Simmons describes how the "Prisoners' Problem" can be solved through parameter substitution in digital signature algorithms.[2][a]
Signature algorithms like ElGamal and DSA have parameters which must be set with random information. He shows how one can make use of these parameters to send a message subliminally. Because the algorithm's signature creation procedure is unchanged, the signature remains verifiable and indistinguishable from a normal signature. Therefore, it is hard to detect if the subliminal channel is used.
Subliminal channels can be classified into broadband and narrowband channel types.
Broadband and narrowband channels can exist in the same datastream.
The broadband channel uses almost all available bits that are available to use. This is commonly understood to mean {≥50% but ≤90%} channel utilization.
Every channel which uses fewer bits is called a narrow-band channel.
The additional used bits are needed for further protection, e.g., impersonation.
The broadband and the narrow-band channels can use different algorithm parameters. A narrow-band channel cannot transport maximal information, but it can be used to send the authentication key or datastream.
Research is ongoing : further developments can enhance the subliminal channel, e.g., allow for establishing a broadband channel without the need to agree on an authentication key in advance. Other developments try to avoid the entire subliminal channel.
Examples
An easy example of a narrowband subliminal channel for normal human-language text would be to define that an even word count in a sentence is associated with the bit "0" and an odd word count with the bit "1". The question "Hello, how do you do?" would therefore send the subliminal message "1".
At signing the parameter has to be set random. For the broadband channel this parameter is instead set with a subliminal message .
Key generation
choose prime
choose prime
calculate generator
choose authentication key and send it securely to the receiver
calculate public key mod
Signing
choose message
(hash function is here substituted with a modulo reduction by 107) calculate message hash value mod mod
instead of random value subliminal message is chosen
calculate inverse of the subliminal message mod
calculate signature value mod mod mod mod
calculate signature value mod mod
sending message with signature triple
Verifying
receiver gets message triple
calculate message hash mod mod
calculate inverse mod
calculate mod mod
calculate mod mod
calculate signature mod mod mod mod
since , the signature is valid
Message extraction on receiver side
from triple (1337; 12, 3)
extract message mod
The formula for message extraction is derived by transposing the signature value calculation formula.
mod
mod
mod
Example: Using a Modulus n = pqr
In this example, an RSA modulus purporting to be of the form n = pq is actually of the form n = pqr, for primes p, q, and r. Calculation shows that exactly one extra bit can be hidden in the digitally signed message. The cure for this was found by cryptologists at the Centrum Wiskunde & Informatica in Amsterdam, who developed a Zero-knowledge proof that n is of the form n = pq.[citation needed] This example was motivated in part by The Empty Silo Proposal.
Example - RSA Case study
Here is a (real, working) PGP public key (using the RSA algorithm), which was generated to include two subliminal channels - the first is the "key ID", which should normally be random hex, but below is "covertly" modified to read "C0DED00D". The second is the base64 representation of the public key - again, supposed to be all random gibberish, but the English-readable message "//This+is+Christopher+Drakes+PGP+public+key//Who/What+is+watcHIng+you//" has been inserted. Adding both these subliminal messages was accomplished by tampering with the random number generation during the RSA key generation phase.
It should be mentioned that this countermeasure has a 1-bit subliminal channel. The reason for that is the problem that a proof can succeed or purposely fail.[9]
Another countermeasure can detect, but not prevent, the subliminal usage of the randomness.[10]
^Gustavus J. Simmons. Subliminal communication is easy using the DSA. In EUROCRYPT ’93: Workshop on the theory and application of cryptographic techniques on Advances in cryptology, pages 218–232, Secaucus, NJ, USA, 1994. Springer-Verlag New York, Inc.
^Gustavus J. Simmons. The subliminal channel in the U.S. Digital Signature Algorithm (DSA), in
Proceedings of the 3rd Symposium on State and Progress of Research in Cryptography (SPRC '93), Rome, Italy, February 15–16, 1993.
^Gustavus J. Simmons. A Secure Subliminal Channel (?). In CRYPTO ’85: Advances in Cryptology, pages 33–41, London, UK, 1986. Springer-Verlag.
^Ross J. Anderson, Serge Vaudenay, Bart Preneel, and Kaisa Nyberg. The Newton Channel. In Proceedings of the First International Workshop on Information Hiding, pages 151–156, London, UK, 1996. Springer-Verlag.
^Yvo Desmedt. Abuses in Cryptography and How to Fight Them. In CRYPTO ’88: Proceedings of the 8th Annual International Cryptology Conference on Advances in Cryptology, pages 375–389, London, UK, 1990. Springer-Verlag.
^Desmedt, Yvo (1996). "Simmons' Protocol is Not Free of Subliminal Channels". Proc. of 9th IEEE Computer Security Foundations Workshop. pp. 170–175. CiteSeerX10.1.1.56.4816.
^Choi, Jong Youl; Golle, Philippe; Jakobsson, Markus (2006). "Tamper Evident Digital Signatures: Protecting Certification Authorities Against Malware". Proceedings of the 2nd IEEE International Symposium on Dependable Autonomic and Secure Computing. CiteSeerX10.1.1.61.9340.
Bruce Schneier. Applied Cryptography, Second Edition: Protocols, Algorithms, and Source Code in C, 2. Ed. Wiley Computer Publishing, John Wiley & Sons, Inc., 1995.