Chosenplaintext

Chosen-plaintext attack (CPA) is a cryptanalytic model in which an attacker can obtain ciphertexts for plaintexts of their choosing in order to learn information about the secret data or the encryption key. The goal is typically to distinguish between the encryptions of two chosen messages or to recover aspects of the plaintext distribution, rather than directly to recover the key. CPA is a foundational notion for assessing confidentiality in encryption schemes.

There are two common variants: non-adaptive CPA and adaptive CPA. In the non-adaptive setting, the attacker submits a set of plaintexts to be encrypted before the challenge phase. In adaptive CPA, the attacker can use the results of previously obtained ciphertexts to craft new plaintexts for subsequent encryptions. In the standard CPA game, the attacker chooses two equal-length messages m0 and m1, a challenger selects a random bit b and returns the ciphertext of m_b under a secret key. The attacker wins if they can guess b with non-negligible advantage over 1/2. If a scheme remains secure under this game, it is said to be CPA-secure.

CPA is closely related to, but weaker than, indistinguishability under chosen-plaintext attacks (IND-CPA). An encryption scheme that is IND-CPA secure is also CPA secure, and IND-CPA is a formalization of the intuition that ciphertexts should reveal no information about the chosen plaintext. CPA security is a key objective in the design of modern symmetric and public-key encryption schemes, serving as a baseline confidentiality guarantee alongside stronger models such as chosen-ciphertext security (CCA).