``Inside Risks'', Comm. ACM, vol 34, no 10, October 1991, page 162 A NATIONAL DEBATE ON ENCRYPTION EXPORTABILITY Clark Weissman (clark@NISD.CAM.UNISYS.COM) Traditionally, cryptography has been an exclusively military technology controlled by the National Security Agency (NSA). Therefore, U.S. International Traffic in Arms Regulations (ITARS) require licenses for all export of modern cryptographic methods. Some methods, such as the Data Encryption Standard (DES), are easily obtained for export to the Coordinating Committee for Multilateral Export Controls (COCOM) countries, but not Soviet block countries, or most third world nations. The recent National Research Council (NRC) report, ``Computers At Risk''  describes advances in computer security uses for cryptography beyond traditional COMSEC (communications security) applications of secure text encoding. These permit business to be conducted over the network and include identification and authentication ``signatures,'' permission credentials transactions, registration and notarizing by third parties, unforgeable integrity checksums, ``indelible'' date/time stamp, non-repudiation of message receipt, and electronic money. These new applications make cryptography a ``dual use'' technology for both civilian and military users. Encryption is used in the civil sector for international banking, electronic information exchange, electronic mail, machine safety, and internetwork commerce. It is the separation of secrecy and authentication encryption that underlies the dual use argument. This separation was made explicit in public key cryptosystems. Industry needs cryptography for vitality and growth, which must be international in scope to address common encryption algorithms, encryption applications, key management and distribution methods irrespective of national boundaries. However, most governments have policies to restrict public access to encryption services in telecommunications. U.S. export controls constrain domestic and international growth of encryption services. Our international trading partners have less severe export restrictions. The U.S. finds itself in a dilemma: harm our economic growth and competitiveness in the expanding world internet products and services industries if we prohibit cryptographic applications, or permit such cryptographic export and potentially weaken military security by providing new encryption capabilities to our adversaries. In both cases U.S. National Security is at issue as noted in two other NRC studies [2,3]. These reports define the agenda and technical foundations for a major encryption policy debate, while there is still time to influence the market place. We risk diminution of our U.S. role in the advancing world market for telecommunications at worst, and lost opportunity to lead the international democratic societies in establishing standard, quality, privacy telecommunication services world wide, at best. The debate is international in applicability. However, U.S. policy on encryption appears most severe, so I urge a U.S. National debate to begin the dialog, and start with some questions. Do we gain more by strengthening our commercial competitiveness and products, upon which the military is increasingly dependent, than we lose by permitting international commonality in cryptographic services, which may weaken military capabilities? Who should debate, Congress, DOD, NSA, NIST, National Security Council, public and private agencies, and industry? Can National Security issues be given a fair hearing if the technical and political facts are classified? Will public confidence be raised or weakened by such debate? The proposed Senate bill S.266 required U.S. cryptographic equipment include government ``trapdoors'' which lessened public confidence. Earlier fears of weakness of the DES have diminished because continuing study and dialog suggest the DES to be free of trapdoors . One practical solution by a vendor for product export license used strong encryption for authentication and weak encryption for secrecy. Is this an acceptable compromise solution out of the dual use dilemma? Western democracies have been strengthened by debate on significant issues of public policy. Encryption policy should likewise be debated in the era of a new world order. REFERENCES: 1. National Research Council (NRC), 1991. Computers at Risk: Safe Computing In the Information Age. Computer Science and Telecommunications Board (CSTB), National Academy Press, Washington, D.C. 2. NRC, 1991. Finding Common Ground: U.S. Export Controls in a Changed Global Environment. CSTB... 3. NRC, 1988. Global Trends in Computer Technology and Their Impact on Export Control. CSTB... 4. Denning, D.E. The Data Encryption Standard: Fifteen Years of Public Scrutiny. Dist. Lecture, 6th Ann. Comp. Security Appl. Conf., IEEE Comp. Soc. Press 1990, pp. x-xv. Clark Weissman is Director of Secure Networks for Unisys Defense Systems, Inc. His career has included advances in security penetrations analysis, virtual machine OS, DBMS, and networks, on such projects as KVM, BLACKER, and DNSIX LANs. He has served on many industry, government, and professional security panels.
The following message contains the text of a paper that I gave as the keynote address at IFIP-SEC '91, the annual conference of IFIP TC11, Security and Privacy. The conference was held in Brighton, England, May 1991. The paper addresses the status and prospects of the "trusted systems" evaluation process in the US, and its relationship to evaluation process developments in Europe and elsewhere. Briefly, I conclude that the current process in the US is not really serving the needs of vendors, users, or security authorities, and that the European ITSEC is not much of an improvement. I also give some suggestions for an improved process, probably not as clearly articulated as if I were rewriting the paper today. At last week's National Computer Security Conference, I ran into quite a few people who seemed interested in the paper but who hadn't seen it. The conference proceedings are available, but perhaps a little obscure. I think that it would be of interest to the RISKS audience. = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = Criteria, Evaluation and the International Environment: where have we been, where are we going? Steven B. Lipner Digital Equipment Corporation Littleton, Massachusetts, USA INTRODUCTION This paper presents a few observations on trusted system evaluation criteria. It begins with a summary of the history of the U.S. criteria and the current state of the European criteria. It then discusses the impact of criteria on the vendors and users of commercial computer systems. After suggesting some guidelines for the developers of new criteria, it goes on to suggest a new direction that may better serve the purposes of vendor and user communities, though at the price of abandoning some long-held beliefs. I should stress at the outset that this paper deals only with commercial computer systems and commercial applications (and with civil government computer applications that are indistinguishable from commercial ones). Defense and national security applications have their own unique attributes -- particularly the need to deal with labelled or classified information. After more than ten years of looking, I am convinced that the unique attributes having to do with labelled information are not required in the commercial and civil sector. I would also offer the caveat that this paper reflects about twenty years of experience in computer security, the last ten gained while working in the employ of a commercial computer manufacturer. While it reflects early experience as a defense security researcher and many discussions with security evaluators, researchers, criteria developers, and most of all, would-be users of secure systems, it is clearly written from a vendor perspective and should be read in that light. THE EVOLUTION OF EVALUATION About ten years ago, the United States Department of Defense issued the directive establishing the Department of Defense Computer Security Center. The primary role of the Center was to establish and operate a program that would evaluate the security properties of commercial computer vendors' products. The theory underlying the Center was that commercial and defense users of computer security products had common needs, and that by evaluating commercial products, the Center would improve the security available both to the defense and commercial users. The Center began its task by drafting, coordinating, and publishing the Trusted Computer System Evaluation Criteria (TCSEC, or Orange Book). The TCSEC was published in 1983, and specifies a range of security evaluation classes that apply to operating systems. By 1986 the Center — now renamed the National Computer Security Center (NCSC) and with scope encompassing the entire U.S. Federal Government — had evaluated a handful of commercial operating systems. As we meet, the NCSC's charter has been reduced to the needs of U.S. government agencies that process defense classified information, and a few dozen operating systems have been evaluated. While the NCSC has undergone its decade-long evolution, the rest of the world has not stood still. U.S. legislation has given the U.S. National Institute for Standards and Technology (NIST) the dominant role in setting security standards for civil government and, by implication, the private sector. NIST has not been granted resources consistent with this responsibility. Recently NIST and NCSC have stated that they will work together "to create a new federal computer security criteria document" that can be applied across the entire U.S. government, including civil and defense sectors. A potentially more significant development is that the governments of the U.K., Germany, France and the Netherlands have begun joint development of the Information Technology Security Evaluation Criteria (ITSEC). The European developers of the ITSEC have begun trial use of the draft, and the European Community has begun the process of establishing EC-wide security criteria based on the ITSEC. The governments of Canada, Australia, and Sweden have also expressed, to varying levels, the intent to develop their own information security criteria. NIST is representing the U.S. government in discussions with the ITSEC developers and EC of common evaluation criteria and processes. LIVING WITH THE TCSEC — THE VENDOR'S VIEW Almost every U.S. manufacturer of computer systems has completed at least one evaluation of a Class C2 operating system. C2 systems are the workhorses of commercial computing. They incorporate user identification and authentication mechanisms, auditing, discretionary access controls, and controls over storage residues. They are tested to insure that the controls work "as advertised". They are thus well-suited to the vast majority of commercial multiuser applications. While the TCSEC is widely condemned as applicable only to protection of data from unauthorized disclosure, a C2 operating system provides basic mechanisms that can be used to enforce a level of data integrity as well. Some users require features beyond those required by Class C2 (such as more controls over the passwords used for authentication) and these are frequently added by vendors. Other users DO NOT require some of the features mandated by Class C2, but those unnecessary features can be "turned off" by system administrators. On the whole, C2 is a good common demoninator. The "goodness" of C2 systems, however is marred by two deficiencies that are especially visible to the vendor: First, no one can tell what a C2 system is; and second, by the time a C2 system is developed and evaluated, it is obsolete. I will spend the next few paragraphs amplifying on these deficiencies. At first blush, it appears easy to tell what is in a C2 system — I summarized such a system two paragraphs above. An early senior manager of the NCSC once observed to me that developers of sytems up through class B1 should almost be able to self-evaluate — to tell whether they had met the requirements by comparing their system to the criteria. Unfortunately, any development manager who has taken a system through the NCSC process knows better. Developers and "evaluators" go through seemingly endless disputes: o Is interprocess communication an "object" requiring access controls and auditing? o How is it that self goup/public access controls for one system meet the C2 criteria while those for another do not? o Is the system's internal design documentation (not required at all by the Orange Book) adequate so that the "evaluators" can tell that all of the relevant tests have been developed and objects controlled? The answers to such questions are not found in the TCSEC. They are defined in a set of "interpretations" of the TCSEC maintained by the "evaluators". Because the interpretation process is often triggered by proprietary aspects of vendors' products, the total set of "interpretations" is not visible outside of the NCSC. As new vendor issues arise, however, new "interpretations" are added and cumulatively imposed on future evaluations of all vendors' products. Since every system is different, there are plentiful issues that require new "interpretations". U.S. vendors refer to this phenomenon as "criteria creep". It has the effect that a 1986 C2 system is NOT the same as a 1990 C2 system, and that no one can tell what a 1992 C2 system will be. The problem of obsolesence results from the evaluation process. The vendor negotiates with the NCSC what C2 means for his system, and eventually gets all of the required features, security tests, and design documentation incorporated in a version. When the version goes to customer field test, the NCSC begins its formal evaluation process — a final review of the documentation, running independent tests and so forth. The NCSC process takes time, and it is likely that by the conclusion of that process, the vendor is shipping the version that succeeds the one under evaluation. Pressures on vendors are to get releases out more often, while the pressures on the NCSC are to do a more thorough job of evaluating. Hence, obsolete evaluated versions. The NCSC has developed a "Rating Maintenance Program" or RAMP that is intended to allow vendors to self-evaluate future versions subject to NCSC review and audit once one version has been evaluated by the NCSC. This process requires the vendor to apply the Class B2 requirements for configuration management (as "interpreted") to systems in classes C2 and B1. These requirements impose on the vendor's development process an overlay of paperwork, checking, bureaucracy and mistrust. For those of you familiar with the U.S. Defense "procurement scandals", it is the paperwork, checking, bureaucracy and mistrust associated with configuration management over all aspects of a development process that makes for $800 hammers and the like. There is considerable resistance to RAMP as specified in the U.S. vendor community. LIVING WITH THE TCSEC — THE USER'S VIEW REAL ENVIRONMENTS The TCSEC states security requirements for multi-user computer operating systems — in effect, for monolithic time-sharing systems. My employer is putting substantial resources into the development of an updated version of its proprietary operating system that will meet the Orange Book criteria for Class C2 and B1. I can, however, state with high confidence that no user will operate that system in its evaluated configuration. The reason for my confidence is that TCSEC evaluations exclude general networking facilities, while essentially all "real" computers are installed in networks. Users will buy the system, install it, and set the controls as best they can. We will document our judgment on the security of systems in networks, in the part of the system security manual that is "outside" the C2 evaluation. The NCSC has issued a "trusted network" interpretation (TNI) of the Orange Book. It sets forth criteria for networks of systems designed, built, installed and managed as though they were a single time-sharing system. However, real users proceed by buying a computer, hooking it up, using it, then adding a second computer, wiring it up to the first, adding a third and so on. At some point, they connect to the Internet, to a world-wide public network or both. The TNI is not comprehensible, much less helpful, to such users. REAL APPLICATIONS A second concern with the TCSEC deals with applications. When I log into the NCSC's system (through a network!) I use it in a "B2" way. I have an account, I send and receive mail, I access files in various sensitivity classes. However, many computer systems are dominated by large data bases and large applications that work "across" users. The users may never see the files and applications program at the base of the Orange Book paradigm. Furthermore, the sharing of information may be controlled by data base systems or applications — not just the operating system. In such a configuration -- common in the commercial world — one installs the application with "privilege" to override the operating system controls, and the evaluated product becomes irrelevant. The hard-earned NCSC evaluation is invalidated by the addition of privileged ("trusted") software that was not part of the configuration evaluated by the NCSC. The NCSC is developing a "data base interpretation" of the TCSEC. Time will tell whether it meets the needs of real secure data base systems. It is clear in any case that end users and software houses need more guidance on the development of secure applications than criteria have yet provided. REAL SYSTEM MANAGEMENT In my work, I have occasionally encountered a user who has experienced a security penetration despite using an evaluated system. The common denominator among such incidents is system configuration and management: the user attempts to install the secure system in his environment, and "gets something wrong" that allows a hostile party to go where he shouldn't be. The trusted system evaluation is irrelevant because the user is doing something he should not -- operating in a network or running an application — to get the job done. Vendors try to document real-world ways of using systems securely. They offer tools to help users with this task. When an unexpected problem arises, the vendors learn from it and update their tools and documentation. Real systems used in real applications are complex, and evaluations are not a substitute for experience. The U.S. evaluation process (the only one where we have a rich experience over time) may actually have evolved into an obstacle to security in the area of system management. As the U.S. evaluators insist that more of the objects (containers that hold information) in a computer system be subject to more controls, the security management documents get thicker and thicker. Developers, writers, and system managers are faced with the challenge of designing, documenting, and finding secure ways of using the systems in the face of a forest of controls and auditing options of interest only to the evaluators. ARE THERE ANY BENEFITS? The paragraphs above paint a fairly bleak picture of the NCSC and Orange Book. The question naturally arises "why bother"? Why do vendors continue to have systems evaluated, and what is the benefit to users? The first answer to the question "why bother" is that even in the world of distributed networks and real applications, operating system security is often fundamental to system security. A C2 operating system does have a cohesive set of controls on which a user can build. For a few applications, one installs the sytem, configures the controls, and goes. More often one must do design, integration, and adaptation. Regardless, the C2 controls are a useful foundation. The second answer is that many public procurements have mandated evaluated systems, and more are likely to do so in the future. Thus vendors have little choice but to develop (at least C2 and B1) evaluated systems. The fact that the NCSC has thus dominated the domain of discourse about secure systems bespeaks a significant accomplishment, though potentially at the price of foreclosing other options. For some organizations, if the answer can't be expressed as an evaluated system, it can't be expressed — although the real world is usually much more complex than the domain of the TCSEC. The third answer is that the NCSC process has been fair in a competitive sense. If the criteria change over time, they are applied fairly at any point in time. The vendors know that they will get an unbiased (though painful) evaluation, while users know that they can use the evaluation class as part of a fair competitive procurement. These three attributes — basic security, a wide base of application, and a fair process — bespeak significant accomplishments. Future criteria writers and evaluators should strive to do as well. EVOLVING EVALUATION When I began to draft this paper, I thought briefly of including some high-level observations on the ITSEC and Canadian CTCPEC. I decided not to do so, because I felt I would be shooting at a "moving target" — by the time I delivered the paper, the criteria in question would likely have been revised and my comments would have become moot. I will try instead to offer a few "timeless" guidelines based largely on real-world experience with the ITSEC and NCSC. THE WORLD MARKET My first comment, then, is that that the computer industry is global and evaluations should likewise be global. The NCSC probably contributed in some measure to the development of the ITSEC by excluding non-US vendors from TCSEC evaluations. Happily, the European evaluators have not chosen to reciprocate by exluding U.S. vendors. However, the development of a set of criteria different from the TCSEC would appear to impose a sufficient obstacle -- especially if ITSEC evaluations are as costly and painful for the vendor as those under the TCSEC. Some have proposed a "feature mapping" scheme that would compare criteria by breaking them down into their finest elements. This scheme is likely to be time-consuming and ineffective, if it is feasible at all. A more sensible approach is for criteria developers to agree — if not on criteria, at least on those classes that are comparable. It should not be necessary for a product to undergo more than one evaluation worldwide — at least at classes up through B1 that are not used to protect the most sensitive defense information and that are most interesting commercially. AMBIGUITY The discussion above of the TCSEC and NCSC made it clear that the descriptions in the TCSEC are not sufficiently explicit. I acknowledge to my chagrin that I was a reviewer of the TCSEC drafts and was as stunned as anyone when "interpretations" started to roll in. My surprise was all the greater since the TCSEC was subject to extensive public review and comment during development, while the "interpretation" process is almost completely conducted behind closed doors. If criteria or standards are intended as mandatory guidance for procurements, they should be very explicit about what features are required, where they must be applied, and what assurances must be provided. The first draft ITSEC was relatively precise about assurance of correctness (though see the next section), but the effectiveness and functionality criteria in the ITSEC, and the functionality criteria in the second draft CTCPEC shared with the TCSEC significant ambiguity. Experience teaches us that we must do better. One might ask "why set feature requirements at all"? The ITSEC answered this question by allowing the sponsor to specify arbitrary security features. The answer to this question goes back to one of the benefits of the TCSEC — a fair and competitive process. If vendors may or must go off completely on their own in selecting security feature sets, competitive procurement will likely suffer as procuring organizations find themselves unable to find any set of security features common to two or more competing vendors. Instead, criteria should be very explicit about the core set of security features required, and allow vendors to "add value" by offering additional security features and functions. ADAPTABLE PROCESS In today's computer industry, there is immense pressure to deliver products faster, with more features and better performance. This pressure is at odds with the sorts of rigid development processes proposed in the first draft ITSEC and the NCSC's RAMP. Development processes for secure commercial products should be consistent with the real commercial development environment. They should not attempt to make computer systems into $800 hammers, nor should they impose an atmosphere of mistrust on the development process. This is not to say that vendors should be allowed to "get away with anything". They should not. But evaluation processes should take into account differences among vendors, the need to repair flaws, and the likely impossibility of preventing them totally. They should also allow for process improvement — a key ingredient in the quest for improved product quality that will yield better security. STABILITY Many of the difficulties with the TCSEC result from the fact that it was not tested until after it had been promulgated. Future criteria should be used on real (not toy) systems in substantially final form before they are made authoritative. NEW DIRECTIONS The suggestions above can guide the development of more useful criteria for the evaluation of secure operating systems. Diligently applied, they might reduce the cost and increase the timeliness of developing secure operating system products. They do not, however, "solve the problem" of computer security. Documents such as criteria or standards that are to meet the needs of users and of the custodians of data that require protection must support the development and installation of real systems. This is a daunting challenge. What can we say about heterogeneous networks? about data bases? about real-time systems and commercial applications? Stories abound in the United States of officials from the NCSC visiting banks, offering them copies of the TCSEC and saying "this is the answer to your security problems". Needless to say, no banker believed that assertion once he had examined the TCSEC, thugh most banks DO use systems that have been evaluated in Class C2. Criteria for secure time-sharing systems will not "make it" in the nineties, but it is not clear that we know enough to write evaluation criteria for networks, data bases or applications. The ITSEC specifies measures for assurance and allow arbitrary functionality; the total composition of the secure system can be up to the end user. However, it seems that few end users would be rich or sophisticated enough to apply the costly ITSEC assurance measures to a unique application system. What then to do? I suggest that we should stabilize criteria as a way of evaluating operating system security, and concentrate on removing the blatant silliness and unpredictability that have crept into the NCSC process. We do not know enough to have criteria for everything, and we shouldn't try. Instead, write guidelines for products and practices "outside" the operating system that embody what we do know and think we know. Offer those guidelines to users with proper humility, try them out, and revise them often. Work with users who have the real problem of combining evaluated operating systems with unevaluated applications, data bases, and networks and see if we can develop suggested techniques and guidelines to apply as needed. Identify useful features and document their attributes in clear language that can be used for competitive procurements. Each user will ultimately select features, products, and custom development to meet his own needs. The most that common standards can do is to identify often-needed sets of products or features and suggest, as application notes, ways of configuring and applying them in real-world situations. This latter sort of guidance will give users help in the all-important area of configuring and managing the products that do meet evaluation criteria. If we listen to real experience, in time the guidelines may improve. When the rate of needed revision slows to the point of "stability", we can think about standards. It may even be that there will be additional areas of application for criteria and evaluation, though I for one am not convinced. IN CLOSING There is an old saying from the American West that goes "You can tell the pioneers; they're the ones with the arrows in their backs". The NCSC went first with security evaluation criteria. They have made mistakes, but they have also changed the way the world does — and thinks about — computer security. It is up to us all now to recognize that evaluation and criteria, at least for the moment, are limited to operating system products. Rather than stretch the paradigm where it has no business going, we should concentrate on establishing stable and economical operating system evaluation processes, but put the major focus of our efforts on more broadly applicable guidelines that help to guide choice by users in the development or selection of cost-effective security measures.
Please report problems with the web pages to the maintainer