Be this the Experiment for multiple COA-security:
$PrivK_{\mathcal{A},\Pi}^{mult}(n)$:
$(m_0^1 , ... , m_0^t,m_1^1 , ... , m_1^t) \leftarrow \mathcal{A}(1^n), |m_0^i|=|m_1^i| \forall i \in [1,t]$
$k\leftarrow Gen(1^n)$
$b \leftarrow \{0,1\}$
$C = (c_b^1 , ... , c_b^t) \leftarrow (Enc_k(m_b^1) , ... , Enc_k(m_b^t))$
$b' \leftarrow \mathcal{A}(C)$
if $b' = b$ return 1 else return 0
If $PrivK_{\mathcal{A},\Pi}^{mult}(n) = 1$ $\mathcal{A}$ wins. For a cryptosystem to have that security, there should not exist an adversary that wins that experiment better than $1/2 + negl(n)$, where $negl(n)$ is a negigible function.
Now I want to construct a cryptosystem that has this security but not KPA- or CPA- or CCA-security. My idea:
- $Gen(1^n)$: Creates a uniform random key $k \leftarrow \{0,1\}^n$
- $Enc_k(m)$: Create a uniform random number $r \leftarrow \{0,1\}^n$ and create $c = m \oplus PRG(k \oplus r) $. Output $(c,r)$
- $Dec_k((c,r))$: Create $m = c \oplus PRG(k \oplus r)$ and output $m$
Assume that PRG is a secure pseudo random generator, then this cryptosystem should be multiple COA-secure (or mult-EAV-IND-secure from Katz & Lindell's textbook (2nd edition))
Is that right or did I overlook something?