Full opinion text
Jośe A. CAJBRANES, Circuit Judge: The plaintiff, an editor, publisher, and part-owner of a newspaper distributed exclusively through electronic means, brings this First Amendment challenge to § 223(d) of the Communications Decency Act of 1996 (“CDA”) criminalizing the use of interactive computer services to display “patently offensive” sexually explicit material such that it is available to persons under the age of eighteen. The plaintiff seeks a preliminary injunction barring application of the section. The three-judge panel, appointed pursuant to 28 U.S.C. § 2284, held that: (1) the plaintiff has not sustained his burden of demonstrating a likelihood of success on his claim that § 223(d) is unconstitutionally vague, but (2) the plaintiff has demonstrated a likelihood of success on his claim that § 223(d) is unconstitutionally overbroad in that it bans protected indecent communication between adults. On this second point, the court concluded that most content providers’ ability to comply with the requirements of the affirmative defenses set out in the statute depends on the actions of third parties, such as software manufacturers, whose cooperation is not required under the statute or otherwise mandated. The technological impossibility of independent compliance with the affirmative defenses renders § 223(d) unconstitutional as an overbroad prohibition on constitutionally protected indecent speech between adults. MEMORANDUM AND ORDER We address here the constitutionality of a provision of the Communications Decency Act of 1996 (“CDA”) with an undeniably worthy goal: to limit the exposure of children to sexually explicit, though not legally obscene, materials available “on line” — that is, capable of being displayed and “accessed” by increasingly common interactive computer services. 47 U.S.C. § 223(d), as added by the CDA on February 8, 1996, criminalizes the use of an interactive computer service to display, in a manner available to persons under eighteen, sexually explicit material that is “patently offensive” by contemporary community standards. Plaintiff Joe Shea, the editor, publisher, and parCowner of a newspaper distributed solely by electronic means, filed this action on February 8, 1996, claiming that § 223(d) is (1) void for vagueness, in that it fails to give ordinary citizens sufficient notice of what conduct will subject them to prosecution or criminal liability; and (2) substantially overbroad, in that it targets a broader category of speech than necessary to achieve the government’s goal and constitutes a ban on certain constitutionally protected speech between adults. As editor of an on-line newspaper, the plaintiff is one of a growing number of citizens who employ an array of widely accessible and constantly evolving media technologies to gather and disseminate information and ideas. In passing the CDA, Congress explicitly recognized that these technologies foster “true diversity of political discourse, unique opportunities for cultural development, and myriad avenues for intellectual activity.” Pub.L. No. 104^104, § 509(a)(3), 110 Stat. 56, 138 (1996) (to be codified at 47 U.S.C. § 230(a)(3)). The range of tools and forums available for users of interactive computer services is astounding: with access to the web of computer networks known as the Internet, a scholar can contact a distant computer and make use of its capabilities; a researcher can peruse the card catalogs of libraries across the globe; users around the world can debate politics, sports, music, and literature. However trivial some of their uses might seem, emerging media technologies quite simply offer an unprecedented number of individual citizens an opportunity to speak and to be heard — at very little cost — by audiences around the world. In that sense, we are encountering a communications medium unlike any we have ever known. In an attempt to limit the availability of certain materials in interactive computer services, Congress enacted a statute of unprecedented sweep: the new § 223(d) purports to regulate not only how commercial purveyors of obscene or pornographic materials may advertise and sell their products on line, but also how private individuals who choose to exchange certain constitutionally protected communications with one another can do so. The question presented is whether our Constitution tolerates this level of governmental intrusion into how adults speak to one another. We conclude, first, that the plaintiff has not sustained his burden of demonstrating a likelihood of success on his claim that § 223(d) is unconstitutionally vague. The definition of material regulated by this section is a familiar one, repeatedly upheld against vagueness challenges in a line of jurisprudence concerning television and radio broadcasting, cable programming, and commercial telephone services. We do, however,' conclude that the plaintiff has demonstrated a likelihood of success on his overbreadth claim, that § 223(d) would serve as a ban on constitutionally protected indecent communication between adults. The Government concedes that strict scrutiny is appropriately applied to this claim and that § 223(d) would, on its own, act as an unconstitutional total ban on indecent communication, protected and unprotected alike, but argues that two affirmative defenses set out in § 223(e)(5) serve to shield adults engaging in constitutionally protected indecent communication from criminal liability. The evidentiary record in this ease compels the conclusion that, given the current state of technology, most adult content providers wishing to engage in constitutionally protected indecent speech will be unable to avail themselves of these affirmative defenses. Only a limited subset of on-line content providers, commercial providers on the World Wide Web, can avail themselves of the defense set out in § 223(e)(5)(B), leaving both non-commercial providers of Web content and content providers using all other modes of on-line communication unprotected. The evidence further demonstrates that content providers’ ability to comply with the terms of the second defense — the so-called good-faith defense — depends on the actions of third parties, such as software manufacturers, whose cooperation is not required under the CDA or otherwise mandated. There is no feasible means, with our current technology, for someone to provide indecent content on line with any certainty that even his best efforts at shielding the material from minors will be “effective,” as the language of the good-faith defense requires. Because neither of the affirmative defenses set out in § 223(e)(5) can, with our current technology, effectively protect adult content providers wishing to engage in constitutionally protected indecent communication, we reach the inescapable conclusion that § 223(d) will serve to chill protected speech. We therefore find that the plaintiff has demonstrated a likelihood of success on the merits of his claim that § 223(d) is unconstitutionally overbroad. We are mindful of our obligation to construe a federal statute to avoid constitutional problems if it is possible to do so, but we are equally mindful of the limits of the judicial power under our Constitution and we decline the Government’s invitation to perform radical surgery on a statute dealing with a difficult problem in a rapidly changing area of technology; in sum, we respectfully decline the invitation to legislate from the bench. In setting aside the challenged provisions, we do not question the legitimacy of the government’s interest in safeguarding children from exposure to certain materials available on line nor suggest that other legislation on another day, carefully tailored to technological realities, may not pass constitutional muster. We also do not consider, nor attempt to delineate, the range of circumstances, if any, in which Congress could now or in the future constitutionally impose content-based restrictions upon communications in the developing medium we explore here. I. Background Plaintiff Joe Shea is the editor-in-chief, part-owner, and publisher of the American Reporter, a daily newspaper distributed solely by electronic means. On February 8, 1996, following the signing of the Telecommunications Act of 1996, Pub.L. No. 104-104, 110 Stat. 56, the American Reporter published an editorial, (Complaint, Ex. 1) criticizing Title V of the Act, known as the Communications Decency Act of 1996 (“CDA”). The editorial contained language arguably falling within the scope of a provision of the CDA criminalizing the transmission or display of certain content in a manner available to minors: Whoever— (1) in interstate or foreign communications knowingly— (A) uses an interactive computer service to send to a specific person or persons under 18 years of age, or (B) uses any interactive computer service to display in a manner available to a person under 18 years of age, any comment, request, suggestion, proposal, image, or other communication that, in context, depicts or describes, in terms patently offensive as measured by contemporary community standards, sexual or excretory activities or organs, regardless of whether the user of such service placed the call or initiated the communication; ... shall be fined under title 18, United States Code, or 'imprisoned not more than two years, or both. Pub.L. No. 104-104, § 502(2)(d), 110 Stat. at 133 (to be codified at 47 U.S.C. § 223(d)). Also on February 8, the plaintiff filed this action on behalf of the American Reporter, seeking a declaration that 47 U.S.C. § 223(d) is unconstitutionally overbroad and vague. On February 17, the plaintiff moved for preliminary injunctive relief to prevent the Department of Justice from enforcing the provision in question. Pursuant to § 561(c) of the Telecommunications Act and 28 U.S.C. § 2284, the Chief Judge of the United States Court of Appeals for the Second Circuit designated this three-judge district court to consider the plaintiffs facial challenge to the constitutionality of § 223(d). The Court heard opening arguments on April 3, 1996. Because the plaintiffs facial challenge to § 223(d) raised the question of whether Internet users can, under current technology, meet the requirements for certain defenses provided for in the Act, the Court concluded that an eviden-tiary hearing would be appropriate. In light of the pendency of consolidated proceedings for preliminary injunctive relief before a three-judge court in the Eastern District of Pennsylvania raising, among other claims, a facial challenge to § 223(d),. see Complaint, American Civil Liberties Union v. Reno, No. 96-963 (E.D.Pa. filed Feb. 8, 1996); Complaint, American Library Ass’n v. U.S. Dep’t of Justice, No. 96-1458 (E.D.Pa. filed Feb. 27,1996) (collectively “the Philadelphia litigation”), the Court directed the parties to consider methods of easing the fact-finding process by entering into a range of stipulations regarding the current state of technology and. incorporating relevant portions of the record in the Philadelphia litigation. The Court received additional demonstrative and testimonial evidence on April 29, April 30, and May 6, and heard closing arguments on June 3, 1996. Following supplemental briefing by the parties, the plaintiffs motion for preliminary injunctive relief was submitted for decision on June 13,199.6. On June 11, 1996, the three-judge court in the Philadelphia litigation concluded, inter olid, that the provision of the CDA here challenged by the plaintiff does not withstand constitutional scrutiny. American Civil Liberties Union v. Reno, 929 F.Supp. 824, 849 (E.D.Pa.1996) (“ACLU/ALA ”). All three judges agreed that the CDA is substantially overbroad, in that it effectively forces many Internet users (specifically, noncommercial, not-for-profit entities and “even many commercial organizations”) to forgo constitutionally protected speech or risk criminal prosecution. Id. at 854-55 (Sloviter, C.J.); id. at 858 (Buekwalter, J.); id. at 870 (Dalzell, J.). Additionally, two of the judges concluded that § 223(d)’s definition of covered speech is unconstitutionally vague. Id. at 856 (Sloviter, C.J.); id. at 858 (Buckwalter, J.). The decision in the Philadelphia litigation does not preclude this Court from deciding the issues presented, to which we now turn. II. Findings of Fact We enter the following findings of fact, many of which are undisputed, the subject of stipulations by the parties, or submitted by the defendant and adopted by us, pursuant to Rule 52(a) of the Federal Rules of Civil Procedure. Although we here consider a so-called facial challenge to a statute, we deemed it appropriate and necessary in the unusual circumstances presented here, and a reasonable exercise of our discretion, to establish a basic record of the facts regarding the new and evolving communications media that is the subject of this legislation. Section 223(d) targets the use of an “interactive computer service” to send or display patently offensive materials. Although § 223 itself contains no definition of that term, the definition applicable to the new 47 U.S.C. § 230 — also added by the CDA — makes clear that the term encompasses means of making “content” available to multiple users both on the vast web of linked networks popularly known as “the Internet” and on other information systems (such as electronic bulletin boards maintained by educational institutions or nonprofit organizations) not physically linked to the Internet. See Pub.L. No. 104-104, § 509(e)(2), 110 Stat. at 139 (to be codified at 47 U.S.C. § 230(e)(2)). We draw upon the stipulations of the parties and the testimony adduced at the three-day eviden-tiary hearing to describe: (1) the nature of the medium targeted by § 223(d), focusing in part on the degree of control that those who transmit content have over who will receive it; (2) the availability of certain categories of potentially objectionable material on line;. (3) the development of software and labeling standards enabling parents to limit their children’s exposure to objectionable on-line content; and (4) the potential for tagging and verification procedures that content providers can use in an effort to shield minors from sexually explicit content that they provide. As we do so, we unavoidably — and with apologies to all others with a similar aversion to “cyberspeak” — adopt some of the terminology that has developed in conjunction with this technology. We endeavor, to the extent possible, to avoid the jargon of this field, and to define our terms wherever possible, for the sake of the clarity of this record and this opinion, as well as for the benefit of any reader required to review our work. A The Development of the Internet Although “the Internet” now formally describes a collection of more than 50,000 networks linking some nine million host computers in ninety countries, it has existed for nearly three decades on a much smaller scale. What we now refer to as the Internet grew out of an experimental project of the Department of Defense’s Advanced Research Projects Administration (“ARPA”) designed to provide researchers with direct access to supercomputers at a few key laboratories and to facilitate the reliable transmission of vital communications. (Declaration of William J. Hoffman (“Hoffman Deck”), Ex. 4, at 11-12) ARPA supplied funds to link computers operated by the military, defense contractors, and universities conducting defense-related research through dedicated phone lines, creating a “network” known as AR-PANet. (Parties’ Stipulations in Preparation for Preliminary Injunction Hearing (“Joint Stip.”) ¶¶ 6-7; Hoffman Deel., Ex. 3, at 3; id. Ex. 4, at 11) Programs on the linked computers implemented a technical scheme known as “packet-switching,” through which a message from one computer to another would be subdivided into smaller, separately addressed pieces of data, known as “packets,” sent independently to the message’s destination and reassembled upon arrival. (Joint Stip. ¶ 9) Each computer on the network was in turn linked to several other computers, creating any number of routes that a communication from one computer could follow to reach its destination. If part of the network were damaged, a portion of the message could be re-routed automatically over any other path to its ultimate destination, a characteristic of the network intended initially to preserve its operability in the event of enemy attack. (Id. ¶¶ 7-8; Hoffman Deel., Ex. 3, at 3; id. Ex. 4, at 12) Having successfully implemented a system for the reliable transfer of information over a computer network, ARPA began to support the development of communications protocols for transferring data between different types of computer networks. Universities, research facilities, and commercial entities began to develop and link together their own networks implementing these protocols; these networks included a high-speed “backbone” network known as NSFNet, sponsored by the National Science Foundation, smaller regional networks, and, eventually, large commercial networks run by organizations such as Sprint, IBM, and Performance Systems International (commonly known as “PSI”). (Hoffman Deel., Ex. 3, at 3; id. Ex. 4, at 13-14) As faster networks developed, most network traffic shifted away from AR-PANet, which formally ceased operations in 1990. (Id. Ex. 3, at 3) What we know as “the Internet” today is the series of linked, overlapping networks that gradually supplanted ARPANet. Because the Internet links together independent networks that merely use the same data transfer protocols, it cannot be said that any single entity or group of entities controls, or can control, the content made publicly available on the Internet or limits, or can limit, the ability of others to access public content. Rather, the resources available to one with Internet access are located on individual computers around the world. (Joint Stip. ¶ 11) It is estimated that as many as forty million individuals have access to the information and tools of the Internet, and that figure is expected to grow to 200 million by the year 1999. (Id. ¶ 3) Access to the Internet can take any one of several forms. First, many educational institutions, businesses, libraries, and individual communities maintain a computer network linked directly to the Internet and issue account numbers and passwords enabling users to gain access to the network directly or by modem. (Id. ¶¶ 12-14) Second, “Internet service providers,” generally commercial entities charging a monthly fee, offer modem access to computers or networks linked directly to the Internet. (Id. ¶ 16) Third; national commercial “on-line services” — such as America Online, CompuServe, Prodigy, and Microsoft Network — allow subscribers to gain access to the Internet while providing extensive content within their own proprietary networks. (Id. ¶ 17) Finally, organizations and businesses can offer access to electronic bulletin-board systems — which, like national on-line services, provide certain proprietary content; some bulletin-board systems in turn offer users links to the Internet. (Id. ¶ 18) B. Categories of Internet Use For our purposes, there are two loose and overlapping categories of Internet use. First, an individual who has secured access to the Internet can correspond or exchange views with one or many other Internet users. Second, a user can locate and retrieve information available on other computers. We explore these categories in greater detail below. As will become clear, distinctions in how Internet content is transmitted affect the degree of control that providers of content have over who will be able to gain access to their communications; we will return to the legal significance of these distinctions at a later juncture. For any communication to take place over the Internet, two pieces of software, adhering to the same communications protocol, are required. A user must have access to certain kinds of “client” software, which enables his computer to communicate with and make requests of remote computers where information is stored; these remote computers must be running “server” software, which provides information in response to requests by client software. (Declaration of Dr. Dan R. Olsen, Jr. (“Olsen Decl”), ¶¶ 13-14) 1. Communicating with Other Internet Users Perhaps the most widely used Internet service is electronic mail, or “e-mail.” Using any one of dozens of available “mailers”— client software capable of reading and writing e-mail — a user is able to address and transmit a message to one or more specific individuals. (Joint Stip. ¶ 21) A user can also “subscribe” to an electronic mailing list on a topic of interest; the user receives a copy of messages posted by other subscribers and, in turn, can post messages for forwarding to the full mailing list. Once a mailing list is established, it is typically maintained using a “mail exploder” — a program such as “listserv” running on the server on which the list resides — that automatically (ie., without human intervention) responds to a user’s request to be added to or removed from the list of subscribers and retransmits messages posted by a subscriber to others on the mailing list. (Id. ¶22) Some mailing lists are “closed”: a user’s request to join the list requires the approval of an individual who maintains the list. (Id.) Mailing lists (both open and- closed) may also be “moderated”: all messages posted to the list are forwarded to a moderator, who approves certain messages and retransmits them to subscribers. (Id.) An individual sending a message that will be retransmitted by a mail exploder program has no way of knowing the e-mail addresses of other subscribers. (Olsen Decl. ¶ 19; Testimony of Gordon C. Gal-ligher, Jr., Tr. at 181) Even if the user could obtain an e-mail address for each subscriber to a particular list, those addresses alone would provide no authoritative information about subscribers. There is no directory that identifies persons using a certain e-mail address. In addition, a user can avoid disclosing his true e-mail address by developing an e-mail “alias” or by using an “anonymous remailer” — a server that purges identifying information from a communication before forwarding it to its destination. (Defendant’s Response to Plaintiffs Request for Admissions (“Defendant’s Adm.”) No. 22; Galligher Test., Tr. at 173) Internet users may also transmit or receive “articles” posted daily to thousands of discussion groups, arranged by subject matter and known as “newsgroups,” available through an electronic bulletin-board system known as “Usenet.” When a user with access to a Usenet server — that is, a computer participating in the Usenet system — posts an article to a particular newsgroup, the server automatically forwards the article to adjacent Usenet servers, which in turn forward it to other servers, until the article is available on all Usenet sites that furnish access to the newsgroup in question. (Joint Stip. ¶23) Once a message reaches a particular Usenet site, it is temporarily stored there so that individual users — running client software, known as a “newsreader,” capable of sorting articles according to header information identifying the newsgroup to which the article was posted — can review and respond to the message. (Id.; Hoffman Decl., Ex. 4, at 129) Some Usenet newsgroups are moderated; messages to the newsgroup are forwarded to an individual who selects those appropriate for distribution. (Joint Stip. V23) Because Usenet articles are distributed to (and made available on) multiple servers, one who posts an article to a newsgroup has no way of knowing who will choose to retrieve it, whether or not the newsgroup is moderated. (Galligher Test., Tr. at 170,174-75) There is no newsgroup equivalent of a “closed” mailing list: access to a particular newsgroup can only be limited by restricting the number of servers participating in the newsgroup. (Testimony of Clay Shirky, Tr. at 251) The Internet also offers opportunities for multiple users to interact in real time. Using a program called “Talk,” two users can exchange messages while they are both on line; a message typed on one user’s computer will appear almost immediately on the other’s screen. (Joint Stip. ¶ 25) Servers running so-called “chat” software, such as Internet Relay Chat (“IRC”), permit multiple users to converse by selecting one of many discussion “channels” active at any time. Commercial on-line services such as America Online, CompuServe, Prodigy, and the Microsoft Network offer their own chat systems for their members. (Id. ¶ 26) Having joined a channel, the user can see and read messages transmitted by other users, each identified by a name the user selects upon joining the channel. (Id. ¶25) Individual participants in IRC discussions know other participants only by the names they choose upon entering the discussion; users can participate anonymously by using a pseudonym. 2. Locating and Retrieving Information on the Internet Individuals with Internet access can take advantage of a number of tools for locating and retrieving information and resources stored on remote computers. One who wishes to make certain articles, files, or software available to other users will set up a server, adhering to certain communications protocols, capable of retrieving and presenting stored information in response to a request from client software using the same communications protocol. (Olsen Decl. ¶¶ 13, 16; Galligher Test., Tr. at 131) a. File-Transfer Protocol (“FTP”) One type of software implements a set of conventions for copying files from a host computer known as “file-transfer protocol” (“FTP”). With appropriate client software, a user with an account on the host computer can contact the server, view a directory of available files, and copy one or more of those files to his own computer. In addition to making files available to users with accounts, thousands of content providers also make files available for “anonymous” retrieval by users who do not possess an account on the host computer. (Hoffman Deck, Ex. 3, at 1-2, 5; id. Ex. 4, at 187; Joint Stip. ¶29) A content provider who makes files available for retrieval by anonymous FTP has no way of discerning who gains access to the files, b. “Gopher ” Servers A second type of server software capable of making available the resources of a host computer is known as a “gopher” program. (Joint Stip. ¶ 30, Hoffman Decl., Ex. 3, at 5) A gopher server presents information in a set of menus, enabling a user who gains access to the server to select a series of increasingly narrow menu items before locating a desired file that can be displayed on or copied to the user’s computer. (Galligher Test., Tr. at 122; Hoffman Decl., Ex. 3, at 5) A content provider who maintains a gopher server ordinarily has no way of knowing who will gain access to the information made available. c. The World Wide Web The third and perhaps best known method of locating and accessing information on the Internet is by exploring the World Wide Web. Documents available on the Web are not collected in any central location; rather, they are stored on servers around the world running Web server software. (Joint Stip. ¶¶ 31, 38, 40) To gain access to the content available on the Web, a user must have a Web “browser” — client software, such as Netscape Navigator, Mosaic, or Internet Explorer, capable of displaying documents formatted in “hypertext markup language” (“HTML”), the standard Web formatting language. (Galligher Test., Tr. at 125; Joint Stip. ¶¶ 31, 43) Each document has an address, known as a Uniform Resource Locator (“URL”), identifying, among other things, the server on which it resides; most documents also contain “links” — highlighted text or images that, when selected by the user, permit him to view another, related Web document. (Joint Stip. ¶ 34) Because Web servers are linked to the Internet through a common communications protocol, known as hypertext transfer protocol (“HTTP”), a user can move seamlessly between documents, regardless of their location; when a user viewing a document located on one server selects a link to a document located elsewhere, the browser will automatically contact the second server and display the document. (Joint Stip. 1Í1Í 34, 37) Some types of Web client software also permit users to gain access to resources available on FTP and gopher sites. A number of “search engines” — such as Yahoo, Magellan, Alta Vista, WebCrawler, and Lycos — are available to help users navigate the World Wide Web. For example, the service Yahoo maintains a directory of documents available on various Web servers. A user can gain access to Yahoo’s server and type a string of characters as a search request. Yahoo returns a list of documents whose entries in the Yahoo directory match the search string and organizes the list of documents by category. (Galligher Test., Tr. at 134; Plaintiffs Ex. 3) Search engines make use of software capable of automatically contacting various Web sites and extracting relevant information. Some search engines, such as Alta Vista, store the information in a database and return it in response to a user request. Others, such as Yahoo, employ a group of individuals to determine whether and how a site should be categorized in the Yahoo directory. (Gal-ligher Test., Tr. at 137; Supplemental Declaration of William J. Hoffman (“Hoffman Supp.Decl.”) Ex. A, at 39-42 (Testimony of Donna L. Hoffman in ACLU/ALA)) As the growth in Internet use and the wide availability of tools and resources to those with access to the Internet suggest, the Internet presents extremely low entry barriers to those who wish to convey Internet content or gain access to it. In particular, a user wishing to communicate through e-mail, newsgroups, or Internet Relay Chat need only have access to a computer with appropriate software and a connection to the Internet, usually available for a low monthly fee. The user then in a sense becomes a public “speaker,” able to convey content, at relatively low cost, to users around the world to whom it may be of interest. Those who possess more sophisticated equipment and greater technical expertise can make content available on the Internet for retrieval by others (known or unknown) by running a server supporting anonymous FTP, a gopher server, or a Web server. Yet content providers need not necessarily run their own servers or have the programming expertise to construct their own sites; they can lease space on a Web server from another or create a “home page” through an on-line commercial service. The ease of entry of many speakers sets interactive computer systems apart from any other more traditional communications medium that Congress has attempted to regulate in the past. With one-way media such as radio and television broadcasting or cable programming, a user is merely a listener or viewer; in the CDA, Congress sought to target “interactive” computer systems through which a listener or viewer, by definition, has the power to become a speaker. The relative ease of speaker entry and the relative parity among speakers accounts for the unprecedented and virtually unlimited opportunities for political discourse, cultural development, and intellectual activity that Congress found to characterize emerging communication technologies. In seeking to describe the range of tools and opportunities for Internet users to “speak,” we recognize that the categories we delineate are far from clean and the technology is far from static. Indeed, by all indications, the way that we conceptualize various media that we have traditionally viewed as distinct — such as cable television, telephones, and computer networks — will change dramatically as these media “converge” into common forms of communication. See Denver Area Educ. Telecommunications Consortium v. FCC (“Denver Area Consortium”), — U.S.-,-& n. 4, 116 S.Ct. 2374, 2402 & n. 4, 135 L.Ed.2d 888 (1996) (Souter, J., concurring); see also Jerry Berman & Daniel J. Weitzner, Abundance and User Control: Renewing the Democratic Heart of the First Amendment in the Age of Interactive Media, 104 Yale L.J. 1619, 1619 n. 1 (1995); Art Kramer, Netwatch: The AJC’s Daily Online Guide, Atl.J. & Const., May 29, 1996, at B04 (describing cable modem technology designed to offer Internet access through existing cable television connections) (Hoffman Supp.Deel., Ex. C, at 3-4). 'Of course, our findings of fact are necessarily time-bound. We can only determine whether the statutory provision at issue here, in light of the technology available during the pen-dency of this case, comports with the First Amendment. C. Sexually Explicit Content on the Internet It is undisputed that there exists some content on the Internet that is — to use the Government’s phrase — “sexually explicit.” (Defendant’s Memorandum of Law, filed March 19, 1996, at 11) The term “sexually explicit” is descriptive rather than legal and does not appear in the statutory provision at issue, but the Government employs it as a shorthand to describe Internet content depicting “sexual or excretory activities or organs” — possibly though not necessarily in a patently offensive way. (Defendant’s Supplemental Memorandum of Law (“Defendant’s Supp. Memo.”), filed June 7, 1996, at 9) That is, the Government does not contend that all sexually explicit material is “patently offensive” and therefore within the scope of the CDA, but claims that there is certainly content available on the Internet that is both sexually explicit and patently offensive. The testimony and demonstration of one of the Government’s expert witnesses, Howard Schmidt, Director of the Air Force Office of Special Investigations, amply confirmed the availability of sexually explicit material on line. Nevertheless, there is no persuasive evidence in the record to suggest, much less prove, that sexually explicit material easily “assaults” an unknowing user' — as in other media, most notably television and radio — or that any substantial proportion of Internet content is sexually explicit. 1. Ease of Access to Sexually Explicit Content The Government urges us to conclude that an Internet user can easily stumble upon sexually explicit material. (Defendant’s Post-Hearing Memorandum of Law (“Defendant’s Post-Hearing Memo.”), filed May 28, 1996, at 31-32) It is important to begin with the general observation that, with the exception of e-mail, no content appears on a user’s screen without the user having first taken some affirmative step. One wishing to read articles posted to a newsgroup must connect to a Usenet server and select the relevant group. To retrieve a file through anonymous FTP or access a gopher server, the user must search for or know the address of a particular server. To gain access to content on the World Wide Web, a user must know the URL of a relevant site or type a keyword into one of several available search engines. Schmidt’s demonstration focused mainly on the availability of sexually explicit content on the World Wide Web. In the absence of any screening software or filter, a user determined to view a site containing sexually explicit material can certainly do so, either by typing a known URL or by searching for key words; One sexually explicit site may, in turn, contain “links” to other such sites. (Defendant’s Exs. 13, 16, 17, 26, 29, 32; Schmidt Test., Tr. at 401-02) While ordinarily a user must affirmatively seek sexually explicit material to view it, on occasion a search not intended to retrieve sexually explicit material may retrieve a link to a sexually explicit site. For example, Schmidt’s searches of “Sleeping Beauty,” “Babe,” and “Little Women” produced a handful of links to sexually explicit sites. (Defendant’s Exs. 15, 18, 27, 31, 38) This demonstration revealed the inevitable imprecision of search engines — a broad search will almost always return some irrelevant results. In the vast majority of eases, the character of a sexually explicit site will be clear from the entry or link that a search engine returns. Nevertheless, there is potential for occasional accidental viewing of sexually explicit material. For example, if a user were to view entries in a WebCrawler search using that program’s standard format as preset by the manufacturer, he would see no summary of the sites’ contents. (Defendant’s Ex. 18; Shirky Test., Tr. at 237-38) One of Schmidt’s searches of “Sleeping Beauty” returned an entiy offering a link to a site containing sexually explicit material; the entry (when viewed apart from other entries on the same page with similar addresses) gave little indication of the site’s contents. (Defendant’s Ex. 15; Shirky Test., Tr. at 238) It is difficult to know how often accidental viewing can occur, but there is no basis in the record for concluding that a user not seeking out sexually explicit material on the Internet will encounter it with any particular frequency. 2. The Availability of Sexually Explicit Content Although Schmidt’s demonstration focused on the World Wide Web, sexually explicit content is available on the Internet through almost any form of Internet communication. Yet there is no evidence that sexually explicit content constitutes a substantial — or even significant — portion of available Internet content. While it is difficult to ascertain with any certainty how many sexually explicit sites are accessible through the Internet, the-president of a manufacturer of software designed to block access to sites containing sexually explicit material testified in the Philadelphia litigation that there are approximately 5,000 to 8,000 such sites, with the higher estimate reflecting the inclusion of multiple pages (each with a unique URL) attached to a single site. (Stipulated Portions of Record in ACLU/ALA (“Stipulated Record”), Ex. M, at 139-40 (Testimony of Ann W. Duvall in ACLU/ALA)) The record also suggests that there are at least thirty-seven million unique URLs. (Galligher Test., Tr. at 144) Accordingly, even if there were twice as many unique pages on the Internet containing sexually explicit materials as this undisputed testimony suggests, the percentage of Internet addresses providing sexually explicit content would be well less than one tenth of one percent of such addresses. It is not disputed that some of the sexually explicit material that the CDA attempts to keep away from minors originates abroad. This is not surprising inasmuch as forty percent of all host computers are located outside the United States. (Joint Stip. ¶3) Although only a tentative approximation is possible, the record suggests that as much as thirty percent of the sexually explicit material currently available on the Internet originates in foreign countries. (Stipulated Record, Ex. L, ¶ 41; id. Ex. M, at 161-62 (Duvall Test.)) D. The Development of Blocking Tools and Labeling Schemes As the Internet has become accessible to more households, several commercial on-line services and software companies have developed features and packages designed to enable parents to limit children’s exposure to potentially inappropriate Internet material. For example, America Online, Prodigy, and Microsoft Network, which permit their subscribers to obtain access to Internet material, offer parental control options free of charge to their members. (Joint Stip. ¶ 67) America Online, for example, allows parents to establish a separate account for their children limited to the service’s own proprietary content. (Id.) In addition, at least one type of screening software, SurfWatch, has a feature allowing parents to block access to all Internet sites except for those that parents choose to make available to their children. (Stipulated Record, Ex. M, at 131 (Duvall Test.)) The Government offered testimony and a demonstration regarding SurfWatch (configured to act as a screening tool, rather than to block all Internet access) and a second type of screening software,’ Cyber Patrol. Surf-Watch and Cyber Patrol maintain lists of sites known to contain sexually explicit material; when operating while a user attempts to retrieve Internet material, access' to sites identified on their programs will be blocked. In addition, the programs block access to sites whose URLs contain particular character'patterns or words, such as “xxx” or “sex,” and block any searches including those character patterns or words. Because of the constant change in the number and location of Internet sites, both SurfWatch and Cyber Patrol offer regular subscription or update services. But even where a parent has properly installed screening software and the software is operational (and configured to block access to certain sites rather than to the entire Internet), it is possible to retrieve some sexually explicit material. The Government’s witness was able to run searches using “Babe” and “Little Women” as key words with screening software running in the background. As with searches performed in the absence of screening software, the searches returned links to sexually explicit materials. Some of the links were not blocked by the screening tool. In addition, the Government’s witness obtained access to sexually explicit material by directly entering URLs obtained from earlier searches conducted without blocking software in the background. The record also shows that blocking software is not widely owned by or used in households with access to the Internet: nearly seventy percent of SurfWatch’s 1,500 subscribers are schools rather than individual households. (Id. at 163-65) Other efforts to assist' parents in filtering and screening material that their children can view on the,Internet are under way. The World Wide Web Consortium (“WC”) has launched the Platform for Internet Content Selection (“PICS”) to develop technical standards for attaching electronic ratings to Internet addresses. (Joint Stip. ¶¶ 47-49; Stipulated Record, Ex. J., at 1; id. Ex. G, at 2-3 (Declaration of Albert Vezza in ACLU/ ALA)) When the system is fully implemented, PICS-compatible client software (including browsers, newsgroup readers, and mail readers); Internet service providers; and commercial on-line services will be able to detect PICS tags and block content based on how a parent has configured the software. (Joint Stip. ¶ 48; Stipulated Record, Ex. G., at 3 (Vezza Decl.)) PICS will thus enable parents to design from an array of categories blocking criteria that suit the parents’ values or needs. The PICS program envisages both rating by content providers and rating by third parties. (Joint Stip. ¶ 48) The vast majority of Internet sites currently remain unrated. Nevertheless, Microsystems Software, Inc. (which manufactures Cyber Patrol) introduced a PICS ratings server in February 1996. (Id. ¶ 54) Cyber Patrol is itself now PICS-compatible; it can screen out material based on its PICS tag. (Id.) In addition, Microsoft released the first PICS-compatible Web browser, Internet Explorer 3.0, on May 28, 1996. The browser allows parents to block children’s access to all unrated Internet sites and to specify appropriate levels of violence or nudity at rated sites. (Hoffman Supp.Decl., Ex. C, at 1-3) In addition to PICS tags, the Government’s expert witness, Dr. Dan Olsen, testified that extent providers wishing to transmit or make available material potentially falling within the scope of the CDA could develop a general practice of inserting a “tag” or “label” — a string of characters, such as “-L18” (for “not less than 18 years”) — into the address or name of a particular site so as to clearly identify the site as unsuitable for minors. To transmit or gain access to Internet content, a user must specify a textual name: one cannot send e-mail without an email address or the name of a mailing list; post an article to a newsgroup without specifying the name of the group; participate in the Internet Relay Chat without specifying a “channel”; or access a file without its address. (Olsen Decl. ¶¶ 22-26) Accordingly, content providers using all significant modes of Internet communication could use a tag to identify their content as “covered” content. For example, when a sender transmits an email message, the message is accompanied by the sender’s address, which contains a “user name” identifying a particular user and a “domain name” assigned to a computer or set of computers. (Id. ¶¶25, 60) If the string -L18 were added to the domain name, all e-mail originating from, that site — regardless of the particular user who transmitted it — would be identified as containing material falling within the scope of the CDA. In the alternative, a particular user name — rather than a domain name — could contain the L18” tag; only e-mail originating under that user name would be tagged. Finally, a tag could be placed in a textual subject line, so as to identify only particular messages (rather than all e-mail sent under a certain user name or from a certain computer) as containing content potentially within the scope of the CDA. (Id. ¶¶ 60-62) Similarly, a tag such as. “-L18” could be added to the name of a newsgroup; an individual user wishing to post an article potentially falling within the scope of the CDA to a newsgroup that does not as a general matter contain such material could insert a tag in the subject line accompanying the article. (Id. ¶¶ 64-65) A tag could also be placed in the name of an IRC channel. Turning to means of making files available for retrieval or viewing by remote users— using an FTP, gopher, or Web server — content providers could insert a specific tag such as “-L18” in a domain name or site name. Thus, as the Government’s expert witness testified, an owner of a Web site named “www.cyberporn.com” could rename the site “www-L18.cyberpom.com”. (Id. IT 51) If a site only contained specific files falling within the scope of the CDA, a content provider could identify those files by adding a tag to the name of the directory in which the file resides or to the file name itself. That is, a file ■ identified with the address “http://www.adult.com/picturel.html/” could be renamed “http://www.adult.com/picturel-L18.html/”; in the alternative, a content provider could place all covered files within a specific directory, such as “http://www.adult.com/pietures-L18/”. (Id. ¶¶ 51-54) A content provider who did not wish to tag an entire file available on a Web server as unsuitable for minors could place a tag within the HTML source code of the file, thus identifying a particular section as subject to the CDA. (Id. ¶ 58) In any of these approaches, tagging content is, in a technical sense, a trivial act. (Id. ¶¶ 59, 62; Stipulated Record, Ex. B, at 56 (Testimony of Scott O. Bradner in ACLU/ALA■)) There is an alternative means to shield minors from sexually explicit content available uniquely to content providers on the World Wide Web: verification of a user’s “adulthood” before allowing him access to a site. A content provider operating a Web server can create and display an electronic form to retrieve information from a user visiting the Web site; after processing the information by using a program such as a Common Gateway Interface (“cgi”) script, the server could grant or deny access to the site. (Shirky Decl. ¶ 21) Not all content providers who make material available on the Web, however, can use programs such as cgi scripts; for example, commercial on-line services such as America Online and CompuServe provide subscribers with the opportunity to post content by configuring their own Web pages but do not permit subscribers to use cgi scripts. (Olsen Test., Tr. at 345) For Web content providers who lack access to cgi scripts, there is no means of age verification. Although some Web providers can query the user of a site for a credit card number, the cost of verification is significant, ranging from sixty cents per transaction to more than a dollar per transaction. (Id. at 341-42) To take advantage of adult access code or adult identification code verification, a content provider would either have to establish and maintain a registration and verification system (or hire someone else to do so) and issue access codes to users — after verifying then-ages — or associate with one of several adult verification services, such as Adult Check, Adult Verification System, First Virtual, Validate, or VeriSign. (Olsen Decl. ¶ 86 & Ex. I; Schmidt Test., Tr. at 203-14; Defendant’s Exs. 6, 7, 8 & 9) Although neither of the Government’s expert witnesses had any firsthand familiarity with adult verification services, advertising materials suggest that an adult can obtain an identification number from a particular service and access any site registered with the service. For example, a user can register with Adult Check for an annual fee of $9.95; when the user attempts to access any site registered with Adult Check, the user is prompted to enter an Adult Check identification number that is checked against the service’s database. (Defendant’s Ex. 6, at 1) If the number is valid, the user is automatically admitted to the site. (Id.) Although most verification services do not charge content providers to register their sites (Id. Exs. 6-8), at least one service does impose a fee on site owners registered with it. (Id. Ex. 9, at 1) Having explored various means of Internet communication, the availability and accessibility of sexually explicit content, the development of blocking software and rating schemes designed to enable parents to shield their children from inappropriate material, and the potential for tagging and verification procedures that content providers can themselves employ in an effort to shield minors from sexually explicit content that they provide, we turn to the governmental regulation in question. III. Discussion 47 U.S.C. § 223(d), as added by the CDA, targets persons who send or display material that, “in context, depicts or describes, in terms patently offensive as measured by contemporary community standards, sexual or excretory activities or organs.” The language of § 223(d) parallels the definition of “indecency” adopted by the FCC in 1975 in the broadcast context, see FCC v. Pacifica Found., 56 F.C.C.2d 94, 98 (1975); an application of this definition to a radio broadcast of the deliberately provocative George Carlin “Filthy Words” monologue was upheld by the Supreme Court in FCC v. Pacifica Found., 438 U.S. 726, 751, 98 S.Ct. 3026, 3041, 57 L.Ed.2d 1073 (1978). Following Pacifica, the FCC applied the indecency standard only narrowly — taking no enforcement action unless material “involved the repeated use, for shock value, of words similar or identical to those satirized in the Carlin ... monologue” — with the result that no broadcasts were found actionable between 1975 and 1987. Action for Children’s Television v. FCC (“ACT I”), 852 F.2d 1332, 1338, 1336 (D.C.Cir.1988) (internal quotation marks omitted). In 1987, however, the FCC, in three rulings in the broadcast context, interpreted its indecency standard more broadly, extending it beyond the particular language at issue in Paci lea. See Infinity Broadcasting Corp., 2 F.C.C.R. 2705 (1987); Regents of the University of California, 2 F.C.C.R. 2703 (1987); Pacifica Found., Inc., 2 F.C.C.R. 2698 (1987). The same standard was Ira-ported, by statute and by regulation, into other contexts, and applies to commercial telephone messages, see Dial Info. Servs. Corp. v. Thornburgh, 938 F.2d 1535, 1540-41 (2d Cir.1991) (quoting Regulations Concerning Indecent Communications by Telephone, 5 F.C.C.R. 4926, 4927 (1990)), cert. denied, 502 U.S. 1072, 112 S.Ct. 966, 117 L.Ed.2d 132 (1992), and cable programming, see 47 U.S.C. § 532(h); 47 C.F.R. §f 76.701(g), 76.702 (1995); Alliance for Community Media v. FCC ("Alliance"), 56 F.3d 105, 129 (D.C.Cir.1995) (in banc), aff'd in part and rev'd in part sub nom. Denver Area Consortium, - U.S. -, 116 S.Ct. 2374, 135 L.Ed.2d 888 (1996). The plaintiff claims principally that § 223(d), as added by the CDA, is unconstitutional on its face because it is vague and substantially overbroad. Where a plaintiff seeks to "stay government action taken in the public interest pursuant to a statutory or regulatory scheme," he must demonstrate a likelihood of success on the merits of his claims and that he will suffer irreparable harm in the absence of an injunction. Able v. United States, 44 F.3d 128, 131 (2d Cir.1995) (per curiam) (internal quotation marks omitted). it is well settled that "[t]he loss of First Amendment freedoms, for even minimal periods of time, unquestionably constitutes irreparable injury." Elrod v. Burns, 427 U.S. 347, 373, 96 S.Ct. 2673, 2690, 49 L.Ed.2d 547 (1976) (plurality opinion). Accordingly, a finding of irreparable harm flows from a court's conclusion that a governmental regulation has a chilling effect on free expression. We examine the plaintiffs vagueness and overbreadth challenges in turn. A. Vagueness We consider first the plaintiffs claim that § 223(d) is unconstitutionally vague-that it fails to convey to persons of ordinary intelligence reasonable notice of. what conduct is prQhibited and creates a danger of arbitrary and discriminatory enforcement. See Grayned v. City of Rockford, 408 U.S. 104, 108-09, 92 S.Ct. 2294, 2298-99, 33 L.Ed.2d 222 (1972). Where a federal statute or regulation fails to supply a fair warning of what will give rise to criminal liabifity, it violates the Due Process Clause of th~ Fifth Amendment; where a statute or regulation purports to limit freedom of expression, its vagueness will also "operatet 1 to inhibit the exercise" of that freedom and violate the First Amendment. Id. (internal quotation marks omitted). As previously noted, § 223(d) essentially codifies the FCC definition of indecency sustained in a particular factual context by the Supreme Court in Pacifica. Although the Pacifica Court never specifically addressed whether the FCC's definition was unconstitutionally vague, the Court's conclusion that the broadcast at issue in Pacifica was "indecent" and the fact that the Court quoted elements of the FCC's indecency definition with approval, see 438 U.S. at 739, 98 S.Ct. at 3035, has been read to foreclose a vagueness challenge to the FCC's definition for indecency in the broadcast medium. See ACT I, 852 F.2d at 1339-40 ("{I]f acceptance of the FCC's generic definition of `indecent' as capable of surviving a vagueness challenge is not implicit in Pacifica, we have misunderstood Higher Authority and welcome correction."); see also Action for Children's Television v. FCC, 932 F.2d 1504, 1508 (D.C.Cir.1991) ("ACT II"), cert. denied sub nom. Children's Legal Found. v. Action for Children's Television, 503 U.S. 913, 112 S.Ct. 1281, 117 L.Ed.2d 507 (1992); Action for Children's Television v. FCC, 58 F.3d 654, 659 (D.C.Cir.1995) (in banc) ("ACT III"), cert. denied sub nom. Pacifica Found, v. FCC, — U.S. -, 116 S.Ct. 701, 133 L.Ed.2d 658 (1996). Relying on the reasoning of Pacifica and ACT I, the courts of appeals have found vagueness challenges to analogous FCC definitions reaching commercial telephone communications and cable programming unavailing. See Dial Info. Servs., 938 F.2d at 1540-41 (indecent commercial telephone messages); Information Providers’ Coalition for the Defense of the First Amendment v. FCC, 928 F.2d 866, 874-76 (9th Cir.1991) (same); Alliance, 56 F.3d at 129 (cable programming). Most recently, the Alliance court’s approach on this question was affirmed by a plurality of the Supreme Court. Denver Area Consortium, — U.S. at--, 116 S.Ct. at 2389-90. In light of Supreme Court and other precedent rejecting claims that the language used by the FCC to define indecency is unconstitutionally vague, we cannot conclude that the plaintiff has demonstrated a likelihood of success on his claim that the incorporation of a virtually identical verbal formula into § 223(d) renders that statute fatally vague. The plaintiff appears to concede that a challenge based solely on the “patently offensive” language is foreclosed, but calls our attention to other purported defects in the statutory language. First, the plaintiff contends that assessment of a work’s “context” in determining whether it is “patently offensive” is highly unpredictable and subjective. (Plaintiff’s Memorandum of Law, filed Feb. 17, 1996, at 23) Second, § 223(d) requires content providers to judge what content will and will not subject them to criminal liability by reference to the “standards” of an unidentified or fictitious “community.” (Plaintiffs Post-Hearing Memorandum of Law (“Plaintiff’s Post-Hearing Memo.”), filed May 21, 1996, at 37) We conclude that neither argument supports a conclusion that § 223(d) is unconstitutionally vague. In addition, we address briefly the basis for our disagreement with the contrary conclusion reached by two of the judges in the Philadelphia litigation. We first address the inclusion in § 223(d) of the phrase “in context.” While the FCC definition that has been applied to television broadcasting since 1987 — the subject of unsuccessful vagueness challenges— has included this phrase, see Infinity Broadcasting Corp., 2 F.C.C.R. at 2705, definitions employed by the FCC with respect to other media have not explicitly included this phrase. See Dial Info. Servs., 938 F.2d at 1540 (indecent commercial telephone messages); Information Providers’ Coalition, 928 F.2d at 869 (same); Alliance, 56 F.3d at 105 (cable programming). Nevertheless, an assessment of a work’s context has always been a component of indecency analysis regardless of the medium; the incorporation of the phrase “in context” merely follows the approach of Pacifica and later cases. See Pacifica, 438 U.S. at 744, 98 S.Ct. at 3037-38 (plurality opinion); id. at 750, 98 S.Ct. at 3041 (majority opinion); Information Providers’ Coalition, 928 F.2d at 876; cf. ACT I, 852 F.2d at 1340 (discussing relevance of social value of material as factor in determining whether material is patently offensive); S.Conf.Rep. 230, 104th Cong., 2d Sess. 189 (“The gravamen of the indecency concept is ‘patent offensiveness.’ Such a determination cannot be made without a consideration of the context of the description or depiction at issue.”). We cannot see how importing certain language that has been used by various courts considering challenges to the definition of indecency renders the CDA unconstitutionally vague. The plaintiff’s second point concerns the ability (or inability) of an Internet content provider to assess what “community standards” govern the transmission or display of patently offensive materials. A communication posted by an individual in New York City to a Usenet server and thereby made available to countless subscribers around the world might indisputably fall outside the scope of what is “indecent” by the standards of New York City, but might subject the individual to criminal prosecution in other federal districts. Nevertheless, in light of the fact that modern communications have long transcended community borders, this problem is not a novel one. Indeed, the definition of obscenity requires a publisher or distributor of arguably obscene material to look to contemporary community standards in various localities into which materials are distributed. See, e.g., Sable Communications of California, Inc. v. FCC, 492 U.S. 115, 125-26, 109 S.Ct. at 2836-37 (1989) (concluding that failure to apply uniform national standard of obscenity does not render statute unconstitutional; “If [the provider’s] audience is comprised of different communities with different local standards, [the provider] ultimately bears the burden of complying with the prohibition on obscene messages.”); see also Miller, 413 U.