I am thinking of a variant of RSA. I know that RSA itself is not semantically secure as it is "deterministic". I have a "randomized" variant of RSA in mind.
Suppose that the public key is $<e,n>$. To encrypt a plaintext $p$, this variant first randomly choose a value $v$, and output the ciphertext as: $$E(p)=<v,(v+p)^e \text{ mod }n>$$ The ciphertext consists of 2 parts, the value of $v$ in clear, and the ciphertext of $(v+p) \text{ mod }n$ computed using the original RSA.
Is this variant of RSA semantically secure? How do I prove it?
No it's not.
As a reminer: Semantic security is equivalent to IND-CPA. Semantic security is less commonly used, because most of the time proofs are less intuitive and more difficult.
In the IND-CPA game, the attacker chooses two messages $m_0,m_1$ and sends them to the challenger. The challenger chooses a bit $b$, and sends $Enc(m_b)$ to the attacker. The attacker wins if he can guess $b$ correctly. And we consider the encryption scheme secure, if the advantage of the attacker compared to random guessing is negligible.
So, let's play the CPA game with this variant:
If you want to have an IND-CPA RSA variant, there is RSA-OAEP. In the random oracle model and with the RSA assumption, this is proven to be IND-CCA2, which is much stronger than IND-CPA.
External links referenced by this document: