dissenting:
The majority’s position rests on two abstractions: first, that a disability plan need not cover “potential risk of relapse,” and, second, that for disability purposes, “a physical condition such as a heart attack ... is different from the risk of relapse into drug use.” Ante at 357-58. Neither abstraction is grounded in law. The text of Continental’s plan designates addiction as a mental disorder and covers mental disorders so long as they make a claimant continuously unable to perform the duties of his previous occupation. All record evidence indicates that, because of his addiction, Stanford cannot return to work in anesthesiology with any reasonable degree of safety. The majority has in effect used its equitable power to authorize an unwritten exception to Continental’s textual promise of coverage, and thereby accomplished an uncommonly harsh result. I respectfully dissent.
I.
The textual case in Stanford’s favor is straightforward: “Disability,” the Plan states, “means that ... injury or sickness causes physical or mental impairment to such a degree of severity that you are: (1) continuously unable to perform the material and substantial duties of your regular occupation; and (2) not gainfully employed.” A claimant who satisfies that definition is entitled to benefits. As Continental states, “[tjhere is no dispute that Stanford has a chemical dependency and addiction,” Brief of Appellee at 10, or that drug addiction qualifies as a “sickness” that causes “mental impairment” (defined in the Plan as those disorders “found in the current diagnostic standards manual of the American Psychiatric Association,” which devotes a full section to substance-related disorders, addiction notably among them). Indeed, that is why Continental paid Stanford benefits for as long as it did. No one doubts that Stanford was unemployed when Continental cut off his benefits. And — remarkably—no one, not even Continental itself, denies that Stanford cannot with any safety perform the duties of his regular occupation as an anesthesia nurse; indeed, not a shred of contrary evidence was ever presented in this case. How, then, can it be that he is denied benefits?
The majority answers by quoting with approval a statement from Continental’s denial letter, “ ‘the policy does not cover potential risk,’ ” and adding on its own the phrase, “of relapse.” Ante at 358. But the phrase “potential risk” is a redundancy; “potential risk” is just risk. The majority’s addition, “of relapse,” might be read as a hmiting reference to addiction, *362or it might not; addicts are not the only medical patients who relapse. The majority offers no further explanation, and Continental’s explanation — “In sum and essence, a risk of relapse is not evidence of a current impairment; instead, it is a future, potential concern.” — just deepens the confusion. Brief of Appellee at 18-19. All agree that Stanford cannot presently return to work in safety, and if we ask why not, the answer must be some existing, not future, impairment — namely, Stanford’s fentanyl addiction.
The chief problem with excluding “potential risk of relapse” from coverage is that the exclusion has no support whatsoever in the language of the Plan. While the majority seeks to couch Continental’s decision as an exercise of discretion, an administrator lacks discretion to disregard the plain terms of its own plan. “The award of benefits under any ERISA plan,” this circuit has said, “is governed in the first instance by the language of the plan itself.” Lockhart v. United Mine Workers of Am.1974 Pension Trust, 5 F.3d 74, 78 (4th Cir.1993). The Plan’s stated definition of disability is functional and encompasses any injury or sickness whose effect is to make one continuously unable to work, as a grave medical risk certainly can. The Plan does contain an “Exclusions and Limitations” section that lists such things as elective cosmetic surgery and pre-existing conditions, but nothing in that list speaks to “potential risk.” In substance, then, the majority is permitting Continental to carve an unwritten exception out of the Plan’s textual promise of coverage. Our circuit does not ordinarily permit this sort of equitable improvisation in ERISA cases, especially when its result is so harsh. See 29 U.S.C. § 1102(a)(1) (2000) (“Every employee benefit plan shall be established and maintained pursuant to a written instrument.”); White v. Provident Life & Accident Ins. Co., 114 F.3d 26, 29 (4th Cir.1997) (rejecting “unwritten modifications of ERISA plans”). There is a fair way to make exceptions to an ERISA plan, and that is to write them down before a claimant comes asking for benefits.
A second problem, as the majority itself admits, is that “Continental’s policy would require [Stanford] to return to work and in fact suffer a relapse in order to qualify for long term disability benefits.” Ante at 359. Stanford’s whole job was to administer drugs to patients, including those very substances to which he is by all accounts addicted. In other words, someone such as Stanford who has struggled back from addiction must now succumb to it again. The majority acknowledges that this result creates a “somewhat troubling — some might say perverse — incentive structure,” and that Stanford “cannot return to his old job” in safety, but comforts itself that Stanford can work “countless other jobs.” Id. This reasoning totally disregards Plan language defining disability as the inability “to perform the material and substantial duties of your regular occupation.” Forcing Stanford to relapse into addiction or lose his benefits would also thwart the very purpose for which disability plans exist: to help people overcome medical adversity if possible, and otherwise to cope with it.
Finally, Continental’s unwritten exception would seem to exclude all medical conditions whose critical effect is to create grave medical risk, conditions that make doing one’s job, though not literally impossible, unreasonably dangerous. Continental’s meditations on “current impairment” versus “potential risk” imply as much, and the position would at least have the textual hook of an unforgiving interpretation of the word “unable.” But “[i]t is a basic tenet of insurance law that an insured is disabled when the activity in question *363would aggravate a serious condition affecting the insured’s health.” Lasser v. Reliance Standard Life Ins. Co., 146 F.Supp.2d 619, 628 (D.N.J.2001). The treatise definition of disability holds that “[t]he insured is considered to be permanently and totally disabled when it is impossible to work without hazarding his or her health or risking his or her life,” 31 John Alan Appleman, Appleman on Insurance § 187.05[A], at 214 (2d ed.2007), a proposition “sufficiently well-settled that in many jurisdictions it travels under the name of the ‘common care and prudence rule,’ ” Lasser, 146 F.Supp.2d at 628. Accord Oppenheim v. Finch, 495 F.2d 396, 398 (4th Cir.1974); 46 C.J.S. Insurance § 1551, at 445 (2007); 44 Am.Jur.2d Insurance § 1470, at 722 (2003).
This is only good sense. Some back conditions leave a patient literally able to lift heavy objects, but at risk of partial paralysis upon doing so; we would not deny disability benefits to a laborer with such a condition. And when busy professionals with cardiac troubles have brought ERISA suits because workplace stress caused a risk of heart attack, they have typically prevailed. See, e.g., Lasser v. Reliance Standard Life Ins. Co., 344 F.3d 381 (3d Cir.2003); Levinson v. Reliance Standard Life Ins. Co., 245 F.3d 1321 (11th Cir.2001). It would not be impossible for an ERISA administrator to buck this legal tradition. But doing so should require what is manifestly absent here— some basis in the text of the plan.
The majority’s response to this final problem gets to what may be the crux of its position. “[T]he risk of a heart attack is different from the risk of relapse into drug use,” the majority explains. Ante at 358. Whether an addict “succumbs to [ ] temptation remains his choice; the heart-attack prone doctor has no such choice” because his condition is “physical.” Id. at 358-59. In one sense, this passage is very welcome, for it would appear to limit an otherwise sweeping exclusion of medical conditions that cause “potential risk” to a more narrow exclusion of addictive relapse alone. In another sense, however, this passage is the most legally ungrounded yet, for it appears to rest on moral considerations of choice and temptation on the one hand, and medical considerations of physical inability on the other, neither of which are to be found in the language of a Plan that puts addiction squarely on all fours with other impairments.1 The moral and medical choices are not this court’s to make. They belong to those who bargained for the Plan — and who have something at stake in it.
II.
The majority presents Continental’s claim as an abstract one about the scope of the Plan’s coverage, and thus never contends with the facts. As the majority puts it, Continental “did not dispute the medical judgment” that returning to work put Stanford at risk of relapse, but rather made a “contractual, not medical” determination that “risk of relapse did not fall *364within the benefit plan’s definition of ‘disability.’ ” Ante at 360. Since I do not think risk of addictive relapse and other medical risk can categorically be excluded from coverage, the proper inquiry in my view is fact-intensive and focuses on a risk’s likelihood and gravity — as one might expect from a definition of disability that turns on an impairment’s “degree of severity.” See Lasser, 344 F.3d at 391 n. 12 (“[Wjhether risk of future effects creates a present disability depends on the probability of the future risk’s occurrence.”). It thus remains to scrutinize Stanford’s evidence and see if he can carry his burden of demonstrating such risk. See Gallagher v. Reliance Standard Life Ins. Co., 305 F.3d 264, 270 (4th Cir.2002) (noting that claimants bear the burden of proving disability).
Stanford has relapsed twice before. The first time was around October 2003, about a week after finishing his first treatment program and before he returned to work. After then enrolling in a more intense, ninety-day treatment program and returning to work (for the first time in almost six months), he promptly relapsed again. By May, he was in a third treatment program, which released him with a note from the treatment team stating that he should return to work only “with the restriction of not having access to narcotics.” Through the rest of 2004, he never returned to work, and stayed clean.
Every medical opinion in the record indicates that Stanford should not return to his job as an anesthesia nurse due to his risk of addictive relapse. First is the already-mentioned note from his third treatment team. Continental also asked Stanford’s treating physician, Dr. David Faulk, his opinion on the matter in a written “functional assessment” in August 2004. Asked to list “specific impairments in [Stanford’s] ability to function” (emphasis in original), Dr. Faulk wrote: “Pt cannot be around narcotics.” In December, Continental called with the same question, and Dr. Faulk repeated his concerns about Stanford’s potential for relapse.2 Finally, in January, Dr. Faulk wrote Continental a letter stating that Stanford is “unable to return to his regular duties as an anesthesia nurse. He cannot be subjected to controlled substances at this time.”
The record also contains an article, which Stanford submitted to Continental in the course of his administrative appeal, about the apparently common problem of anesthesiologists becoming addicted to the drugs they administer. See Eric. B. Hed-berg, Anesthesiologists: Addicted to the Drugs They Administer, ASA Newsletter (Am. Soc’y of Anesthesiologists, Park Ridge, 111.), May 2001. The article, authored by a medical director of an addiction treatment facility, states that only about half of opiate-addicted anesthesia personnel can return to their profession even after substantial treatment. It also contains a list of seven factors, any one of which indicates that an addicted anesthesia specialist should “[n]ever return to clinical anesthesiology.” Stanford underlines three of them: “Significant relapse despite adequate treatment,” “Lacks confidence to return to the operating room and not self-administer anesthetic drugs,” and “Significant Axis I or II psychopathology,” such as Stanford’s ongoing depression.
*365Finally, the narcotic to which Stanford became addicted, fentanyl, is a fearsome drug, which used properly has “an analgesic potency of about 80 times that of morphine” and used recreationally has “biological effects ... indistinguishable from those of heroin” but potentially “hundreds of times more potent.” Drug Enforcement Admin., U.S. Dep’t of Justice, Drugs of Abuse 25-26 (2005). Illicit use began among medical personnel. Id. Indeed, the federal courts have seen a number of disability disputes featuring an anesthesia specialist addicted to the drug. See, e.g., Shafer v. Preston Mem’l Hosp. Corp., 107 F.3d 274 (4th Cir.1997); Allen v. Minn. Life Ins. Co., 216 F.Supp.2d 1377 (N.D.Ga.2001); Laucks v. Provident Cos., No. 1CV971507, 1999 WL 33320463 (M.D.Pa. Oct.29, 1999); Holzer v. MBL Life Assurance Corp., No. 97 Civ. 5834(TPG), 1999 WL 649004 (S.D.N.Y. Aug.25, 1999); Vedernikov v. W. Va. Univ., 55 F.Supp.2d 518 (N.D.W.Va.1999).
To balance out the scale, Continental offers only repeated references to Stanford’s seven months of outpatient therapy and clean living prior to the benefits cutoff. As Stanford argues, this evidence standing alone is so scant as to violate ERISA regulations. See, e.g., 29 C.F.R. § 2560.503 — l(h)(3)(iii) (2007) (“[I]n deciding an appeal of any adverse benefit determination that is based in whole or in part on a medical judgment, ... the appropriate named fiduciary shall consult with a health care professional.... ”). Continental is so persistently abstract, to judge from the record, because it has nothing else to work with. Where a claimant presents substantial evidence of disability and an administrator presents almost nothing in response, the appropriate outcome is an award of benefits.
III.
My inquiry thus far has been an inquiry of law, for I think the moral opprobrium that underlies the special exclusion for drug addicts is not grounded in the language of the Plan or the evidence in this case. But if we do take up the moral issue, I believe my colleagues mistake the moral balance. Mr. Stanford is not currently taking drugs; he is trying to cease taking drugs. We should give people like him a chance to get back on their feet. To put him to the cruel choice of losing his disability benefits or returning to the environment that impelled his addiction is not right. Judge-made exceptions are often assumed to be humane, while law is thought to be a cold, hard thing. But equity here is a sword that strikes against the needy but unfavored. Law would be kinder.
. It is true that some lower courts have taken the majority’s view, but others have not, and no appellate court has yet addressed the issue. Compare Hellman v. Union Cent. Life Ins. Co., 175 F.Supp.2d 1044 (M.D.Tenn.2001) (holding that a recovering, substance-addicted anesthesiologist’s risk of relapse may render him unable to return to his profession, depending on the facts), and Brosnan v. Provident Life & Accident Ins. Co., 31 F.Supp.2d 460 (E.D.Pa.1998) (same), with Allen v. Minn. Life Ins. Co., 216 F.Supp.2d 1377 (N.D.Ga. 2001) (holding that a recovering, substance-addicted anesthesiologist's risk of relapse does not render him unable to return to his profession), and Laucks v. Provident Cos., No. 1CV971507, 1999 WL 33320463 (M.D.Pa. Oct.29, 1999) (same).
. The majority states that in December 2004, Dr. Faulk told Continental “that Stanford no longer suffered any impairment that would prevent him from performing the duties of his occupation as a nurse anesthetist.” Ante at •-•. This is baffling. The record contains the Continental investigator’s notes from the conversation, which state: “[Confirmed that Dr is saying that clmt has no impairment that would prevent him from doing his occ except for the potential for relapse.”