Mittelbach, Arno Andreas (2016)
Random Oracles in the Standard Model.
Technische Universität Darmstadt
Dissertation, Erstveröffentlichung
Kurzbeschreibung (Abstract)
Provable security is a fundamental concept of modern cryptography (see, e.g., Katz and Lindell; Introduction to Modern Cryptography, Chapter 1, 2007). In order to argue about security, we first require a precise and rigorous definition of what security means (e.g., a definition of secure encryption). Such a security definition, in particular, contains a description of the capabilities of an adversary, i.e., a model of the adversary which tries to come as close to reality as possible. Given a security model, the provable security approach is to provide a mathematical proof that a given construction achieves the desired level of security. In other words, we prove that no (efficient) adversary can break the security with good probability given that it behaves as defined within the security model. Typically, proofs reduce the security of a construction to an unproven cryptographic hardness-assumption---we show that the existence of an adversary violates an assumption---which, preferably, is as simple as possible. Furthermore, any assumption needs to be stated precisely. Examples of cryptographic hardness assumptions can be complexity-theoretic assumptions, such as, P vs. NP, the assumed existence of objects, such as, one-way functions, or long standing open problems from number theory, such as, factoring large integers or computing the discrete log in certain groups.
A widely used technique for the construction of cryptographic schemes is the so-called random oracle methodology, introduced in 1993 by Bellare and Rogaway (CCS, 1993). As before, we start with a precise model of security which is extended to include a uniformly random function which may be evaluated by any party (including the construction and adversary). As a random function has a huge, if not infinite description, parties cannot be given its code but, instead, are provided with black-box access to an oracle which evaluates the function for them. This oracle is called the random oracle. Then, as before, a mathematical proof is given that a construction is secure as per definition relative to the random oracle. Finally, to implement the scheme in practice, the random oracle is replaced by a cryptographic hash function (such as SHA-3).
No mathematical model can fully capture reality and, thus, cast in the framework of provable security, we may consider a random oracle security model as being somewhat further away from reality than a standard security model: in addition to the abstractions of the standard security model it is assumed that adversaries do not make any use of the code of a hash function; it is assumed that they do not even evaluate the hash function on their own but use an external device for the evaluation (i.e., black-box access). Now, if we trust schemes that are designed according to the provable security approach, should we then also trust schemes devised via the random oracle methodology?
This question has lead to a huge debate within the cryptographic community and has been discussed for more than two decades. The discussion is fueled by results showing that the extension of security models to include a random oracle may produce provably secure schemes that cannot be securely implemented. In 1998, Canetti, Goldreich, and Halevi (STOC, 1998) showed that schemes exist that are inherently insecure but which should be secure according to the random oracle methodology. In more detail, they present a public-key encryption scheme which an adversary can trivially attack when given the code of the hash function that was used to replace the random oracle. Note that this differs from, e.g., side-channel attacks: while here also the attack is successful because it is outside the model, implementations can, potentially, protect against these attacks and, thus, secure implementations may exist. With the scheme presented by Canetti et al., on the other hand, there is, provably, no secure implementation although the scheme is secure in the random oracle model.
Despite these negative results, many of the schemes that we trust on a daily basis---examples include the standardized public-key encryption scheme RSA-OAEP as well as the standardized signature schemes RSA-PSS and DSA---only have proofs in the random oracle model. Similarly, for many advanced cryptographic concepts, including IND-secure deterministic public-key encryption, correlated-input secure hash functions, universal hardcore functions, and many others, we (so far) only have constructions in the random oracle model. One reason for the success of random oracles is that they allow to design very efficient and natural schemes. Furthermore, the power of random oracles enables us to realize concepts which we would not know how to implement without random oracles. A third, and very compelling argument in favor of the random oracle methodology is that the random oracle heuristic seems to be a good one: no random oracle scheme which was not designed to portrait inconsistencies of the random oracle model has been attacked due to the use of random oracles. However, if a scheme is, indeed, secure, should we then not be able to understand and pinpoint the underlying source of hardness?
In this thesis we study random oracles with the help of program obfuscation (in particular indistinguishability obfuscation and point-function obfuscation). The study of obfuscation has a long tradition in computer science, and specifically in cryptography, but only recent advancements gave rise to the first candidate constructions of provably secure general-purpose indistinguishability obfuscators (Garg et al.; FOCS, 2013). Intuitively, a program obfuscator takes as input a program and produces a functionally equivalent but unintelligible program, i.e., the obfuscated program hides how it operates. Conceptually, this is very close to one of the fundamental abstractions made within the random oracle model where hash functions do not have an explicit and efficient description but can be evaluated only via black-box access to the random oracle. While an obfuscated hash function still has an explicit and efficient description, the description should hide the way the function works and, thus, intuitively, should be of no help to any adversary.
Using obfuscation-based techniques we show how to instantiate the random oracle in various cryptographic constructions. Amongst others, we obtain the first standard model (i.e., without random oracles) candidate construction for a universal hardcore function with long output, a q-query correlated-input secure hash function, and q-query IND-secure deterministic public-key encryption. We obtain our positive results by instantiating various forms of universal computational extractors. The universal computational extractor (UCE) framework was introduced by Bellare, Hoang, and Keelveedhi (CRYPTO, 2013) to provide (very strong) standard-model notions of hash functions that allow instantiating random oracles in a wide range of applications. Intriguingly, even though obfuscation allows us to show how to replace random oracles in certain situations, it also allows us to show limitations of the random oracle methodology as well as of UCEs. Using obfuscation-based techniques, we prove that several concrete UCE assumptions (including all originally proposed assumptions) cannot hold in case indistinguishability obfuscation exists. (We note that these negative results inspired the weaker UCE notions that lay at the core of our positive constructions.) Assuming the existence of indistinguishability obfuscation also allows us to extend the uninstantiability techniques of Canetti et al. (STOC, 1998) and show that a large class of random-oracle transformations are not sound. This affects the Encrypt-with-Hash transformation (Bellare et al.; CRYPTO, 2007) to construct deterministic public-key encryption, as well as the widely used Fujisaki--Okamoto transformation (Fujisaki, Okamoto; CRYPTO, 1999) which transforms weak public-key encryption schemes into strong public-key encryption schemes.
An often repeated criticism of random-oracle uninstantiability results is that the schemes only fail to be secure because they are designed to do so and, furthermore, their artificial design conflicts good cryptographic practice (see, for example, Koblitz and Menezes; Journal of Cryptology, 2007). Similar criticism can be voiced also for our counterexamples to the general applicability of the above mentioned random-oracle transformations. While this does not refute the mathematical validity of such uninstantiability results, we do, however, also present a very different counterexample to the soundness of the random oracle methodology: we show that if indistinguishability obfuscation exists, then a strong variant of point-function obfuscation (which can be similarly interpreted as a strong form of symmetric encryption) cannot be achieved without the help of random oracles while at the same time there are simple and elegant constructions in the random oracle model. We note that the same holds also for our negative results for UCEs.
In summary, we develop techniques to work with obfuscation which allow us to show that the existence of indistinguishability obfuscation implies that various random oracle techniques may lead to insecure schemes. Our results suggest, once again, that we should be careful with random oracle proofs and we hope that they spark further research to overcome the necessity to use random oracles in the first place. Concerning the latter, we make first steps by proposing new and widely applicable UCE notions together with standard-model candidate constructions showing that UCEs may, indeed, be a viable alternative to the use of random oracles.
Typ des Eintrags: | Dissertation | ||||
---|---|---|---|---|---|
Erschienen: | 2016 | ||||
Autor(en): | Mittelbach, Arno Andreas | ||||
Art des Eintrags: | Erstveröffentlichung | ||||
Titel: | Random Oracles in the Standard Model | ||||
Sprache: | Englisch | ||||
Referenten: | Fischlin, Prof. Dr. Marc ; Ran, Prof. Dr. Canetti | ||||
Publikationsjahr: | Dezember 2016 | ||||
Ort: | Darmstadt | ||||
Datum der mündlichen Prüfung: | 7 Dezember 2015 | ||||
URL / URN: | http://tuprints.ulb.tu-darmstadt.de/5222 | ||||
Kurzbeschreibung (Abstract): | Provable security is a fundamental concept of modern cryptography (see, e.g., Katz and Lindell; Introduction to Modern Cryptography, Chapter 1, 2007). In order to argue about security, we first require a precise and rigorous definition of what security means (e.g., a definition of secure encryption). Such a security definition, in particular, contains a description of the capabilities of an adversary, i.e., a model of the adversary which tries to come as close to reality as possible. Given a security model, the provable security approach is to provide a mathematical proof that a given construction achieves the desired level of security. In other words, we prove that no (efficient) adversary can break the security with good probability given that it behaves as defined within the security model. Typically, proofs reduce the security of a construction to an unproven cryptographic hardness-assumption---we show that the existence of an adversary violates an assumption---which, preferably, is as simple as possible. Furthermore, any assumption needs to be stated precisely. Examples of cryptographic hardness assumptions can be complexity-theoretic assumptions, such as, P vs. NP, the assumed existence of objects, such as, one-way functions, or long standing open problems from number theory, such as, factoring large integers or computing the discrete log in certain groups. A widely used technique for the construction of cryptographic schemes is the so-called random oracle methodology, introduced in 1993 by Bellare and Rogaway (CCS, 1993). As before, we start with a precise model of security which is extended to include a uniformly random function which may be evaluated by any party (including the construction and adversary). As a random function has a huge, if not infinite description, parties cannot be given its code but, instead, are provided with black-box access to an oracle which evaluates the function for them. This oracle is called the random oracle. Then, as before, a mathematical proof is given that a construction is secure as per definition relative to the random oracle. Finally, to implement the scheme in practice, the random oracle is replaced by a cryptographic hash function (such as SHA-3). No mathematical model can fully capture reality and, thus, cast in the framework of provable security, we may consider a random oracle security model as being somewhat further away from reality than a standard security model: in addition to the abstractions of the standard security model it is assumed that adversaries do not make any use of the code of a hash function; it is assumed that they do not even evaluate the hash function on their own but use an external device for the evaluation (i.e., black-box access). Now, if we trust schemes that are designed according to the provable security approach, should we then also trust schemes devised via the random oracle methodology? This question has lead to a huge debate within the cryptographic community and has been discussed for more than two decades. The discussion is fueled by results showing that the extension of security models to include a random oracle may produce provably secure schemes that cannot be securely implemented. In 1998, Canetti, Goldreich, and Halevi (STOC, 1998) showed that schemes exist that are inherently insecure but which should be secure according to the random oracle methodology. In more detail, they present a public-key encryption scheme which an adversary can trivially attack when given the code of the hash function that was used to replace the random oracle. Note that this differs from, e.g., side-channel attacks: while here also the attack is successful because it is outside the model, implementations can, potentially, protect against these attacks and, thus, secure implementations may exist. With the scheme presented by Canetti et al., on the other hand, there is, provably, no secure implementation although the scheme is secure in the random oracle model. Despite these negative results, many of the schemes that we trust on a daily basis---examples include the standardized public-key encryption scheme RSA-OAEP as well as the standardized signature schemes RSA-PSS and DSA---only have proofs in the random oracle model. Similarly, for many advanced cryptographic concepts, including IND-secure deterministic public-key encryption, correlated-input secure hash functions, universal hardcore functions, and many others, we (so far) only have constructions in the random oracle model. One reason for the success of random oracles is that they allow to design very efficient and natural schemes. Furthermore, the power of random oracles enables us to realize concepts which we would not know how to implement without random oracles. A third, and very compelling argument in favor of the random oracle methodology is that the random oracle heuristic seems to be a good one: no random oracle scheme which was not designed to portrait inconsistencies of the random oracle model has been attacked due to the use of random oracles. However, if a scheme is, indeed, secure, should we then not be able to understand and pinpoint the underlying source of hardness? In this thesis we study random oracles with the help of program obfuscation (in particular indistinguishability obfuscation and point-function obfuscation). The study of obfuscation has a long tradition in computer science, and specifically in cryptography, but only recent advancements gave rise to the first candidate constructions of provably secure general-purpose indistinguishability obfuscators (Garg et al.; FOCS, 2013). Intuitively, a program obfuscator takes as input a program and produces a functionally equivalent but unintelligible program, i.e., the obfuscated program hides how it operates. Conceptually, this is very close to one of the fundamental abstractions made within the random oracle model where hash functions do not have an explicit and efficient description but can be evaluated only via black-box access to the random oracle. While an obfuscated hash function still has an explicit and efficient description, the description should hide the way the function works and, thus, intuitively, should be of no help to any adversary. Using obfuscation-based techniques we show how to instantiate the random oracle in various cryptographic constructions. Amongst others, we obtain the first standard model (i.e., without random oracles) candidate construction for a universal hardcore function with long output, a q-query correlated-input secure hash function, and q-query IND-secure deterministic public-key encryption. We obtain our positive results by instantiating various forms of universal computational extractors. The universal computational extractor (UCE) framework was introduced by Bellare, Hoang, and Keelveedhi (CRYPTO, 2013) to provide (very strong) standard-model notions of hash functions that allow instantiating random oracles in a wide range of applications. Intriguingly, even though obfuscation allows us to show how to replace random oracles in certain situations, it also allows us to show limitations of the random oracle methodology as well as of UCEs. Using obfuscation-based techniques, we prove that several concrete UCE assumptions (including all originally proposed assumptions) cannot hold in case indistinguishability obfuscation exists. (We note that these negative results inspired the weaker UCE notions that lay at the core of our positive constructions.) Assuming the existence of indistinguishability obfuscation also allows us to extend the uninstantiability techniques of Canetti et al. (STOC, 1998) and show that a large class of random-oracle transformations are not sound. This affects the Encrypt-with-Hash transformation (Bellare et al.; CRYPTO, 2007) to construct deterministic public-key encryption, as well as the widely used Fujisaki--Okamoto transformation (Fujisaki, Okamoto; CRYPTO, 1999) which transforms weak public-key encryption schemes into strong public-key encryption schemes. An often repeated criticism of random-oracle uninstantiability results is that the schemes only fail to be secure because they are designed to do so and, furthermore, their artificial design conflicts good cryptographic practice (see, for example, Koblitz and Menezes; Journal of Cryptology, 2007). Similar criticism can be voiced also for our counterexamples to the general applicability of the above mentioned random-oracle transformations. While this does not refute the mathematical validity of such uninstantiability results, we do, however, also present a very different counterexample to the soundness of the random oracle methodology: we show that if indistinguishability obfuscation exists, then a strong variant of point-function obfuscation (which can be similarly interpreted as a strong form of symmetric encryption) cannot be achieved without the help of random oracles while at the same time there are simple and elegant constructions in the random oracle model. We note that the same holds also for our negative results for UCEs. In summary, we develop techniques to work with obfuscation which allow us to show that the existence of indistinguishability obfuscation implies that various random oracle techniques may lead to insecure schemes. Our results suggest, once again, that we should be careful with random oracle proofs and we hope that they spark further research to overcome the necessity to use random oracles in the first place. Concerning the latter, we make first steps by proposing new and widely applicable UCE notions together with standard-model candidate constructions showing that UCEs may, indeed, be a viable alternative to the use of random oracles. |
||||
Alternatives oder übersetztes Abstract: |
|
||||
URN: | urn:nbn:de:tuda-tuprints-52224 | ||||
Sachgruppe der Dewey Dezimalklassifikatin (DDC): | 000 Allgemeines, Informatik, Informationswissenschaft > 004 Informatik | ||||
Fachbereich(e)/-gebiet(e): | 20 Fachbereich Informatik > Kryptographie und Komplexitätstheorie 20 Fachbereich Informatik |
||||
Hinterlegungsdatum: | 24 Jan 2016 20:55 | ||||
Letzte Änderung: | 24 Jan 2016 20:55 | ||||
PPN: | |||||
Referenten: | Fischlin, Prof. Dr. Marc ; Ran, Prof. Dr. Canetti | ||||
Datum der mündlichen Prüfung / Verteidigung / mdl. Prüfung: | 7 Dezember 2015 | ||||
Export: | |||||
Suche nach Titel in: | TUfind oder in Google |
Frage zum Eintrag |
Optionen (nur für Redakteure)
Redaktionelle Details anzeigen |