WALD, Circuit Judge:
Appellant Ralph W. McGehee is a former Central Intelligence Agency (CIA) officer. When he joined the CIA, McGehee signed an agreement that on its face bars him from revealing classified information without prior CIA approval. After the CIA censored portions of a manuscript he wrote, McGehee sought a declaratory judgment that the CIA classification and censorship scheme violates the first amendment and that, even if the scheme were constitutional, his article contains no properly classified material. The district court rejected McGehee’s first amendment challenge, and found, after giving deference to the CIA’s judgment, that the CIA had properly classified the censored materials.
We affirm. The CIA classification and censorship scheme protects critical national interests. Furthermore, the classification criteria applied in this case — for “secret” information — specify the nature of information subject to censorship with sufficient particularity to satisfy the applicable constitutional tests for first amendment restraints on former CIA employees. The district court also properly gave deference to the CIA’s detailed and reasoned explanation of its classification decisions. We agree that the CIA reasonably classified as “secret” the censored portions of McGehee’s article.
I. Background
In 1952, when he joined the CIA, McGehee signed a secrecy agreement, promising not to divulge “classified information” obtained by virtue of his employment “unless specifically authorized in writing” by the CIA to do so.1 Pursuant to this agreement, on March 20, 1981, he submitted an article to the CIA for prepublication review. The article asserted that the CIA had mounted a campaign of deceit to convince the world that the “revolt of the poor natives against a ruthless U.S.-backed oligarchy” in El Salvador was really “a Soviet/Cuban/Bulgarian/Vietnamese/PLO/Ethiopian/Nicaraguan/International Terrorism challenge to the United States.” 2 To lend plausibility to his assertion, McGehee’s article proceeded “to review a few examples of ... CIA disinformation programs” in Iran, Vietnam, Chile, and Indonesia.3 Four days later, on March 24, 1981, the CIA notified McGehee that portions of his article contained “secret” information, and accordingly withheld permission to publish those portions.
The CIA employs three classification levels. . The most sensitive information — information that, if disclosed, “reasonably could be expected to cause exceptionally grave damage to the national security” — the CIA classifies “Top Secret.” Exec. Order No. 12,065 § 1-102, 3 C.F.R. 190, 191 (1979) (current version at Exec. Order No. 12,356 § 1.1(a)(1), 3 C.F.R. 166, 167 (1983)).4 The least restricted, but still sensitive, information — information that, if disclosed, “reasonably could be expected to cause identifiable damage to the national security” — the [1140]*1140CIA classifies “Confidential.” Id. § 1-104, 3 C.F.R. at 191.
The CIA classified the censored portions of McGehee’s article “Secret.” “Secret” is the middle level classification between “Top Secret” and “Confidential”:
“Secret” shall be applied only to information, the unauthorized disclosure of which reasonably could be expected to cause serious damage to the national security.
Id. § 1-103, 3 C.F.R. at 191. The CIA found that McGehee’s identification of countries where the CIA had established bases would likely damage U.S. relations with those countries, and that his discussion of the details of a CIA operation in Indonesia would likely disclose secret intelligence methods and help identify secret intelligence sources. The CIA therefore concluded that portions of McGehee’s article threatened to cause “serious damage to the national security.”
The Nation magazine published McGehee’s article on April 11, 1981 with the censored portions deleted.5 McGehee then sought judicial review in the district court, challenging (1) the constitutionality of the CIA’s classification and censorship scheme, and (2) the propriety, under that scheme, of classifying portions of his article “secret.”
The district court summarily rejected McGehee’s constitutional challenge. In the district court’s view “the critical question is whether that information has been properly classified and thus [is] subject to censorship.”6 After a “de novo review of the [affidavits] submitted for in camera inspection” by the parties, and “exercising] judicial deference to the administrative expertise” of the CIA, the district court found that the CIA had “properly classified the documents and [was] warranted in [its] censorship.” 7
In this court, McGehee renews his claim that the CIA classification and censorship scheme violates the first amendment, and that, even if the scheme is constitutional, the CIA improperly classified portions of his article.
We affirm, holding that (1) when balanced against the first amendment interests in public disclosure of former agents’ writings, the CIA scheme of classifying and censoring “secret” information is constitutional because (a) the government has a substantial interest in assuring secrecy in the conduct of foreign intelligence operations, and (b) the criteria for what constitutes “secret” information are neither over-broad, considering the governmental interest the scheme protects, nor excessively vague, considering the particularity with which the criteria offer guidance to the censor; (2) in reviewing whether specified information reasonably could be expected to cause actual serious harm if divulged, courts should accord deference to the CIA’s reasoned explanation of its classification decision; and (3) in this case, the CIA properly classified the censored portions of McGehee’s article.
II. The Constitutional Standard for Reviewing the Censorship Scheme
In Snepp v. United States, 444 U.S. 507, 509 n. 3, 100 S.Ct. 763, 765 n. 3, 62 L.Ed.2d 704 (1980) (per curiam), the Supreme Court held that the CIA could, consistent with the first amendment, recover damages for breach of a secrecy agreement under which a former agent promised to submit CIA-related writings to the CIA for prepublication clearance. The Court found the secrecy agreement to be “a reasonable means for protecting” the “secrecy of information important to our national security and the appearance of confidentiality so essential to the effective operation of our foreign intelligence service. Id. In Snepp, the former agent published CIA-related information without submitting his manuscript for prepublication review. The government’s action in Snepp, therefore, [1141]*1141did “not depend upon whether [Snepp’s] book actually contained classified information .... The Government simply claim[ed] that ... Snepp should have given the CIA an opportunity to determine whether the material he proposed to publish would compromise classified information or sources.” Id. at 511, 100 S.Ct. at 766.
In this case, by contrast, McGehee adhered to his secrecy agreement. He submitted his manuscript for prepublication review, and deleted portions of his article in accordance with the CIA’s orders. At issue here is the constitutionality of the CIA’s substantive criteria and scheme for deciding how to classify, and thereby censor, writings of former agents.
We note, to begin with, that McGehee’s secrecy agreement applies only when he seeks to publish “classified information” that “has come or shall come to [his] attention by virtue of [his] connection with the Central Intelligence Agency.”8 The agreement does not extend to unclassified materials or to information obtained from public sources. The government may not censor such material, “contractually or otherwise.” United States v. Marchetti, 466 F.2d 1309, 1313 (4th Cir.), cert. denied, 409 U.S. 1063, 93 S.Ct. 553, 34 L.Ed.2d 516 (1972). The government has no legitimate interest in censoring unclassified materials. Moreover, when the information at issue derives from public sources, the agent’s special relationship of trust with the government is greatly diminished if not wholly vitiated.9 Accord Snepp v. United States, 444 U.S. at 513 n. 8, 100 S.Ct. at 767 n. 8 (dictum) (“if in fact information is unclassified or in the public domain, neither the CIA nor foreign agencies would be concerned”).
As in McGehee’s case, the CIA requires all of its employees to enter into a secrecy agreement as a condition of employment.10 This fact is critical to our first amendment analysis because numerous Supreme Court decisions establish that:
[T]he State has interests as an employer in regulating the speech of employees that differ significantly from those it possesses in connection with regulation of the speech of the citizenry in general. The problem ... is to arrive at a balance between the interests of the [employee], as a citizen, in commenting upon matters of public concern and the interest of the State, as an employer, in promoting the [1142]*1142efficiency of the public services it performs through its employees.
Pickering v. Board of Education, 391 U.S. 563, 568, 88 S.Ct. 1731, 1734, 20 L.Ed.2d 811 (1968); see Connick v. Myers, - U.S. -, 103 S.Ct. 1684, 1686, 75 L.Ed.2d 708 (1983) (quoting Pickering). We must, then, apply a balancing test in determining whether the CIA’s censorship of ex-agents’ writings violates the first amendment.
The Snepp Court discussed the relevant balancing standard as if it were well settled:
[T]his Court’s case makes clear that ... the CIA [can] act[ ] to protect substantial government interests by imposing reasonable restrictions on employee activities that in other contexts might be protected by the First Amendment.
444 U.S. at 509 n. 3, 100 S.Ct. at 765 n. 3 (1980). A review of relevant cases, however, shows that the precise standard for balancing is not that well settled.
In United States Civil Service Commission v. National Association of Letter Carriers, 413 U.S. 548, 93 S.Ct. 2880, 37 L.Ed.2d 796 (1973), the Supreme Court upheld the Hatch Act’s proscription on partisan politicking by government employees. In doing so, the Court quoted the Pickering balancing test and then provided the terse explanation that:
[T]he balance [Congress] has ... struck is sustainable by the obviously important interests sought to be served by the .. . Hatch Act.
413 U.S. at 564, 93 S.Ct. at 2890. The Court concluded with a lengthy analysis of the statute, construing it narrowly to avoid overbreadth and vagueness problems. Id. at 568-81, 93 S.Ct. at 2892-98.
Cole v. Richardson, 405 U.S. 676, 92 S.Ct. 1332, 31 L.Ed.2d 593 (1972), the culmination of a series of Supreme Court cases reviewing loyalty oaths required of government employees, sheds some additional light on the proper balancing standard. Under Cole, a loyalty oath must not require an employee to forego “protected speech activities” and must not be unduly vague; moreover, inquiry into associational activities must be “narrowly confined” to avoid infringement upon constitutional rights. Id. at 680, 92 S.Ct. at 1335.
More closely related to the case before us is Brown v. Glines, 444 U.S. 348, 100 S.Ct. 594, 62 L.Ed.2d 540 (1980). There the Court upheld against a first amendment challenge an Air Force regulation that prohibited “ ‘any [Air Force] member ... in uniform or ... in a foreign country’ from soliciting signatures on a petition without first obtaining authorization from the appropriate commander,” id. at 349, 100 S.Ct. at 597 (quoting Air Force Reg. 30-1(9) (1971)), and provided that “ ‘[n]o member of the Air Force will distribute or post any printed or written material ... within any Air Force installation without permission of the commander ...,’” id. at 349-50, 100 S.Ct. at 597 (quoting Air Force Reg. 35-15(3)(a)(1) (1970)). The Court upheld this scheme of precirculation clearance because it “protects] a substantial Government interest unrelated to the suppression of free speech,” id. at 354,100 S.Ct. at 599, namely, the “ ‘overriding demands of discipline and duty’ ” in the military services, id. (quoting Parker v. Levy, 417 U.S. 733, 744, 94 S.Ct. 2517, 2556, 41 L.Ed.2d 439 (1974)). Moreover, the Court found that the regulations “restrict speech no more than is reasonably necessary to protect th[is] substantial governmental interest.” Id. at 355, 100 S.Ct. at 600.
We discern from these cases two consistent themes, best articulated in Brown v. Glines. First, restrictions on the speech of government employees must “protect a substantial government interest unrelated to the suppression of free speech.”11 Id. at [1143]*1143354, 100 S.Ct. at 599; see Snepp, 444 U.S. at 509 n. 3, 100 S.Ct. at 765 n. 3 (“substantial government interest”); National Association of Letter Carriers, 413 U.S. at 564, 93 S.Ct. at 2890 (“important interests”). Second, the restriction must be narrowly drawn to “restrict speech no more than is necessary to protect the substantial government interest.” Brown v. Glines, 444 U.S. at 355, 100 S.Ct. at 600; see National Association of Letter Carriers, 413 U.S. at 580, 93 S.Ct. at 2897 (Hatch Act not impermissibly overbroad); Cole v. Richardson, 405 U.S. at 680, 92 S.Ct. at 1335 (loyalty oath must be narrowly confined).
With these two principles in mind, we now turn to examine the CIA scheme for classification and censorship of “secret” information.
III. The Constitutional Standard Applied to the CIA Scheme for “Secret” Information
The CIA classified portions of McGehee’s articles as “secret,” believing that disclosure “reasonably could be expected to cause serious damage to the national security.” Exec. Order No. 12,065 § 1-103, 3 C.F.R. 190, 191 (1979). We hold that the CIA censorship of “secret” information contained in former agents’ writings and obtained by former agents during the course of CIA employment does not violate the first amendment.
The censorship of “secret” information “protect[s] a substantial governmental interest unrelated to the suppression of free expression.” Brown v. Glines, 444 U.S. at 354, 100 S.Ct. at 599. Indeed, the government “has a compelling interest in protecting ... the secrecy of information important to our national security .... ” Snepp v. United States, 444 U.S. at 509 n. 3, 100 S.Ct. at 765 n. 3 (emphasis added). Information properly classified as “secret” does possess such importance by virtue of its potential for causing “serious damage to the national security.”
Further, the classification criteria for “secret” information reasonably confine the resulting censorship to cases in which a substantial governmental interest is served. The criteria do not sweep too broadly because they impede disclosure only when it poses a reasonable probability of “serious” harm. In addition, as the following discussion explains, the classification criteria are not excessively vague.
The term “national security” is defined for classification purposes as “the national defense and foreign relations of the United States.” So defined, the term is inherently vague. See Halperin v. Kissinger, 606 F.2d 1192, 1200 (D.C.Cir.1979), aff’d by an equally divided Court, 452 U.S. 713, 101 S.Ct. 3132, 69 L.Ed.2d 367 (1981); cf. United States v. United States District Court, 407 U.S. 297, 314, 320, 92 S.Ct. 2125, 2135, 2138, 32 L.Ed.2d 752 (1972) (noting the “inherent vagueness of the domestic security concept”); Zweibon v. Mitchell, 516 F.2d 594, 653-54 (D.C.Cir.1975) (en banc) (the concept “affecting foreign relations” vague and subject to abuse). In this case, however, [1144]*1144the governing Executive Order adds specificity to the definition by enumerating the types of information that may be considered for classification:
Information may not be considered for classification unless it concerns:
(a) military plans, weapons, or operations;
(b) foreign government information;
(c) intelligence activities, sources or methods;
(d) foreign relations or foreign activities of the United States;
(e) scientific, technological, or economic matters relating to the national security;
(f) United States Government programs for safeguarding nuclear materials or facilities; or
(g) other categories of information which are related to national security and which require protection against unauthorized disclosure as determined by the President, by a person designated by the President ..., or by an agency head.12
Exec. Order No. 12,065 § 1-301, 3 C.F.R. 190, 193 (1979).
To be sure, this attempt to limit the classification of documents has its shortcomings. In particular, item (e) refers back to the concept of national security itself, thereby undermining the attempt to eliminate vagueness, and item (g) grants roving classification authority to specified individuals. The CIA, however, did not invoke items (e) or (g) in this case.13 We therefore need not consider today whether classification decisions that place great reliance upon items (e) or (g) would be defective due to the asserted vagueness of the Executive Order.14
Item (d) also refers broadly to “foreign relations or foreign activities of the United States.” Standing alone, such a classification standard might be excessively vague. See Zweibon v. Mitchell, 516 F.2d at 653-55. The CIA, however, has articulated narrower standards to guide classification decisions under item (d). See Agency Information Security Program Handbook: Classifying, Declassifying, Marking and Safeguarding [1145]*1145National Security Information, HHB No. 70-2 f 9d (Nov. 28, 1978). The guidelines pertinent to this case15 are sufficiently precise to withstand a challenge for unconstitutional vagueness.
Moreover, one of the principal dangers of imprecise standards is that they may deter legitimate expression. When an administrative scheme is designed to avoid the deterrence of legitimate speech, however, that danger is alleviated in part. Thus in Civil Service Commission v. National Association of Letter Carriers, the Supreme Court rejected a vagueness challenge to the Hatch Act, in part because:
the Commission ha[d] established a procedure by which an employee in doubt about the validity of a proposed course of conduct may seek and obtain advice from the Commission and thereby remove any doubt there may be as to the meaning of the law ....
413 U.S. at 580, 93 S.Ct. at 2897-98. Similarly, in this case the Executive Order, by establishing a preclearance procedure for determining whether intended publications contain classified material, engenders less of a chilling effect on free speech. This is not to say, of course, that imprecise standards would not still present an intolerable burden by allowing the censor unwarranted discretion in vetoing material, but, as we have said, we do not find the standards for classifying material as “secret” unconstitutionally vague. And we note that the agent may seek judicial review of the CIA’s classification decision.16
Finally, McGehee argues that the classification standard for “confidential” information — “information, the unauthorized disclosure of which reasonably could be expected to cause identifiable damage to the national security” — should be invalidated because it is “too broad and vague to satisfy the First Amendment.”17 Because, as explained below, McGehee lacks standing to assert overbreadth and vagueness challenges to the “confidential” standard, we need not consider them in this case.
The censored portions of McGehee’s writings were classified “secret,” not “confidential.” 18 Thus he attempts to challenge the [1146]*1146Executive Order on the ground that it may conceivably be unconstitutionally applied to others, in situations other than his own. Generally, litigants may not properly raise such claims. See, e.g., Broadrick v. Oklahoma, 413 U.S. 601, 610, 93 S.Ct. 2908, 2915, 37 L.Ed.2d 830 (1973); United States v. Raines, 362 U.S. 17, 22, 80 S.Ct. 519, 523, 4 L.Ed.2d 524 (1960). When the litigant asserts a first amendment challenge, however, courts sometimes make an exception and hear “attacks on overly broad [rules] with no requirement that the person making the attack demonstrate that his own conduct could not be regulated by a [rule] drawn with the requisite narrow specificity.” Dombrowski v. Pfister, 380 U.S. 479, 486, 85 S.Ct. 1116, 1121, 14 L.Ed.2d 22 (1965).
This exception, however, does not extend automatically to all first amendment challenges. Courts devised the exception “because of a judicial prediction or assumption that the [rule]’s very existence may cause others not before the court to refrain from constitutionally protected speech.” Broadrick v. Oklahoma, 413 U.S. at 612, 93 S.Ct. at 2916. Accordingly, overbreadth analysis should not be deployed when a limiting construction could save the rule from its constitutional defects, see, e.g., Dombrowski v. Pfister, 380 U.S. at 491, 85 S.Ct. at 1123; Cox v. New Hampshire, 312 U.S. 569, 61 S.Ct. 762, 85 L.Ed. 1049 (1941), or when its deterrent effect is neither real nor substantial, see, e.g., Young v. American Mini Theatres, Inc., 427 U.S. 50, 96 S.Ct. 2440, 49 L.Ed.2d 310 (1976); Broadrick v. Oklahoma, 413 U.S. at 615, 93 S.Ct. at 2917.
In this case, the Executive Order does, under some constructions, touch constitutionally protected speech. As McGehee observes, “identifiable damage” might include “a minimal amount of harm which may not be sufficient in all cases to justify suppression of speech.”19 Yet future construction of the order20 could reasonably delimit the scope of the order. The CIA or a reviewing court might formulate standards for “confidential” information that require damage to “important ... national security” interests. Snepp v. United States, 444 U.S. at 509 n. 3,100 S.Ct. at 765 n. 3. Alternatively, the classification scheme might be narrowed to exclude much of what falls under the rubric of “confidential” materials. To paraphrase the Supreme Court, “[e]ven if the [‘confidential’ standard] were to be considered in some respects unconstitutionally overbroad, we would not invalidate the entire [classification scheme]. The remainder of the [scheme] ... covers a whole range of ... constitutionally proscribable ... conduct.” Civil Service Commission v. National Association of Letter Carriers, 413 U.S. at 580-81, 93 S.Ct. at 2897-98.
We also conclude that the classification scheme, taken as a whole, does not result in real or substantial deterrence of protected speech. The “secret” and “top secret” classifications place constitutional burdens on the speech of former CIA agents. We are not prepared to hold today that the burdens [1147]*1147placed on such speech by the “confidential ’ classification are sufficiently heavy or widespread to render the entire classification scheme substantially overbroad and therefore invalid on its face. Cf. id. at 581, 93 S.Ct. at 2898 (same conclusion regarding an overbreadth challenge to the Hatch Act on the grounds that a prohibition against political endorsements, not applied to the plaintiff, rendered the entire statute void). In addition, as we noted above, the CIA classification and censorship scheme reduces deterrence, because prepublication review alleviates a former agent’s fear that his disclosure of non-sensitive information might result in liability. Thus the CIA’s scheme reduces any disfavored chilling effect.
McGehee also lacks standing to challenge the “confidential” standard for vagueness. A litigant cannot properly challenge a rule for vagueness when it clearly applies to him. See Parker v. Levy, 417 U.S. 733, 756, 94 S.Ct. 2547, 2562, 41 L.Ed.2d 439 (1974) (“One to whose conduct a statute clearly applies may not successfully challenge it for vagueness.”).
Accordingly, we uphold against first amendment challenge the CIA classification and censorship scheme for “secret” information. We decline to rule whether the “confidential” standard passes constitutional muster because McGehee lacks standing to mount such a challenge. We now turn to the question whether the CIA properly classified portions of McGehee’s article as “secret.”
IV. The Standard for Judicial Keview of Individual CIA Censorship Decisions
The district court found that the CIA “properly classified the [relevant] documents and [was] warranted in their censorship” of portions of McGehee’s article.21 We agree, although we take this opportunity to clarify the standard of judicial review appropriate to a case such as this one.
This case arises in a posture significantly different from a request for release of CIA documents under the Freedom of Information Act (FOIA). In a FOIA case, an individual seeks to compel release of documents in the government’s possession. Here, by contrast, McGehee wishes publicly to disclose information that he already possesses, and the government has ruled that his secrecy agreement forbids disclosure.
This difference between seeking to obtain information and seeking to disclose information already obtained raises McGehee’s constitutional interests in this case above the constitutional interests held by a FOIA claimant. As a general rule, citizens have no first amendment right of access to traditionally nonpublic government information. See, e.g., Houchins v. KQED, Inc., 438 U.S. 1, 8-9, 98 S.Ct. 2588, 2593, 57 L.Ed.2d 553 (1978) (plurality opinion); id. at 16, 98 S.Ct. at 2597 (Stewart, J., concurring); Saxbe v. Washington Post Co., 417 U.S. 843, 849, 94 S.Ct. 2811, 2814, 41 L.Ed.2d 514 (1974); Pell v. Procunier, 417 U.S. 817, 831-32, 94 S.Ct. 2800, 2808-09, 41 L.Ed.2d 495 (1974). A litigant seeking release of government information under FOIA, therefore, relies upon a statutory entitlement — as narrowed by statutory exceptions — and not upon his constitutional right to free expression.
In this case, however, McGehee wishes to publish information he possesses, and the CIA wishes to silence him. Although neither the CIA’s administrative determination nor any court order in this case constitutes a prior restraint in the traditional sense upon McGehee or any other party,22 [1148]*1148the entire scheme of prepublication review is designed for the purpose of preventing publication of classified information. McGehee therefore has a strong first amendment interest in ensuring that CIA censorship of his article results from a proper classification of the censored portions. Cf. Alfred A. Knopf, Inc. v. Colby, 509 F.2d 1362, 1367 (4th Cir.) (“the deletion items should be suppressed only if they are found to be both classified and classifiable under the Executive Order”), cert. denied, 421 U.S. 992, 95 S.Ct. 1999, 44 L.Ed.2d 482 (1975).
We must accordingly establish a standard for judicial review of the CIA classification decision that affords proper respect to the individual rights at stake while recognizing the CIA's technical expertise and practical familiarity with the ramifications of sensitive information. We conclude that reviewing courts should conduct a de novo review of the classification decision, while giving deference to reasoned and detailed CIA explanations of that classification decision.
We begin with an examination of the standard employed in the review of FOIA requests for classified information, because the scope of judicial review in this case should be at least that broad. The FOIA calls for de novo judicial review of an agency decision, and places the burden on the agency to justify its claim of exemption. 5 U.S.C. § 552(a)(4)(B). See, e.g., Salisbury v. United States, 690 F.2d 966, 970 (D.C.Cir.1982); Military Audit Project v. Casey, 656 F.2d 724, 738 (D.C.Cir.1981). At the same time, courts are to “accord substantial weight to an agency’s affidavit concerning the details of the classified status of the disputed record” because “the Executive departments responsible for national defense and foreign policy matters have unique insights into what adverse affects [sic] might occur as a result of a particular classified record.” S.Rep. No. 1200, 93d Cong., 2d Sess. 12, U.S.Code & Admin.News 1974, p. 6267 (1974) (Conference Report on the FOIA Amendments).
This circuit has on many occasions reviewed whether the denial of a FOIA request properly fell within the FOIA exemption for classified documents. 5 U.S.C. § 552(b)(1). In these cases, we have established that CIA explanations for its classification decisions should be neither “conclusory, merely reciting statutory standards, [nor] too vague [n]or sweeping.” Hayden v. National Security Agency, 608 F.2d 1381, 1387 (D.C.Cir.1979), cert. denied, 446 U.S. 937, 100 S.Ct. 2156, 64 L.Ed.2d 790 (1980). At the same time, “[o]nce satisfied that the proper procedures have been followed and that the information logically falls into the exemption claimed, the courts ‘need not go further to test the expertise of the agency, or to question its veracity when nothing appears to raise the issue of good faith.’ ” Gardels v. Central Intelligence Agency, 689 F.2d 1100, 1104-05 (D.C.Cir.1982) (quoting Weissman v. Central Intelligence Agency, 565 F.2d 692, 697 (D.C.Cir.1977)); see also Halperin v. Central Intelligence Agency, 629 F.2d 144, 148 (D.C.Cir.1980). Similarly, in Alfred A. Knopf, Inc. v. Colby, supra, the Fourth Circuit, after invoking the FOIA standard, announced a “presumption of regularity” for CIA classification decisions. 509 F.2d at 1368.
Because the present case implicates first amendment rights, however, we feel compelled to go beyond the FOIA standard of review for cases reviewing CIA censorship pursuant to secrecy agreements. While we believe courts in securing such determinations should defer to CIA judgment as to the harmful results of publication, they must nevertheless satisfy themselves from the record, in camera or otherwise, that the CIA in fact had good reason to classify, and therefore censor, the materials at issue. Accordingly, the courts should require that CIA explanations justify censorship with reasonable specificity, demonstrating a logical connection between the deleted information and the reasons for classification. These should not rely on a “presumption of regularity” if such rational [1149]*1149explanations are missing. We anticipate that in camera review of affidavits, followed if necessary by further judicial inquiry, will be the norm. Cf. Hayden v. National Security Agency, 608 F.2d at 1387 (“in camera review is neither necessary nor appropriate” in FOIA cases once the agency demonstrates “(1) that it followed proper classification procedures, and (2) that by its description the document logically falls within the claimed exemption”). Moreover, unlike FOIA cases,23 in cases such as this both parties know the nature of the information in question. Courts should therefore strive to benefit from “criticism and illumination by [the] party with the actual interest in forcing disclosure.” Vaughn v. Rosen, 484 F.2d 820, 825 (D.C.Cir.1973) (describing the defects in the FOIA procedures), cert. denied, 415 U.S. 977, 94 S.Ct. 1564, 39 L.Ed.2d 873 (1974). This was, in fact, the procedure employed by the district court here.24
We are, of course, well aware that judicial review of CIA classification decisions, by reasonable necessity, cannot second-guess CIA judgments on matters in which the judiciary lacks the requisite expertise. Due to the “mosaic-like nature of intelligence gathering,” Salisbury v. United States, 690 F.2d at 971, for example, “[w]hat may seem trivial to the uninformed[] may appear of great moment to one who has a broad view of the scene and may put the questioned item of information in context,” United States v. Marchetti, 466 F.2d 1309, 1318 (4th Cir.), cert. denied, 409 U.S. 1063, 93 S.Ct. 553, 34 L.Ed.2d 516 (1972). But while the CIA’s tasks include the protection of the national security and the maintenance of the secrecy of sensitive information, the judiciary’s tasks include the protection of individual rights. Considering that “speech concerning public affairs is more than self-expression; it is the essence of self-government,”25 and that the line between information threatening to foreign policy and matters of legitimate public concern is often very fine26 see New York Times Co. v. United States, 403 U.S. 713, 91 S.Ct. 2140, 29 L.Ed.2d 822 (1971), courts must assure themselves that the reasons for classification are rational and plausible ones.
V. The Application of the Standard to This Case
We have made such a judgment in this case, and, after examining the detailed in camera affidavits of the parties, we conclude that the CIA properly classified the information at issue. The CIA affidavits give us reason to believe that disclosure of the censored portions of McGehee’s article could reasonably be expected to cause serious damage to the national security. The affidavit offers reasonably convincing and detailed evidence of a serious risk that intelligence sources and methods would be compromised by disclosures proposed by McGehee. We also believe, on the basis of plausible scenarios put forward in the CIA affidavit, that the United States could suffer significant strategic and diplomatic set[1150]*1150backs as a result of the disclosure of the deleted information. These risks identified in the CIA affidavits do not, of course, rise to the level of certainty, but we believe they are real and serious enough to justify the classification decision in this case.
Accordingly, the judgment of the district court is
Affirmed.
Separate Statement of
The Supreme Court has ruled that secrecy contracts, such as the one involved here, are a legal condition of CIA employment; today this court finds that the “secret” classification censorship scheme employed by the CIA has been constitutionally applied in this case.
I write separately to stress, however, that neither the agency’s nor our analysis takes account of any separate public right to know critical albeit classified facts about the activities of our intelligence agencies.1 See Afshar v. CIA, 702 F.2d 1125 (D.C.Cir.1983); cf. N.Y. Times v. United States, 403 U.S. 713, 728-30, 91 S.Ct. 2140, 2148-49, 29 L.Ed.2d 822 (1970) (Stewart, J., concurring). It would of course be extremely difficult for judges to “balance” the public’s right to know against an acknowledged national security risk, and I do not believe we are currently authorized to do so. However, it seems important in view of recent revelations about past indiscretions in the name of national security, for some governmental institution, if not the classification system itself, to conduct such a balance. As Emerson explained, “history ... give[s] value to the present hour and its duty.” By not weighing the value to the public of knowing about particularly relevant episodes in the intelligence agencies’ history, we may undermine the public’s ability to assess the government’s performance of its duty. Economic and criminal sanctions against agents who violate the preclearance and agency classification scheme are justifiable. But with no mechanism in the system for balancing the public’s right to know with possible risks to security, those sanctions can also result in the permanent loss of information critical to public debate. Our decision today, reflecting current restraints on our authority, cannot and does not fill the public’s need for such a balance.