretation with the basic summary plan description objective: clear, simple communication. See §§2(a), 102(a), 29 U. S. C. § 1001(a), 1022(a) (2006 ed.). To make the language of a plan summary legally binding could well lead plan administrators to sacrifice simplicity and comprehensibility in order to describe plan terms in the language of lawyers. Consider the difference between a will and the summary of a will or between a property deed and its summary. Consider, too, the length of Part I of this opinion, and then consider how much longer Part I would have to be if we had to include all the qualifications and nuances that a plan drafter [438]*438might have found important and feared to omit lest they lose all legal significance. The District Court’s opinions take up 109 pages of the Federal Supplement. None of this is to say that plan administrators can avoid providing complete and accurate summaries of plan terms in the manner required by ERISA and its implementing regulations. But we fear that the Solicitor General’s rule might bring about complexity that would defeat the fundamental purpose of the summaries.
For these reasons taken together we conclude that the summary documents, important as they are, provide communication with beneficiaries about the plan, but that their statements do not themselves constitute the terms of the plan for purposes of § 502(a)(1)(B). We also conclude that the District Court could not find authority in that section to reform CIGNA’s plan as written.
If § 502(a)(1)(B) does not authorize entry of the relief here at issue, what about nearby § 502(a)(3)? That provision allows a participant, beneficiary, or fiduciary “to obtain other appropriate equitable relief” to redress violations of (here relevant) parts of ERISA “or the terms of the plan.” 29 U. S. C. § 1132(a)(3) (emphasis added). The District Court strongly implied, but did not directly hold, that it would base its relief upon this subsection were it not for (1) the fact that the preceding “plan benefits due” provision, § 502(a)(1)(B), provided sufficient authority; and (2) certain cases from this Court that narrowed the application of the term “appropriate equitable relief,” see, e. g., Mertens, 508 U. S. 248; Great-West, 534 U. S. 204. Our holding in Part II-A, supra, removes the District Court’s first obstacle. And given the likelihood that, on remand, the District Court will turn to and rely upon this alternative subsection, we consider the court’s second concern. We find that concern misplaced.
[439]*439We have interpreted the term “appropriate equitable relief” in §502(a)(3) as referring to “‘those categories of relief’ ” that, traditionally speaking (i. e., prior to the merger of law and equity), “‘were typically available in equity.’” Sereboff, 547 U. S., at 361 (quoting Mertens, 508 U. S., at 256). In Mertens, we applied this principle to a claim seeking money damages brought by a beneficiary against a private firm that provided a trustee with actuarial services. We found that the plaintiff sought “nothing other than compensatory damages” against a nonfiduciary. Id., at 253, 255 (emphasis deleted). And we held that such a claim, traditionally speaking, was legal, not equitable, in nature. Id., at 255.
In Great-West, we considered a claim brought by a fiduciary against a tort-award-winning beneficiary seeking monetary reimbursement for medical outlays that the plan had previously made on the beneficiary’s behalf. We noted that the fiduciary sought to obtain a lien attaching to (or a constructive trust imposed upon) money that the beneficiary had received from the tort-case defendant. But we noted that the money in question was not the “particular” money that the tort defendant had paid. And, traditionally speaking, relief that sought a lien or a constructive trust was legal relief, not equitable relief, unless the funds in question were “particular funds or property in the defendant’s possession.” 534 U. S., at 213 (emphasis added).
The case before us concerns a suit by a beneficiary against a plan fiduciary (whom ERISA typically treats as a trustee) about the terms of a plan (which ERISA typically treats as a trust). See LaRue v. DeWolff, Boberg & Associates, Inc., 552 U. S. 248, 253, n. 4 (2008); Varity Corp., supra, at 496-497. It is the kind of lawsuit that, before the merger of law and equity, respondents could have brought only in a court of equity, not a court of law. 4 A. Scott, W. Fratcher, & M. Ascher, Trusts §24.1, p. 1654 (5th ed. 2007) (hereinafter Scott & Ascher) (“Trusts are, and always have been, the bai[440]*440liwiek of the courts of equity”); Duvall v. Craig, 2 Wheat. 45, 56 (1817) (a trustee was “only suable in equity”).
With the exception of the relief now provided by § 502(a)(1)(B), Restatement (Second) of Trusts §§ 198(l)-(2) (1957) (hereinafter Second Restatement); 4 Scott & Ascher §24.2.1, the remedies available to those courts of equity were traditionally considered equitable remedies, see Second Restatement § 199; J. Adams, Doctrine of Equity: A Commentary on the Law as Administered by the Court of Chancery 61 (7th Am. ed. 1881) (hereinafter Adams); 4 Scott & Ascher § 24.2.
The District Court’s affirmative and negative injunctions obviously fall within this category. Mertens, supra, at 256 (identifying injunctions, mandamus, and restitution as equitable relief). And other relief ordered by the District Court resembles forms of traditional equitable relief. That is because equity chancellors developed a host of other “distinctively equitable” remedies — remedies that were “fitted to the nature of the primary right” they were intended to protect. 1 S. Symons, Pomeroy’s Equity Jurisprudence § 108, pp. 139-140 (5th ed. 1941) (hereinafter Pomeroy). See generally 1 J. Story, Commentaries on Equity Jurisprudence § 692 (12th ed. 1877) (hereinafter Story). Indeed, a maxim of equity states that “[ejquity suffers not a right to be without a remedy.” R. Francis, Maxims of Equity 29 (1st Am. ed. 1823). And the relief entered here, insofar as it does not consist of injunctive relief, closely resembles three other traditional equitable remedies.
First, what the District Court did here may be regarded as the reformation of the terms of the plan, in order to remedy the false or misleading information CIGNA provided. The power to reform contracts (as contrasted with the power to enforce contracts as written) is a traditional power of an equity court, not a court of law, and was used to prevent fraud. See Baltzer v. Raleigh & Augusta R. Co., 115 U. S. 634, 645 (1885) (“[I]t is well settled that equity would reform [441]*441the contract, and enforce it, as reformed, if the mistake or fraud were shown”); Hearne v. Marine Ins. Co., 20 Wall. 488, 490 (1874) (“The reformation of written contracts for fraud or mistake is an ordinary head of equity jurisdiction”); Bradford v. Union Bank of Tenn., 13 How. 57, 66 (1852); J. Eaton, Handbook of Equity Jurisprudence § 306, p. 618 (1901) (hereinafter Eaton) (courts of common law could only void or enforce, but not reform, a contract); 4 Pomeroy § 1375, at 1000 (reformation “chiefly occasioned by fraud or mistake,” which were themselves concerns of equity courts); 1 Story § § 152— 154; see also 4 Pomeroy § 1375, at 999 (equity often considered reformation a “preparatory step” that “establishes the real contract”).
Second, the District Court’s remedy essentially held CIGNA to what it had promised, namely, that the new plan would not take from its employees benefits they had already accrued. This aspect of the remedy resembles estoppel, a traditional equitable remedy. See, e. g., E. Merwin, Principles of Equity and Equity Pleading §910 (H. Merwin ed. 1895); 3 Pomeroy §804. Equitable estoppel “operates to place the person entitled to its benefit in the same position he would have been in had the representations been true.” Eaton § 62, at 176. And, as Justice Story long ago pointed out, equitable estoppel “forms a very essential element in... fair dealing, and rebuke of all fraudulent misrepresentation, which it is the boast of courts of equity constantly to promote.” 2 Story § 1533, at 776.
Third, the District Court injunctions require the plan administrator to pay to already retired beneficiaries money owed them under the plan as reformed. But the fact that this relief takes the form of a money payment does not remove it from the category of traditionally equitable relief. Equity courts possessed the power to provide relief in the form of monetary “compensation” for a loss resulting from a trustee’s breach of duty, or to prevent the trustee's unjust enrichment. Restatement (Third) of Trusts § 95, and Com[442]*442ment a (Tent. Draft No. 5, Mar. 2, 2009) (hereinafter Third Restatement); Eaton §§211-212, at 440. Indeed, prior to the merger of law and equity this kind of monetary remedy against a trustee, sometimes called a “surcharge,” was “exclusively equitable.” Princess Lida of Thurn and Taxis v. Thompson, 305 U. S. 456, 464 (1939); Third Restatement § 95, and Comment a; G. Bogert & G. Bogert, Trusts and Trustees § 862 (rev. 2d ed. 1995) (hereinafter Bogert); 4 Scott & Ascher §§ 24.2, 24.9, at 1659-1660, 1686; Second Restatement § 197; see also Manhattan Bank of Memphis v. Walker, 130 U. S. 267, 271 (1889) (“The suit is plainly one of equitable cognizance, the .bill being filed to charge the defendant, as a trustee, for a breach of trust”); 1 J. Perry, A Treatise on the Law of Trusts and Trustees § 17, p. 13 (2d ed. 1874) (common-law attempts “to punish trustees for a breach of trust in damages, .. . w[ere] soon abandoned”).
The surcharge remedy extended to a breach of trust committed by a fiduciary encompassing any violation of a duty imposed upon that fiduciary. See Second Restatement § 201; Adams 59; 4 Pomeroy § 1079; 2 Story §§ 1261, 1268. Thus, insofar as an award of make-whole relief is concerned, the fact that the defendant in this case, unlike the defendant in Mertens, is analogous to a trustee makes a critical difference. See 508 U. S., at 262-263. In sum, contrary to the District Court's fears, the types of remedies the court entered here fall within the scope of the term “appropriate equitable relief” in § 502(a)(3).
Ill
Section 502(a)(3) invokes the equitable powers of the District Court. We cannot know with certainty which remedy the District Court understood itself to be imposing, nor whether the District Court will find it appropriate to exercise its discretion under § 502(a)(3) to impose that remedy on remand. We need not decide which remedies are appropriate on the facts of this case in order to resolve the parties’ dispute as to the appropriate legal standard in determin[443]*443ing whether members of the relevant employee class were injured.
The relevant substantive provisions of ERISA do not set forth any particular standard for determining harm. They simply require the plan administrator to write and to distribute written notices that are “sufficiently accurate and comprehensive to reasonably apprise” plan participants and beneficiaries of “their rights and obligations under the plan.” § 102(a); see also §§ 104(b), 204(h). Nor can we find a definite standard in the ERISA provision, § 502(a)(3) (which authorizes the court to enter “appropriate equitable relief” to redress ERISA “violations”). Hence any requirement of harm must come from the law of equity.
Looking to the law of equity, there is no general principle that “detrimental reliance” must be proved before a remedy is decreed. To the extent any such requirement arises, it is because the specific remedy being contemplated imposes such a requirement. Thus, as CIGNA points out, when equity courts used the remedy of estoppel, they insisted upon a showing akin to detrimental reliance, i e., that the defendant’s statement “in truth, influenced the conduct of” the plaintiff, causing “prejudic[e].” Eaton §61, at 175; see 3 Pomeroy § 805. Accordingly, when a court exercises its authority under § 502(a)(3) to impose a remedy equivalent to estoppel, a showing of detrimental reliance must be made.
But this showing is not always necessary for other equitable remedies. Equity courts, for example, would reform contracts to reflect the mutual understanding of the contracting parties where “fraudulent suppression^], omission[s], or insertion[s],” 1 Story §154, at 149, “materially] . . . affect[ed]” the “substance” of the contract, even if the “complaining part[y]” was negligent in not realizing its mistake, as long as its negligence did not fall below a standard of “reasonable prudence” and violate a legal duty, 3 Pomeroy §§ 856, 856b, at 334, 340-341. See Baltzer, 115 U. S., at 645; Eaton § 307(b).
[444]*444Nor did equity courts insist upon a showing of detrimental reliance in cases where they ordered “surcharge.” Rather, they simply ordered a trust or beneficiary made whole following a trustee’s breach of trust. In such instances equity courts would “mold the relief to protect the rights of the beneficiary according to the situation involved.” Bogert §861, at 4. This flexible approach belies a strict requirement of “detrimental reliance.”
To be sure, just as a court of equity would not surcharge a trustee for a nonexistent harm, 4 Scott & Ascher §24.9, a fiduciary can be surcharged under § 502(a)(3) only upon a showing of actual harm — proved (under the default rule for civil cases) by a preponderance of the evidence. That actual harm may sometimes consist of detrimental reliance, but it might also come from the loss of a right protected by ERISA or its trust-law antecedents. In the present case, it is not difficult to imagine how the failure to provide proper summary information, in violation of the statute, injured employees even if they did not themselves act in reliance on summary documents — which they might not themselves have seen — for they may have thought fellow employees, or informal workplace discussion, would have let them know if, say, plan changes would likely prove harmful. We doubt that Congress would have wanted to bar those employees from relief.
The upshot is that we can agree with CIGNA only to a limited extent. We believe that, to obtain relief by surcharge for violations of §§ 102(a) and 104(b), a plan participant or beneficiary must show that the violation injured him or her. But to do so, he or she need only show harm and causation. Although it is not always necessary to meet the more rigorous standard implicit in the words “detrimental reliance,” actual harm must be shown.
We are not asked to reassess the evidence. And we are not asked about the other prerequisites for relief. We are asked about the standard of prejudice. And we conclude [445]*445that the standard of prejudice must be borrowed from equitable principles, as modified by the obligations and injuries identified by ERISA itself. Information-related circumstances, violations, and injuries are potentially too various in nature to insist that harm must always meet that more vigorous “detrimental harm” standard when equity imposed no such strict requirement.
IV
We have premised our discussion in Part III on the need for the District Court to revisit its determination of an appropriate remedy for the violations of ERISA it identified. Whether or not the general principles we have discussed above are properly applicable in this case is for it or the Court of Appeals to determine in the first instance. Because the District Court has not determined if an appropriate remedy may be imposed under § 502(a)(3), we must vacate the judgment below and remand this case for further proceedings consistent with this opinion.
It is so ordered.
Justice Sotomayor took no part in the consideration or decision of this case.