Is there a way of maintaining malleability in a homomorphic encryption system while making it infeasible to perform chosen ciphertext attacks?

Question

• Is there a way of maintaining malleability in a homomorphic encryption system while making it infeasible to perform chosen ciphertext attacks?

I have been reading about homomorphic encryption and malleable cryptosystems lately and have found it fascinating. I still have a lot of reading to do, however, I came across a statement in my reading that suggested malleability is inherently counter to security against chosen ciphertext attacks.

While I read up more about this relationship, I am curious to learn if there a way to maintain malleability while making a chosen ciphertext attack computationally infeasible? Why or why not?

Answer 1

I know of two lines of work on this question. It is indeed possible to allow malleability but still make some guarantees in the presence of a chosen-ciphertext attack:

• Manoj Prabhakaran & Mike Rosulek: Reconciling Non-malleability with Homomorphic Encryption.

• Dan Boneh and Gil Segev and Brent Waters: Targeted Malleability: Homomorphic Encryption for Restricted Computations.

Both papers present encryption schemes (and security definitions) that allow malleability $\textsf{Enc}(m) \leadsto \textsf{Enc}(T(m))$ for some set of allowed transformations $T$ (as a feature), but where any other kind of malleability is infeasible.

As a concrete example, suppose the only allowable transformation is the identity transformation. Then it is possible to transform $\textsf{Enc}(m)$ into another "fresh" encryption of the same (unknown) $m$. But it is infeasible to transform $\textsf{Enc}(m)$ into any $m' \ne m$ that is related to $m$. This special case is called "rerandomizable RCCA" encryption.

The first paper is my work, a combination of 3 of our conference papers; the one most relevant to your question is this one. Our construction has additional security requirement: a "transformed" ciphertext obtained via $\textsf{Enc}(m) \leadsto \textsf{Enc}(T(m))$ should be indistinguishable from a "fresh" ciphertext (even to the private-key holder). We only consider the case of unary transformations, since n-ary transformations (i.e., combining several ciphertexts in a transformation) are impossible under these definitions.

The second paper does not have this extra requirement --- so "transformed" ciphertexts look different than "fresh" ciphertexts. They use an approach of appending a ZK proof that an allowable transformation was used on some original ciphertext.